Overclock.net - An Overclocking Community - Reply to Topic

Thread: [Tom's Hardware] AMD EPYC Rome Benched: 64 Cores, 128 Threads Boosting to 2.2 GHz (for now) Reply to Thread
Title:
Message:

Register Now

In order to be able to post messages on the Overclock.net - An Overclocking Community forums, you must first register.
Please enter your desired user name, your email address and other required details in the form below.
User Name:
If you do not want to register, fill this field only and the name will be used as user name for your post.
Password
Please enter a password for your user account. Note that passwords are case-sensitive.
Password:
Confirm Password:
Email Address
Please enter a valid email address for yourself.
Email Address:

Log-in


  Additional Options
Miscellaneous Options

  Topic Review (Newest First)
05-21-2019 06:05 AM
Asterox
Quote: Originally Posted by mothergoose729 View Post
This is the same benchmark with 2 of intels Xeon 8180s.

https://ranker.sisoftware.co.uk/show...781f2cff7&l=en

And this is two EPYC 32 cores

https://wccftech.com/amd-epyc-7601-benchmarks-analysis/

Assuming the score would double with a dual socket configuration (and it usually does) that would yield a result of 1553 GOPS, which is up to a 16.1% higher score than the dual Xeon 8180s, and a 21.1% higher score than dual EPYC 32 cores.

The performance per core though is pretty terrible. The average clock speed for the run was 2.2ghz and it had twice the cores, compared to the 2.7ghz average clock speed of EPYC 32 cores with half the threads, yet only achieving a modest 21% score increase, instead of what you would expect to be a 81% performance increase with similar IPC. There is a lot of performance left on the table. Whether that is due to poor scaling in the sisoft sandra benchmark, errors in the silicon, or some other issue, who can say. These numbers might not even be accurate.

Put another way, two EPYC 32 core CPUs are 30% faster than a single one of these 64 cores CPU, assuming linear scaling with clock speed.
This is old news from November last year, 64 Core Epyc Rome have 2.3ghz stock CPU clock.You must use software/OS that can in full use 64/128 server CPU.

"This si server CPU, there is no errors in the silicon or again this is just simple ordinary or day to day 64/128 CPU".

https://www.anandtech.com/show/13598...awk-at-235-ghz
05-20-2019 05:47 PM
NightAntilli
Quote: Originally Posted by KyadCK View Post
That is untrue. Performance alone (IPC x clock speed) is linear. If everything on a CPU is scaled up 40%, it will be 40% faster in everything. Performance per watt is not linear, as both speed (linear) and voltage (linear) are multiplied for final wattage and combined with power gating may result in less performance than reported clock speed would otherwise say. There may also be bottlenecks in, say, RAM controllers or otherwise that limit a core's ability to work, but that is not the same as the CPU not scaling; the Core's performance (IPC) is irrelevant of external factors.

In this case you can not use clock speed as your only metric anyway as they are two very different designs. It may still run on Zen cores, but the CPUs overall layout is vastly changed from the first iteration, to the point that you may as well be comparing it to an IBM chip. For a start we know there is more FPU logic on the die (2x if I'm remembering right) than previous Ryzen designs, which would obviously make a significant impact all on it's own.
At bold part... Provided RAM is not a bottleneck, which we know influences Ryzen heavily, because it influences how infinity fabric works.
05-20-2019 11:23 AM
EniGma1987
Quote: Originally Posted by mmonnin View Post
Pretty sure that's not the formula for wattage...
Speed is linear increase, voltage is exponential increase, and capacitance is mostly flat overall.
P = a * C * V2 * f


P is power (Watts). V is voltage (Volts). f is the frequency (cycles/sec), or equivalently, the number of seconds/cycle. And C is capacitance (Farads). A Farad is Coulombs/Volt, the charge divided by the EMF (Electromotive Force) potential.
https://software.intel.com/en-us/blo...obvious-pt-2-2
05-20-2019 11:03 AM
7850K
Quote: Originally Posted by KyadCK View Post
That is untrue. Performance alone (IPC x clock speed) is linear. If everything on a CPU is scaled up 40%, it will be 40% faster in everything.
years ago I recall reading the stilt talk about steamroller in Kaveri CPUs having a performance wall around 4.6Ghz. higher clockspeeds did not increase performance linearly, or even much at all. perhaps it is a rare case.
05-20-2019 09:36 AM
DNMock
Quote: Originally Posted by mmonnin View Post
Pretty sure that's not the formula for wattage...
he forgot amps.
05-20-2019 08:46 AM
mmonnin
Quote: Originally Posted by KyadCK View Post
That is untrue. Performance alone (IPC x clock speed) is linear. If everything on a CPU is scaled up 40%, it will be 40% faster in everything. Performance per watt is not linear, as both speed (linear) and voltage (linear) are multiplied for final wattage and combined with power gating may result in less performance than reported clock speed would otherwise say. There may also be bottlenecks in, say, RAM controllers or otherwise that limit a core's ability to work, but that is not the same as the CPU not scaling; the Core's performance (IPC) is irrelevant of external factors.

In this case you can not use clock speed as your only metric anyway as they are two very different designs. It may still run on Zen cores, but the CPUs overall layout is vastly changed from the first iteration, to the point that you may as well be comparing it to an IBM chip. For a start we know there is more FPU logic on the die (2x if I'm remembering right) than previous Ryzen designs, which would obviously make a significant impact all on it's own.
Pretty sure that's not the formula for wattage...
05-20-2019 07:34 AM
KyadCK
Quote: Originally Posted by NightAntilli View Post
Performance doesn't scale linearly. So having a lower clock might show a better IPC than it can achieve at higher clocks.
That is untrue. Performance alone (IPC x clock speed) is linear. If everything on a CPU is scaled up 40%, it will be 40% faster in everything. Performance per watt is not linear, as both speed (linear) and voltage (linear) are multiplied for final wattage and combined with power gating may result in less performance than reported clock speed would otherwise say. There may also be bottlenecks in, say, RAM controllers or otherwise that limit a core's ability to work, but that is not the same as the CPU not scaling; the Core's performance (IPC) is irrelevant of external factors.

In this case you can not use clock speed as your only metric anyway as they are two very different designs. It may still run on Zen cores, but the CPUs overall layout is vastly changed from the first iteration, to the point that you may as well be comparing it to an IBM chip. For a start we know there is more FPU logic on the die (2x if I'm remembering right) than previous Ryzen designs, which would obviously make a significant impact all on it's own.
05-20-2019 05:32 AM
NightAntilli
Quote: Originally Posted by Hwgeek View Post
EPYC Rome 32c VS EPYC 7551 in SiSoftware Processor Multi-Media
Benchmark, How they managed to boost IPC >50%?
Scary if it's true.
https://www.tomshardware.com/news/am...ecs,39373.html
https://ranker.sisoftware.co.uk/show...187f4c9f9&l=en
https://ranker.sisoftware.co.uk/show...385f6cbf3&l=en
Performance doesn't scale linearly. So having a lower clock might show a better IPC than it can achieve at higher clocks.
05-19-2019 11:50 AM
Defoler
Quote: Originally Posted by mothergoose729 View Post
Put another way, two EPYC 32 core CPUs are 30% faster than a single one of these 64 cores CPU, assuming linear scaling with clock speed.
I would say that is expected considering power and heat constraints.

While it might be slower, it doesn't matter when you have a system that needs 128/256 cores.
05-19-2019 11:27 AM
Hwgeek EPYC Rome 32c VS EPYC 7551 in SiSoftware Processor Multi-Media
Benchmark, How they managed to boost IPC >50%?
Scary if it's true.
https://www.tomshardware.com/news/am...ecs,39373.html
https://ranker.sisoftware.co.uk/show...187f4c9f9&l=en
https://ranker.sisoftware.co.uk/show...385f6cbf3&l=en
This thread has more than 10 replies. Click here to review the whole thread.

Posting Rules  
You may post new threads
You may post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off