[gentoo-user] PCIe x1 or PCIe x4 SATA controller card

[Dale]

Howdy,

As some know, my inventory on hard drives is growing.  I'm at about a
dozen between main rig and backups.  Most are 14TB or above.  I was
pestering ebay the other day and noticed there is a PCIe card that adds
ports that is different.  I had a 4 port one but just replaced it with a
10 port.  It is a PCIe x1 card.  It has the narrow connector.  While
looking around tho, I found a few cards that are PCIe x4.  It has a
wider connector.  According to my google searches, PCIe x4 is faster
than PCIe x1.  It's why some cards are PCIe x8 or x16.  I think video
cards are usually x16.  My question is, given the PCIe x4 card costs
more, is it that much faster than a PCIe x1?  I'm not driving the drives
to the max or anything during normal use but when I use pvmove, it does
pretty much max them out.  I try to connect drives in a way that data
moving from place to place takes the best path but sometimes the future
surprises me. 

Is it better to have PCIe x4 instead? 

Thanks for the info.

Dale

:-)  :-) 

P. S.  My drive list according to my notes.  2 16TBs, 3 14TBs, 1 10TB
and 2 8TBs drive.  I may have a couple smaller ones laying around
somewhere.  I'm pretty sure there is a 6TB lurking about somewhere. 

[Rich Freeman]

Post by Dale
According to my google searches, PCIe x4 is faster
than PCIe x1. It's why some cards are PCIe x8 or x16. I think video
cards are usually x16. My question is, given the PCIe x4 card costs
more, is it that much faster than a PCIe x1?

It could be slower than PCIe x1, because you didn't specify the version.

PCIe uses lanes. Each lane provides a certain amount of bandwidth
depending on the version in use.

For example, a v1 4x card has 1 GB/s of bandwidth. A v4 1x card has
2GB/s of bandwidth.

Note that slot size is only loosely coupled with the number of lanes.
Lots of motherboards have a second 16x slot that only provides 4-8
lanes to save on the cost of a PCIe swich. You can also use adapters
to connect a 16x card to a 1x slot, or you might find a motherboard
that has an open-ended slot so that you can just fit a 16x card onto
the 1x slot. It will of course only use a single lane that way.

So what you need to do is consider the following:

1. How much bandwidth do you actually need? If you're using spinning
disks you aren't going to sustain more than 200MB/s to a single drive,
and the odds of having 10 drives using all that bandwidth are pretty
low. If you're using SSDs then you're more likely to max them out
since the seek cost is much lower.
2. What PCIe version does your motherboard support? Sticking a v4
card on an old motherboard that only supports v2 is going to result in
it running at v2 speeds, so don't pay a premium for something you
won't use. Likewise, if they cut down on the number of lanes assuming
they'll have more bandwidth you might have less than you expected to
have.

Then look up the number of lanes and the PCIe version and see what you
can expect:
https://en.wikipedia.org/wiki/PCI_Express#History_and_revisions

I think odds are you aren't going to want to pay a premium if you're
just using spinning disks. If you actually wanted solid state storage
then I'd also be avoiding SATA and trying to use NVMe, though doing
that at scale requires a lot of IO, and that will cost you quite a
bit. There is a reason your motherboard has mostly 1x slots - PCIe
lanes are expensive to support. On most consumer motherboards they're
only handled by the CPU, and consumer CPUs are very limited in how
many they offer. Higher end motherboards may have a switch and offer
more lanes, but they'll still bottleneck if they're all maxed out
getting into the CPU. If you buy a server CPU for several thousand
dollars one of the main features they offer is a LOT more PCIe lanes,
so you can load up on NVMes and have them running at v4-5. (Typical
NVMe uses a 4x M.2 slot, and of course you can have 16x cards offering
multiples of those.)

The whole setup is pretty analogous to networking. If you have a
computer with 4 network ports you can bond them together and run them
to a switch that supports this with 4 cables, and get 4x the
bandwidth. However, you can also get a single connection to run at
higher speeds (1Gb, 2.5Gb, 10Gb, etc), and you can do both. PCIe
lanes are just like bonded network cables - they are just pairs of
signal wires that use differential signaling, just like twisted pairs
in an ethernet cable. Longer slots just add more of them. Everything
is packet switched, so if there are more lanes it just spreads the
packets across them. Higher versions mean higher speeds in each lane.

[Dale]

Post by Rich Freeman

Post by Dale
According to my google searches, PCIe x4 is faster
than PCIe x1. It's why some cards are PCIe x8 or x16. I think video
cards are usually x16. My question is, given the PCIe x4 card costs
more, is it that much faster than a PCIe x1?

It could be slower than PCIe x1, because you didn't specify the version.
PCIe uses lanes. Each lane provides a certain amount of bandwidth
depending on the version in use.
For example, a v1 4x card has 1 GB/s of bandwidth. A v4 1x card has
2GB/s of bandwidth.
Note that slot size is only loosely coupled with the number of lanes.
Lots of motherboards have a second 16x slot that only provides 4-8
lanes to save on the cost of a PCIe swich. You can also use adapters
to connect a 16x card to a 1x slot, or you might find a motherboard
that has an open-ended slot so that you can just fit a 16x card onto
the 1x slot. It will of course only use a single lane that way.
1. How much bandwidth do you actually need? If you're using spinning
disks you aren't going to sustain more than 200MB/s to a single drive,
and the odds of having 10 drives using all that bandwidth are pretty
low. If you're using SSDs then you're more likely to max them out
since the seek cost is much lower.
2. What PCIe version does your motherboard support? Sticking a v4
card on an old motherboard that only supports v2 is going to result in
it running at v2 speeds, so don't pay a premium for something you
won't use. Likewise, if they cut down on the number of lanes assuming
they'll have more bandwidth you might have less than you expected to
have.
Then look up the number of lanes and the PCIe version and see what you
https://en.wikipedia.org/wiki/PCI_Express#History_and_revisions
I think odds are you aren't going to want to pay a premium if you're
just using spinning disks. If you actually wanted solid state storage
then I'd also be avoiding SATA and trying to use NVMe, though doing
that at scale requires a lot of IO, and that will cost you quite a
bit. There is a reason your motherboard has mostly 1x slots - PCIe
lanes are expensive to support. On most consumer motherboards they're
only handled by the CPU, and consumer CPUs are very limited in how
many they offer. Higher end motherboards may have a switch and offer
more lanes, but they'll still bottleneck if they're all maxed out
getting into the CPU. If you buy a server CPU for several thousand
dollars one of the main features they offer is a LOT more PCIe lanes,
so you can load up on NVMes and have them running at v4-5. (Typical
NVMe uses a 4x M.2 slot, and of course you can have 16x cards offering
multiples of those.)
The whole setup is pretty analogous to networking. If you have a
computer with 4 network ports you can bond them together and run them
to a switch that supports this with 4 cables, and get 4x the
bandwidth. However, you can also get a single connection to run at
higher speeds (1Gb, 2.5Gb, 10Gb, etc), and you can do both. PCIe
lanes are just like bonded network cables - they are just pairs of
signal wires that use differential signaling, just like twisted pairs
in an ethernet cable. Longer slots just add more of them. Everything
is packet switched, so if there are more lanes it just spreads the
packets across them. Higher versions mean higher speeds in each lane.

This is why I asked.  I didn't even think about the different PCIe
versions available.  My mobo, had to go dig out the manual, says it is
PCIe 2.0.  This is a Gigabyte 970A-UD3P motherboard.  Yes, I'm thinking
about building a new rig.  Turn this into a NAS maybe.  Anyway, I'm
assuming 2.0 isn't the slowest or fastest but as you point out, it'll be
a bottleneck.  Everything has a bottleneck somewhere. 

The difference in price isn't that large.  What I did was, I saw the
PCIe x1 and bought it.  It supported Linux.  Later on I noticed the PCIe
x4 and then wondered if I should upgrade to that.  Given the limits of
my mobo and the fact I won't be maxing out the drives anyway, I don't
see the need to upgrade.  Your info pretty much makes that clear.  I
might, just might, see a small difference when using pvmove.  Maybe. 
Given that it generally maxes out the drive as it is, even if it does go
faster, it won't be much.  Add in that when I start a pvmove, I go nap
and do other things anyway, I won't notice it.  Last pvmove took a
little over 19 hours.  Even 20 or 30 minutes isn't much difference in
the grand scheme of things. 

Another question.  My rig is getting a bit aged.  I have a AMD FX-8350 8
core CPU running at 4GHz.  I also have 32GBs of memory.  I've read that
Intel currently has the best bang for buck on CPUs nowadays.  I'm open
to the idea of switching.  As far as speed goes, if I built a new rig
that is using a reasonably cost CPU and memory, would I see any real
improvements?  When I say 'reasonably cost', I usually find the
fastest/newest then drop down a bit to get out of that 'brand new' price
point.  Generally, the difference in price is quite large but the
difference in speed isn't that much.  Also, I got hard drives, I don't
spend much on video cards either since I don't game, except solitaire. 
So, let's say a mobo, CPU and memory.  What price range would I need to
look for?  Just a rough idea.  I figure the CPU will be a few hundred. 
Memory may be half that.  Mobo will likely be close to $200 or so.  I'm
thinking $500 to $700 or so.  Then comes case, video card and all that. 
The CPU, memory and mobo is the ones I try to buy all at once from the
same vendor, like Newegg or Tigerdirect.  Thoughts?  Am I close?

Thanks for the info.  At least I know I'm good on drive speed.  For my
use anyway. 

Dale

:-)  :-) 

[Mark Knecht]

On Mon, Mar 13, 2023 at 7:35 AM Dale <***@gmail.com> wrote:
<SNIP>

Another question. My rig is getting a bit aged. I have a AMD FX-8350 8
core CPU running at 4GHz. I also have 32GBs of memory. I've read that
Intel currently has the best bang for buck on CPUs nowadays. I'm open
to the idea of switching. As far as speed goes, if I built a new rig
that is using a reasonably cost CPU and memory, would I see any real
improvements?

I think it all depends on what you're going to use the machine for and
whether you really use all your CPU for extended periods of time.

For all the hours my machines run they are mostly idle, in the sense
that even if I'm keeping the machine busy watching a movie, doing
backups, browsing the web, even on my older machines none of those
use more than 10-15% of my older machines. The only two things I do
which drove the purchase of my new machine were:

1) Studio level audio recording using Mixbus32C (the for-pay version
of the Open Source project called Ardour)

2) Astrophotography photo processing using the for-pay program
called PixInsight.

Mixbus32C issues are more based around real-time performance
and use of both the Linux and Windows versions, and being able
to transfer projects back and forth between both platforms. I've
never heard you talk about using Windows, nor doing anything
that takes real-time capabilities so that probably doesn't apply.

PixInsight is the processor hog. It can use all my 32GB memory
(and more) and it can run for hours using 100% of my CPU so
it's the one that drove my eventual purchase of a Ryzen 9 5950X.

PixInsight has a benchmark program built in and all the results
are open to look at:

https://pixinsight.com/benchmark/index.php?sort=cpu&os=all

Interestingly I didn't find your processor even on the list and
the top says it covers about 3000 CPU models. You might
take a look at this when you boil your processor choices down
to 2 or 3.

Note that for the specific processor type you can open up the
group and look at individual machines. Most/many include what
motherboard they were running so that can assist you making
choices also.

Hope this helps,
Mark

[Dale]

Post by Mark Knecht
<SNIP>

Another question.  My rig is getting a bit aged.  I have a AMD FX-8350 8
core CPU running at 4GHz.  I also have 32GBs of memory.  I've read that
Intel currently has the best bang for buck on CPUs nowadays.  I'm open
to the idea of switching.  As far as speed goes, if I built a new rig
that is using a reasonably cost CPU and memory, would I see any real
improvements?

I think it all depends on what you're going to use the machine for and
whether you really use all your CPU for extended periods of time. 
For all the hours my machines run they are mostly idle, in the sense
that even if I'm keeping the machine busy watching a movie, doing 
backups, browsing the web, even on my older machines none of those 
use more than 10-15% of my older machines. The only two things I do
1) Studio level audio recording using Mixbus32C (the for-pay version
of the Open Source project called Ardour)
2) Astrophotography photo processing using the for-pay program
called PixInsight.
Mixbus32C issues are more based around real-time performance
and use of both the Linux and Windows versions, and being able
to transfer projects back and forth between both platforms. I've
never heard you talk about using Windows, nor doing anything 
that takes real-time capabilities so that probably doesn't apply.
PixInsight is the processor hog. It can use all my 32GB memory
(and more) and it can run for hours using 100% of my CPU so
it's the one that drove my eventual purchase of a Ryzen 9 5950X.
PixInsight has a benchmark program built in and all the results
https://pixinsight.com/benchmark/index.php?sort=cpu&os=all
Interestingly I didn't find your processor even on the list and
the top says it covers about 3000 CPU models. You might
take a look at this when you boil your processor choices down
to 2 or 3.
Note that for the specific processor type you can open up the
group and look at individual machines. Most/many include what
motherboard they were running so that can assist you making 
choices also.
Hope this helps,
Mark

One of my concerns isn't just speed, it's the age of things like caps
and such on the mobo.  This mobo is around a decade old.  While it is
supposed to be a top of the line board, it's still got caps which tend
to be a weak spot.  I seem to recall looking when I bought this mobo
that it does have Japanese caps which are the best.  Thing is, even they
go bad sometimes.  While this machine is old, it is still pretty fast. 
I really wish I knew the life expectancy of a Gigabyte mobo like this. 
It claims to be 'ultra durable' and given it is in a Cooler Master
HAF-932 case with those large fans, it does run pretty cool, heat tends
to age caps and make the stink get out. 

The other reason, I'm just curious if I build a new rig if I should be
looking to really upgrade by a lot or just get parts that are newer and
less likely to fail due to age.  When I went from previous rig which was
single core to current rig which originally had a 4 core CPU, it was
about 6 to 7 times faster.  When I upgraded to a 8 core, it speed up
some more.  It was a noticeable improvement both times over original
single core rig.  Thing is, it seems CPU frequencies have pretty much
maxed out.  I think pushing above 4.5GHz or so is difficult to do. 
Dang, that is fast.  Over twice the frequency of a microwave oven for
goodness sake.  They seem to be making them more efficient, adding
cores/threads and such as that.  We had a long thread several years back
talking about reaching the max on frequency of CPUs and such.  It's
almost like we need a whole new technology now to make things faster as
far as the CPU frequencies go. 

In the past, I used a list on Tom's Hardware to pick CPUs.  I usually
started about 4 or 5 CPUs down the list, from fastest to slowest, and
started checking prices.  Sometimes a CPU that costs $500 can only be
just a fraction faster than a $200 CPU.  Given that my rig, as you point
out, sits here and waits on me to do something most of the time, that's
a lot of money for something I won't see much time savings on.  I might
add tho, I do sometimes convert videos from 1080p to 720p.  That makes
the CPU max out pretty good.  Compiling Libreoffice, Firefox etc also
maxes out the CPU but those are what, once a month or so???

I was also wondering what a mobo/CPU/memory combo would cost nowadays. 
Maybe someone who recently built a decent rig recalls how much they paid
for those three.  I don't go cheap on power supply but I don't require a
lot for a video card or anything.  Some spend half their money on a
video card alone but I just don't need anything that fancy.  I got a
Nvidia GeForce GTX 650 that drives both my monitor and my TVs through a
splitter and it does just fine.  Heck, the video card fan is pretty much
at idle and the temps cool most all the time so I can't be pushing it to
hard.  Usually, mobo and CPU is the main part of my cost.  Power supply
right behind that.

This is some good info tho.  Maybe someone who built a rig recently can
chime in on costs, US dollar would be nice.  ;-)

Thanks.

Dale

:-)  :-) 

[Frank Steinmetzger]

Post by Mark Knecht

Another question.  My rig is getting a bit aged.  I have a AMD FX-8350 8
core CPU running at 4GHz.  I also have 32GBs of memory.  I've read that
Intel currently has the best bang for buck on CPUs nowadays.  I'm open
to the idea of switching.  As far as speed goes, if I built a new rig
that is using a reasonably cost CPU and memory, would I see any real
improvements?

I think it all depends on what you're going to use the machine for and
whether you really use all your CPU for extended periods of time. 

This! My mini PC with its passive 10 W Celeron N5100 is enough for desktop
use, including encrypted storage. But maybe not for Gentoo. :)

Post by Mark Knecht
[
]
PixInsight has a benchmark program built in and all the results
https://pixinsight.com/benchmark/index.php?sort=cpu&os=all
Interestingly I didn't find your processor even on the list

That’s probably because the FX processors are ooooold. Old and hungry. ^^

Sometimes a CPU that costs $500 can only be just a fraction faster than a
$200 CPU.

That’s still the case today for those impatient gamer enthusiasts who are
after the “longest bars” [in benchmarks]. The same goes for power
consumption. With Zen 4, AMD of course launched the fastest X-processors
first with a gargantuan power demand. A few months later the non-X were
released. They used 40 % or so less power at a performance cost of maybe 10
% (not actual numbers, but figuratively speaking from memory).

Given that my rig, as you point
out, sits here and waits on me to do something most of the time, that's
a lot of money for something I won't see much time savings on.  I might
add tho, I do sometimes convert videos from 1080p to 720p.  That makes
the CPU max out pretty good.  Compiling Libreoffice, Firefox etc also
maxes out the CPU but those are what, once a month or so???

Intel and AMD are giving themselves quite a race these days about who offers
more bang for the buck, or rather, more bang. In the past, Intel used to
have more to offer at the lower end (below 100 € CPUs, like Pentiums and
i3’s, while AMD was milking the market with high-end chips due to their
limited manufacturing capacities).

If you want to save money and aim for a low-cost AMD APU (processor with
integrated graphics), you can get an older 3000-series Ryzen for a two-digit
price. It’ll still be much faster than your old FX at a fraction of the
power consumption. Like the 4300G, which is twice as fast for half the
electricity. With today’s processors, basically none of the socktetable
models are too slow unless you have specific performance requirements.

With each generation, the architecture becomes more efficient, meaning more
instructions per cycle, lower consumption and so on. The max frequency is
not really the driving force behind performance increase anymore due to
efficiency issues at higher frequencies.

Here are some benchmark comparisons from cpubenchmark.net:

Processor year power cores single-core score multi-core score
FX-8350 2012 125 W 8/8 1580 6026
i5-4590 2014 84 W 4/4 2086 5356
i5-10400 2020 65 W 6/12 2580 12258
R3 4300G 2020 65 W 4/8 2557 11017
R5 5600G 2021 65 W 6/12 3185 19892
R5 7600X 2022 145 W 6/12 4213 28753

Sources:
https://www.cpubenchmark.net/singleThread.html#desktop-thread
https://www.cpubenchmark.net/cpu.php?cpu=AMD+FX-8350+Eight-Core&id=1780
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5-4590+%40+3.30GHz&id=2234
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5-10400+%40+2.90GHz&id=3737
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+3+4300G&id=3808
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+5+5600G&id=4325
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+5+7600X&id=5033

You can see the increase in performance. My old i5-4590, at half the cores,
can keep up with your FX, even though it is only 1œ years younger. Ryzens
used to be more efficient in multi workloads (look at the 2020 entries). But
I’m not too sure about current generations due to Intel’s big-little
concept.
DDR5 and PCIe5 have higher requirements at signal quality, making the boards
and components much more expensive (and, again, more power hungry). That’s
why, even though DDR4 platforms are on their way out technologically, they
are still an economically sound choice.

I was also wondering what a mobo/CPU/memory combo would cost nowadays. 
Maybe someone who recently built a decent rig recalls how much they paid
for those three.  I don't go cheap on power supply but I don't require a
lot for a video card or anything.  Some spend half their money on a
video card alone but I just don't need anything that fancy.

Any current Intel non-F CPU (F means no graphics) can cover your graphics
need. Finally, AMD caught up and started shipping a minimal graphics chip in
all of their processors with Zen 4, but as I said, that platform is still
expensive.

I got a Nvidia GeForce GTX 650 that drives both my monitor and my TVs
through a splitter and it does just fine.

How cute. This should be about twice as fast as the integrated graphics in
my 8-year-old i5. So you’ll be fine with *any* integrated graphics (which
will also cut down on idle consuption, compared with a dGPU).

This is some good info tho.  Maybe someone who built a rig recently can
chime in on costs, US dollar would be nice.  ;-)

As mentioned, DDR5 is still expensive. With DDR4 platforms getting older,
their prices are going down. The Ryzen 5 5600G is an excellent and efficient
processor (it’s basically a laptop chip in a desktop socket) and currently
can be had for around 125 € (including taxes of course, not sure about US
prices). It has over twice the single- and thrice the multi-core performance
of your FX chip. Its graphics are way overkill for you, but you never know.
;-)
If you want to keep yout GPU, there’s also the Ryzen 5 5500, it has no
graphics and is only minutely slower than the 5600G, but can be had for less
than 100 €.


So, in summary (talking German consumer prices, meaning all taxes included,
but I think you can assume very similar $ pricse) for a not too fancy¹ system:

Processor 120 € (or up top 150 € for a current i3/i5)
RAM 60 € 32 GB DDR4 (cheap RAM, low latency costs more, but has no real use
for your use case)
Board 100..120 € depending on I/O needs and quality.

Going DDR5 means an increase in budget by at least 100 € for a 32 GB system.


¹ As far as I can see, compiling packages is the most taxing thing you do,
which is why I don’t see you needing a big-rig processor. (Though I
understand the nice feeling you get from having one.)

[Frank Steinmetzger]

Post by Frank Steinmetzger

Post by Mark Knecht
[
]
PixInsight has a benchmark program built in and all the results
https://pixinsight.com/benchmark/index.php?sort=cpu&os=all
Interestingly I didn't find your processor even on the list

That’s probably because the FX processors are ooooold. Old and hungry. ^^

Correction: it is there. Search for just the number 8350 and you’ll find it.

[Dale]

Post by Frank Steinmetzger

Post by Mark Knecht

Post by Dale
Another question.  My rig is getting a bit aged.  I have a AMD FX-8350 8
core CPU running at 4GHz.  I also have 32GBs of memory.  I've read that
Intel currently has the best bang for buck on CPUs nowadays.  I'm open
to the idea of switching.  As far as speed goes, if I built a new rig
that is using a reasonably cost CPU and memory, would I see any real
improvements?

I think it all depends on what you're going to use the machine for and
whether you really use all your CPU for extended periods of time. 

This! My mini PC with its passive 10 W Celeron N5100 is enough for desktop
use, including encrypted storage. But maybe not for Gentoo. :)

Post by Mark Knecht
[…]
PixInsight has a benchmark program built in and all the results
https://pixinsight.com/benchmark/index.php?sort=cpu&os=all
Interestingly I didn't find your processor even on the list

That’s probably because the FX processors are ooooold. Old and hungry. ^^

Sometimes a CPU that costs $500 can only be just a fraction faster than a
$200 CPU.

That’s still the case today for those impatient gamer enthusiasts who are
after the “longest bars” [in benchmarks]. The same goes for power
consumption. With Zen 4, AMD of course launched the fastest X-processors
first with a gargantuan power demand. A few months later the non-X were
released. They used 40 % or so less power at a performance cost of maybe 10
% (not actual numbers, but figuratively speaking from memory).

Given that my rig, as you point
out, sits here and waits on me to do something most of the time, that's
a lot of money for something I won't see much time savings on.  I might
add tho, I do sometimes convert videos from 1080p to 720p.  That makes
the CPU max out pretty good.  Compiling Libreoffice, Firefox etc also
maxes out the CPU but those are what, once a month or so???

Intel and AMD are giving themselves quite a race these days about who offers
more bang for the buck, or rather, more bang. In the past, Intel used to
have more to offer at the lower end (below 100 € CPUs, like Pentiums and
i3’s, while AMD was milking the market with high-end chips due to their
limited manufacturing capacities).
If you want to save money and aim for a low-cost AMD APU (processor with
integrated graphics), you can get an older 3000-series Ryzen for a two-digit
price. It’ll still be much faster than your old FX at a fraction of the
power consumption. Like the 4300G, which is twice as fast for half the
electricity. With today’s processors, basically none of the socktetable
models are too slow unless you have specific performance requirements.
With each generation, the architecture becomes more efficient, meaning more
instructions per cycle, lower consumption and so on. The max frequency is
not really the driving force behind performance increase anymore due to
efficiency issues at higher frequencies.
Processor year power cores single-core score multi-core score
FX-8350 2012 125 W 8/8 1580 6026
i5-4590 2014 84 W 4/4 2086 5356
i5-10400 2020 65 W 6/12 2580 12258
R3 4300G 2020 65 W 4/8 2557 11017
R5 5600G 2021 65 W 6/12 3185 19892
R5 7600X 2022 145 W 6/12 4213 28753
https://www.cpubenchmark.net/singleThread.html#desktop-thread
https://www.cpubenchmark.net/cpu.php?cpu=AMD+FX-8350+Eight-Core&id=1780
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5-4590+%40+3.30GHz&id=2234
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5-10400+%40+2.90GHz&id=3737
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+3+4300G&id=3808
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+5+5600G&id=4325
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+5+7600X&id=5033
You can see the increase in performance. My old i5-4590, at half the cores,
can keep up with your FX, even though it is only 1½ years younger. Ryzens
used to be more efficient in multi workloads (look at the 2020 entries). But
I’m not too sure about current generations due to Intel’s big-little
concept.
DDR5 and PCIe5 have higher requirements at signal quality, making the boards
and components much more expensive (and, again, more power hungry). That’s
why, even though DDR4 platforms are on their way out technologically, they
are still an economically sound choice.

I was also wondering what a mobo/CPU/memory combo would cost nowadays. 
Maybe someone who recently built a decent rig recalls how much they paid
for those three.  I don't go cheap on power supply but I don't require a
lot for a video card or anything.  Some spend half their money on a
video card alone but I just don't need anything that fancy.

Any current Intel non-F CPU (F means no graphics) can cover your graphics
need. Finally, AMD caught up and started shipping a minimal graphics chip in
all of their processors with Zen 4, but as I said, that platform is still
expensive.

I got a Nvidia GeForce GTX 650 that drives both my monitor and my TVs
through a splitter and it does just fine.

How cute. This should be about twice as fast as the integrated graphics in
my 8-year-old i5. So you’ll be fine with *any* integrated graphics (which
will also cut down on idle consuption, compared with a dGPU).

This is some good info tho.  Maybe someone who built a rig recently can
chime in on costs, US dollar would be nice.  ;-)

As mentioned, DDR5 is still expensive. With DDR4 platforms getting older,
their prices are going down. The Ryzen 5 5600G is an excellent and efficient
processor (it’s basically a laptop chip in a desktop socket) and currently
can be had for around 125 € (including taxes of course, not sure about US
prices). It has over twice the single- and thrice the multi-core performance
of your FX chip. Its graphics are way overkill for you, but you never know.
;-)
If you want to keep yout GPU, there’s also the Ryzen 5 5500, it has no
graphics and is only minutely slower than the 5600G, but can be had for less
than 100 €.
So, in summary (talking German consumer prices, meaning all taxes included,
Processor 120 € (or up top 150 € for a current i3/i5)
RAM 60 € 32 GB DDR4 (cheap RAM, low latency costs more, but has no real use
for your use case)
Board 100..120 € depending on I/O needs and quality.
Going DDR5 means an increase in budget by at least 100 € for a 32 GB system.
¹ As far as I can see, compiling packages is the most taxing thing you do,
which is why I don’t see you needing a big-rig processor. (Though I
understand the nice feeling you get from having one.)

This is all good info.  I went to Tom's Hardware and found their list by
computing power.  I try to find a generic power rating since what I use
my rig for is more generic.  No need looking at a chart for gaming. 
;-)  Anyway, I was looking at a somewhat costly Ryzen 7 5800x3d or a
Ryzen 7 7700.  I need to look at the details because I like having my
own video card.  That way I can use Nvidia but switch to something else
if the need should arise.  Plus, if the video stops working, replace
card instead of whole mobo.  I also have to have two outputs.  One for
desktop, one for TV.  Based on your info tho, I could go down more in
price and still have a much better CPU than the current one.

One other thing, the mobos I keep finding have few PCIe slots.  Some
have 2 maybe 3.  That's getting to be to few for me.  I have a ethernet
card, SATA expansion card plus a couple other things in mine that I
use.  Then my next thing, a case.  The cases I find have a ton of
lights, which I hate, but as far as layout and such, they suck.  Some
cost a arm and leg and they are worthless to me.  I found one the other
day that is fairly plain, holds 8 or 10 hard drives and has reasonably
good cooling.  I'm hoping I can get it.  I don't think even Cooler
Master makes a case like what I got anymore.  I need more drive space
but I love the cooling of my current case.  The fans don't spin very
fast but they move a LOT of air, quietly. 

Usually I look forward to building a new rig.  Trying to find things I
like takes the fun out of it.  I'll get there tho.  Eventually. 

Thanks for all the info.  It helps me to know if I build a new rig, I
will see a benefit speed wise.  I want to get something out of it.  lol

Dale

:-)  :-) 

[Frank Steinmetzger]

Post by Frank Steinmetzger
If you want to save money and aim for a low-cost AMD APU (processor with
integrated graphics), you can get an older 3000-series Ryzen for a two-digit
price. It’ll still be much faster than your old FX at a fraction of the
power consumption. Like the 4300G, which is twice as fast for half the
electricity. With today’s processors, basically none of the socktetable
models are too slow unless you have specific performance requirements.
[
]
[
]
You can see the increase in performance. My old i5-4590, at half the cores,
can keep up with your FX, even though it is only 1œ years younger. Ryzens
used to be more efficient in multi workloads (look at the 2020 entries). But
I’m not too sure about current generations due to Intel’s big-little
concept.

Post by Dale
I was also wondering what a mobo/CPU/memory combo would cost nowadays. 
Maybe someone who recently built a decent rig recalls how much they paid
for those three.  I don't go cheap on power supply but I don't require a
lot for a video card or anything.  Some spend half their money on a
video card alone but I just don't need anything that fancy.

Any current Intel non-F CPU (F means no graphics) can cover your graphics
need. Finally, AMD caught up and started shipping a minimal graphics chip in
all of their processors with Zen 4, but as I said, that platform is still
expensive.

Post by Dale
I got a Nvidia GeForce GTX 650 that drives both my monitor and my TVs
through a splitter and it does just fine.

How cute. This should be about twice as fast as the integrated graphics in
my 8-year-old i5. So you’ll be fine with *any* integrated graphics (which
will also cut down on idle consuption, compared with a dGPU).
[
]

This is all good info.  I went to Tom's Hardware and found their list by
computing power.  I try to find a generic power rating since what I use
my rig for is more generic.  No need looking at a chart for gaming. 
;-)  Anyway, I was looking at a somewhat costly Ryzen 7 5800x3d or a
Ryzen 7 7700.

The 5800x3d got a huge boost from its cache. It may be the most powerful AM4
socket processor around. But it is aimed mostly at gamers, because that’s
where the cache mostly shines.

Plus, if the video stops working, replace card instead of whole mobo.

Onboard graphics sits in the CPU. The mobo just routes the signal. And if
any of that breaks, you can still fall back to a slot-in GPU.

I also have to have two outputs.  One for desktop, one for TV.

Most boards have at least two outputs, nowadays usually one HDMI and DP.

One other thing, the mobos I keep finding have few PCIe slots.  Some
have 2 maybe 3.  That's getting to be to few for me.

Gamer boards tend to skimp on ports, because those people generally care
mostly for their GPU (plus design and RGB). I believe you use ATX boards,
right? Here is a list of AM4 ATX boards with at least four PCIe slots,
totalling to 70 models (and still 31 with five slots):
https://skinflint.co.uk/?cat=mbam4&xf=18869_4.
Feel free to play with the filters yourself.

Then my next thing, a case.  The cases I find have a ton of lights, which
I hate, but as far as layout and such, they suck.  Some cost a arm and leg
and they are worthless to me.

Unilluminated ATX cases for at least five 3.5″ drives:
https://skinflint.co.uk/?cat=gehatx&xf=17572_ohne%7E536_5%7E9691_ATX
Or for eight drives:
https://skinflint.co.uk/?cat=gehatx&xf=17572_ohne%7E536_8%7E9691_ATX

There are some nice bland black boxes among them, notably the Fractal Define
series.

I found one the other day that is fairly plain, holds 8 or 10 hard drives
and has reasonably good cooling.  I'm hoping I can get it. 

Here, Linus is showcasing an 8-drives storage machine in a Fractal Define R4:

And here a system with 18 drives 🀪 in a Fractal Define 7XL:

Usually I look forward to building a new rig.  Trying to find things I
like takes the fun out of it.  I'll get there tho.  Eventually. 

I’ve been thinking for two years now. But my old PC keeps on running. I even
play some Cities Skylines with its 8-year-old intel iGPU. I just can’t bring
myself to discard that or to dosh out the money. And I can’t decide for a
form factor. I want to have all at once and use them depending of the mood
of the day. :D

[Rich Freeman]

Post by Frank Steinmetzger
Gamer boards tend to skimp on ports, because those people generally care
mostly for their GPU (plus design and RGB).

Well, that, and the CPU only has so many PCIe lanes and adding ports
beyond that requires a switch. Also, if they have two 16x slots to
allow for dual graphics cards those eat up quite a few of the lanes
(even if one isn't actually 16x).

Post by Frank Steinmetzger
http://youtu.be/DpJViwtct5g
http://youtu.be/FAy9N1vX76o

I think a lot of the consumer cases have been moving away from
accommodating hard drives and making more room for gigantic GPUs.

All that said, I have largely abandoned the crusade of trying to
squeeze a dozen hard drives into one host, in favor of distributed
filesystems. If you're only putting a few drives per host and having
more hosts then it becomes pretty easy to find hardware that works.

Finally, for any system that will be running 24x7 I'd suggest
optimizing for power use per unit of computation (which is a hard
figure to find), and idle power use (unless you actually do something
that keeps the server busy 24x7). Usually newer processors will do
better here. The up-front costs of a CPU are likely to be dwarfed by
the cost of powering it. ARM is of course advantageous if you don't
need too much horsepower or RAM. Unfortunately ARM boards with lots
of RAM are crazy-expensive so it isn't a great option if you need more
than a few GB.

There has been interest in using mini PCs from corporate used sales as
servers, and I'm thinking about building storage around a solution
like this. The drives would need to be external of course, but USB3
is plenty fast for hard drives. However, it is hard to find easy to
lookup metrics on power use and stats like USB3/etc - most filters on
used product sale sites just have filters for RAM and maybe CPU. You
do need to be careful as some of those systems could have high power
draw or lack USB3 or even gigabit LAN, making them unsuitable for 24x7
storage. The price and form factor can be very attractive though, and
power use still tends to be low since large companies do think about
those costs.

--
Rich





--
Rich

[Dale]

<<<SNIP>>>
With each generation, the architecture becomes more efficient, meaning more
instructions per cycle, lower consumption and so on. The max frequency is
not really the driving force behind performance increase anymore due to
efficiency issues at higher frequencies.
Processor year power cores single-core score multi-core score
FX-8350 2012 125 W 8/8 1580 6026
i5-4590 2014 84 W 4/4 2086 5356
i5-10400 2020 65 W 6/12 2580 12258
R3 4300G 2020 65 W 4/8 2557 11017
R5 5600G 2021 65 W 6/12 3185 19892
R5 7600X 2022 145 W 6/12 4213 28753
https://www.cpubenchmark.net/singleThread.html#desktop-thread
https://www.cpubenchmark.net/cpu.php?cpu=AMD+FX-8350+Eight-Core&id=1780
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5-4590+%40+3.30GHz&id=2234
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5-10400+%40+2.90GHz&id=3737
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+3+4300G&id=3808
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+5+5600G&id=4325
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+5+7600X&id=5033
You can see the increase in performance. My old i5-4590, at half the cores,
can keep up with your FX, even though it is only 1½ years younger. Ryzens
used to be more efficient in multi workloads (look at the 2020 entries). But
I’m not too sure about current generations due to Intel’s big-little
concept.
DDR5 and PCIe5 have higher requirements at signal quality, making the boards
and components much more expensive (and, again, more power hungry). That’s
why, even though DDR4 platforms are on their way out technologically, they
are still an economically sound choice.
<<<SNIP>>>
¹ As far as I can see, compiling packages is the most taxing thing you do,
which is why I don’t see you needing a big-rig processor. (Though I
understand the nice feeling you get from having one.)

It's been a while.  I been getting some things ready for garden time and
a few spring projects as well.  I looked at a few lists of CPU
processors.  This is a bit pricey but I may try to buy a AMD Ryzen 9
5900X 12-Core @ 3.7 GHz.  It has 4 more cores but clock speed is a
little slower.  Even just comparing number of cores and the fairly close
clock speed, that alone should make it a bit faster.  Add in that they
make them run code more efficiently now, should be a good bit better.  I
usually try to aim for 4 or 5 times more processing power.  I suspect
this may help with encryption as well since newer CPUs have extra code
just for that on there now.  Most of the mobos also handle a lot more
memory as well.  I have 32GBs now.  Most support 64GB and I think I saw
a 128GB version somewhere. 

Just comparing CPU to CPU, what would you expect as far as increase in
speed?  I'm not expecting a exact number, just curious as to how much
difference I could reasonably expect. 

As to another reply, I've looked at the following cases. 

Fractal Design Node 804  Lots of hard drive space.
Gamemax Master M905  Lots of drive space here too.  A little like my
current HAF-932 case.
Thermaltake Tower 900 Tons of hard drive space.  Looks really large tho. 

Most of those have a fairly close price range depending where I buy.  My
biggest thing, hard drive space.  I added up the other day, including
backups and such, I have around 100TBs of hard drive space.  I made a
list.  2 16TB, 3 14TB, 1 10TB, 2 8TB and a 6TB.  I may have another one
lurking about somewhere.  Most of the larger ones are in my main rig. 

Trying to figure out how much improvement I can expect.  I'm mostly just
worried about the age of my current rig tho. 

Dale

:-)  :-) 

[Peter Humphrey]

It's been a while. I been getting some things ready for garden time and
a few spring projects as well. I looked at a few lists of CPU
processors. This is a bit pricey but I may try to buy a AMD Ryzen 9
little slower.

I have one of those processors. I can't give you benchmarks or anything, but
in practice at 3.7GHz it blows the socks off my older i7-5820K at 3.3 MHz.

[Frank Steinmetzger]

<<<SNIP>>>
With each generation, the architecture becomes more efficient, meaning more
instructions per cycle, lower consumption and so on. The max frequency is
not really the driving force behind performance increase anymore due to
efficiency issues at higher frequencies.
Processor year power cores single-core score multi-core score
FX-8350 2012 125 W 8/8 1580 6026
i5-4590 2014 84 W 4/4 2086 5356
i5-10400 2020 65 W 6/12 2580 12258
R3 4300G 2020 65 W 4/8 2557 11017
R5 5600G 2021 65 W 6/12 3185 19892
R5 7600X 2022 145 W 6/12 4213 28753
https://www.cpubenchmark.net/singleThread.html#desktop-thread
https://www.cpubenchmark.net/cpu.php?cpu=AMD+FX-8350+Eight-Core&id=1780
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5-4590+%40+3.30GHz&id=2234
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5-10400+%40+2.90GHz&id=3737
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+3+4300G&id=3808
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+5+5600G&id=4325
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+5+7600X&id=5033
[
]

It's been a while.  I been getting some things ready for garden time and
a few spring projects as well.  I looked at a few lists of CPU
processors.  This is a bit pricey but I may try to buy a AMD Ryzen 9
little slower.  Even just comparing number of cores and the fairly close
clock speed, that alone should make it a bit faster.  Add in that they
make them run code more efficiently now, should be a good bit better.  I
usually try to aim for 4 or 5 times more processing power.  I suspect
this may help with encryption as well since newer CPUs have extra code
just for that on there now.  Most of the mobos also handle a lot more
memory as well.  I have 32GBs now.  Most support 64GB and I think I saw
a 128GB version somewhere. 
Just comparing CPU to CPU, what would you expect as far as increase in
speed?  I'm not expecting a exact number, just curious as to how much
difference I could reasonably expect. 

Since I personally don’t have any experience with high-performance
contemporary CPUs and can’t remember all those reviews I read in my newsfeed
from time to time, I tend to visit benchmark sites like the aforementioned
cpubenchmark.net. Those provide comparable numbers of synthetic and/or
real-world benchmarks for both single- and multi-core use cases.

The Phoronix Test Suite is another notable name, and also very
linux-centric. I haven’t used that one myself yet, but have a look and click
around:
https://openbenchmarking.org/suites/pts

It’s open source, so you can run it on your own machine to get comparison
numbers.

[Wols Lists]

Thanks for any light you can shed on this.  Googling just leads to a ton
of confusion.  What's true 6 months ago is wrong today.  :/  It's hard
to tell what still applies.

Well, back in the days of the megahurtz wars, a higher clock speed
allegedly meant a faster CPU. Now they all run about 5GHz, and anything
faster would break the speed of light ... so how they do it nowadays I
really don't know ...

Cheers,
Wol

[Rich Freeman]

Post by Wols Lists

Thanks for any light you can shed on this. Googling just leads to a ton
of confusion. What's true 6 months ago is wrong today. :/ It's hard
to tell what still applies.

Well, back in the days of the megahurtz wars, a higher clock speed
allegedly meant a faster CPU. Now they all run about 5GHz, and anything
faster would break the speed of light ... so how they do it nowadays I
really don't know ...

Effective instructions per clock in general, and increasing core count
are some of the big ones. I say effective because I think IPC is
somewhat synthetic and idealized, and there are MANY bottlenecks in a
CPU.

Efficiency improvements that allow a CPU to boost for longer or
increase sustained clock speed also help.

Getting more done in a clock cycle can happen in many ways:
1. Actually reducing the number of cycles needed to complete an
instruction at the most elementary level.
2. Using speculative execution to increase the number of instructions
run in parallel.
3. Improving branch prediction to maximize your speculative execution budget.
4. Reducing the cost of prediction errors by reducing pipelines/etc.
5. Better cache to reduce wait time.
6. Better internal IO to reduce wait time.
7. Better external IO (esp RAM) to reduce wait time.

Those are just ones I've thought of offhand. I'm sure there are tons
of info on which ones matter the most in practice, and things I
haven't thought of.

[Frank Steinmetzger]

I used to use the bogomips number as a rough guide.  Thing is, the new
CPU has a lower bogomips number than my current CPU does.  That doesn't
seem right.

Bogomips seems to be veeeery simple, because it takes the current frequency
into account. So the number will be low when your PC idles and very high
when you compile something. The “bogo” stands for bogus for a reason.

From Wikipedia [https://en.wikipedia.org/wiki/BogoMips]:
“It is not usable for performance comparisons among different CPUs.”

So, I guess that number no longer means much.  So, I went
digging on the site you linked to.  I found this but not sure what to
make of it. 
https://openbenchmarking.org/vs/Processor/AMD+Ryzen+9+5900X+12-Core,AMD+FX-8350+Eight-Core
Some tests, my CPU is faster.  Most, the new one is faster.

Your CPU is not faster at any of them. Look at the label at the top of each
graph; for some tests, lower is better (as in “time taken for a task”).

For instance, the GnuPG test for encrypting a 2 GB file takes 11.6 seconds
on the Ryzen, and 19.4 on your CPU. The test is single-threaded, so for this
kind of task, you can expect around a ⅔ increase in performance per core (or
rather, thread). OTOH, the m-queens 1.2 test is multi-threaded and you get
39 s vs 238 s, meaning over 5 times more performance. Probably at lower
electricity draw.

I'm trying to figure if I'd be better in the
long run to buy that expensive CPU or pick one of the cheaper ones you
mentioned.  I started off with a 4 core on current rig and went to 8
core and slightly higher frequency.  Money wise it was pretty painless. 
I could do that again with new rig.

Of course, only you can answer that in the end. Write down what you need and
what you care about. Weigh those factors. Then decide. Raw CPU power,
electricity bill, heat budget (cooling, noise, dust), the “new and shiny”
factor (like DDR5), and price. As I mentioned earlier, the 7xxx-X series are
hotheads. But when run with a lower power budget, they are very efficient
(which is basically what the non-X do).

Compiles will speed up no matter what CPU you choose. But where else do you
need compute power? Video transcodes can be done in the background, and
there is also a limit to what parallelisation can achieve. Encryption is
also a non-issue for you. Even my 10 year old i3 in the NAS can encrypt over
1 GB per second, IIRC.

[Rich Freeman]

Post by Frank Steinmetzger
Bogomips seems to be veeeery simple, because it takes the current frequency
into account. So the number will be low when your PC idles and very high
when you compile something. The “bogo” stands for bogus for a reason.

Just to add to this, you need to also keep in mind its purpose. The
kernel needs to be able to measure timings that wouldn't make sense to
measure using a timer chip (lots of reasons for this). So it uses a
timer chip to calibrate a delay loop. I don't know what instructions
are being executed in the delay loop but the obvious design goal with
a delay loop is maximum consistency with a short enough time per
iteration that you have sufficient resolution. The BogoMIPS output is
just telling you what the calibration factor was for each cycle of the
loop. It is about as synthetic a benchmark as you can get, and it
measures how quickly your CPU can execute code designed to do nothing
more than waste time.

Post by Frank Steinmetzger
Of course, only you can answer that in the end. Write down what you need and
what you care about. Weigh those factors. Then decide. Raw CPU power,
electricity bill, heat budget (cooling, noise, dust), the “new and shiny”
factor (like DDR5), and price. As I mentioned earlier, the 7xxx-X series are
hotheads. But when run with a lower power budget, they are very efficient
(which is basically what the non-X do).

Are they actually hotheads on an energy consumed per unit of work
basis? As you say, they're efficient. If the CPU has 2x the power
draw, but does 2.5x as much work in a unit of time than the "cooler"
CPU you're comparing it to, then actually doing any job is going to
consume less electricity and produce less heat - it is just doing it
faster.

Max sustained power draw matters for cooling and electrical design
(the latter being something users typically don't try to change). It
isn't really a measure of thermal efficiency since that requires
incorporating some measure of work getting done.

A recent trend is upping the power draw of CPUs/GPUs to increase their
throughput, but as long as efficiency remains the same, it creates
some thermal headaches, but doesn't actually make the systems use more
energy for a given amount of work. Of course if you throw more work
at them then they use more energy.

[Frank Steinmetzger]

Post by Rich Freeman

Post by Frank Steinmetzger
Of course, only you can answer that in the end. Write down what you need and
what you care about. Weigh those factors. Then decide. Raw CPU power,
electricity bill, heat budget (cooling, noise, dust), the “new and shiny”
factor (like DDR5), and price. As I mentioned earlier, the 7xxx-X series are
hotheads. But when run with a lower power budget, they are very efficient
(which is basically what the non-X do).

Are they actually hotheads on an energy consumed per unit of work
basis? As you say, they're efficient. If the CPU has 2x the power
draw, but does 2.5x as much work in a unit of time than the "cooler"
CPU you're comparing it to, then actually doing any job is going to
consume less electricity and produce less heat - it is just doing it
faster.
[
]
A recent trend is upping the power draw of CPUs/GPUs to increase their
throughput, but as long as efficiency remains the same, it creates
some thermal headaches, but doesn't actually make the systems use more
energy for a given amount of work. Of course if you throw more work
at them then they use more energy.

Back in the day, CPUs were sold to run at an optimum work point, meaning a
compromise between silicon wafer yield, power consumption and performance.
Some of the chips were so good, they had the potential for overclocking,
meaning they are stable enough to be clocked higher and to handle the heat.
(But at no guarantee from the manufacturer, I presume. So if you grill it,
it’s your loss.) And heat there was: you could increase a CPU from 4 GHz to
4.4 GHz (10 % increase), but at a lot more power draw than just 10 %. The
performance curve flattens at the high end; processing power does not scale
linearly with power consumption beyond a certain point (else we would do it
already).

These days, modern high-end CPUs seem to come over-clocked from the factory.
Instead, if the user wants to run at a more efficient mode, the BIOS offers
ways to tune down the power budget. You lose 10..20 % in performance, but
gain 20 K in cooling and 30 % or more in power consumption.

10 years ago, when the very efficient Core architecture swept the market,
the high-end “extreme” Haswell models drew 140 W. [0] Comare that to current
generations [1] (Intel) or [2] (AMD), those go beyond 200 W. Of course they
are much much faster, but average-Joe doesn’t need that.

Looking at concrete examples, the Ryzen 7900 has 3.7 GHz sustained max
frequency (meaning no thermal throttling) at 65 W. The 7900X has 4.7 GHz (a
quarter more) and 200 MHz more boost frequency, but is rated at 2œ times the
wattage. The TDP does not tell you how much power the chip takes at most
anymore (it can actually take much more in bursts or when it is still cool),
but for how much thermal energy the cooling system needs to be designed in
order to keep up the maximum (non-turbo, I think) frequency under load. This
means that for a short time or on a low number of cores, the non-X can
sustain almost as much boost clock as the X (it is the same silicon, after
all), but once the cooling can’t keep up, it will throttle.

I’m not very good at explaining the math or providing hard numbers from
memory, because all I know about this matter is from reading the occasional
review. So please have a read yourself (see below). Another reason to take
my word with a grain of salt: I am biased towards environmentally friendly
choices. Power may still be cheap where you live, but every kWh produced has
an impact on the globe.


Power efficiency (“points per Watt” metric):
https://www.techpowerup.com/review/amd-ryzen-9-7900x/24.html
Ryzen 7 5700G (i.e. laptop APU): 240.7 points
Ryzen 7 5700X: 84.5
Ryzen 7 7700X: 83.0
Ryzen 9 7900X: 47.2 at stock (meaning no down-scaling)

A comparison at https://www.xda-developers.com/amd-ryzen-9-7900-review/
shows only around 10 % more performance for the 7900X vs. the 7900:
“The Ryzen 9 7900 is essentially the 7900X without PBO enabled, but it
would be a waste to spend more money on essentially the same chip to then
underclock it for better thermal performance. It's a better value choice
to pick up the Ryzen 9 7900 and then boost up to 7900X-level performance
through a simple BIOS toggle. After this has been carried out, performance
is pretty much identical.”

Some more reading fodder:
https://www.anandtech.com/show/18693/the-amd-ryzen-9-7900-ryzen-7-7700-and-ryzen-5-5-7600-review-ryzen-7000-at-65-w-zen-4-efficiency


[0] https://en.wikipedia.org/wiki/Haswell_(microarchitecture)
[1] https://en.wikipedia.org/wiki/Alder_Lake
[2] https://en.wikipedia.org/wiki/Zen_4

[Wol]

Post by Frank Steinmetzger
Back in the day, CPUs were sold to run at an optimum work point, meaning a
compromise between silicon wafer yield, power consumption and performance.
Some of the chips were so good, they had the potential for overclocking,
meaning they are stable enough to be clocked higher and to handle the heat.
(But at no guarantee from the manufacturer, I presume. So if you grill it,
it’s your loss.)

I remember one supplier, can't remember exactly the details, but it was
something like "we supply overclocked chips to save you money. If you
fry your chip within (18 months it might have been) we'll replace the
chip with one spec'd at the higher price".

They'd done the maths, and it was something like the chip would probably
survive the warranty, and once the warranty expired, chip prices would
have fallen to the point the customer could use the savings and replace
a failed chip. Win win ...

Cheers,
Wol

[Peter Humphrey]

Post by Frank Steinmetzger
Compiles will speed up no matter what CPU you choose. But where else do you
need compute power? Video transcodes can be done in the background, and
there is also a limit to what parallelisation can achieve. Encryption is
also a non-issue for you. Even my 10 year old i3 in the NAS can encrypt over
1 GB per second, IIRC.

Another heavy-load case is BOINC projects [1], which load all CPU threads, or
a proportion of them, with floating-point calculations. I run 24 threads
continuously on this Ryzen 9 and 9 threads on the older I7, except during
Gentoo updates.

I haven't looked into CPU comparisons for this kind of load, preferring to
rely on workstations from a known high-performance system builder [2].

1. https://boinc.berkeley.edu/ - the Berkeley Open Infrastructure for Network
Computing.

2. https://armari.com/ . They specialise in systems for financial trading in
the City of London, where (I'm told) milliseconds count, as well as
reliability of course.

[Peter Humphrey]

I looked at a few lists of CPU processors. This is a bit pricey but I may
clock speed is a little slower. Even just comparing number of cores and the
fairly close clock speed, that alone should make it a bit faster. Add in
that they make them run code more efficiently now, should be a good bit
better.

That's the CPU I have. It's double-threaded, and it just flies. :)

I usually try to aim for 4 or 5 times more processing power. I suspect
this may help with encryption as well since newer CPUs have extra code
just for that on there now. Most of the mobos also handle a lot more
memory as well. I have 32GBs now. Most support 64GB and I think I saw
a 128GB version somewhere.

My machine was built by Armari, and it has 64GB. Even that isn't enough to
accommodate more than one huge package emerge at a time - which is why I'd
like to see the new feature I've been bleating about. I might ask them if I
can double it.

[Peter Humphrey]

Post by Peter Humphrey
My machine was built by Armari, and it has 64GB. Even that isn't enough to
accommodate more than one huge package emerge at a time - which is why I'd
like to see the new feature I've been bleating about. I might ask them if I
can double it.

It turns out that I can double it to 128 GB, but at a cost of course. I'm now
musing over whether I can justify it. I'll also have to consider whether
portage can make effective use of it.

I also discovered that Armari are a big player. They've supplied well over 100
huge systems to CERN for the LHC. I wish I had CERN's money! :)