Message boards :
Number crunching :
Ryzen 3000 and floating point
Message board moderation
Author | Message |
---|---|
Send message Joined: 15 Nov 14 Posts: 602 Credit: 24,371,321 RAC: 0 |
AMD has made significant changes to their CPU architecture which help deliver twice the throughput of their first generation Zen architecture. The major points include an entirely redesigned execution pipeline, major floating point advances which doubled the floating point registers to 256-bit and double bandwidth for load/store units. https://wccftech.com/amd-ryzen-3000-cpus-7nm-zen-2-first-epyc-rome-q3-2019-launch/ Is this of much use to the LHC projects? I have not seen much comparison between Intel and AMD here. |
Send message Joined: 15 Nov 14 Posts: 602 Credit: 24,371,321 RAC: 0 |
Very good. It is on my radar for next year, and I wanted a project that could use it. |
Send message Joined: 15 Nov 14 Posts: 602 Credit: 24,371,321 RAC: 0 |
Best yet that epyc keeps tempting you ... I am not sure how you feed them with fast enough memory. And the switch fabric (if that is what it uses) introduces latency. That might be OK for a server, but not so great for crunching. For the price, you could probably get at least two Ryzens, which would also give you an additional PCIe slot. So I think I will pass, but let me know if you get one. |
Send message Joined: 24 Oct 04 Posts: 1118 Credit: 49,729,593 RAC: 13,163 |
We do have lots of AMD's here but the only place I have seen comparisons was over at Einstein where we tend to compare CPU's and GPU's quite often. (we have Ryzens here too) |
Send message Joined: 2 May 07 Posts: 2090 Credit: 158,966,396 RAC: 124,782 |
|
Send message Joined: 18 Dec 15 Posts: 1688 Credit: 103,129,915 RAC: 120,329 |
https://lhcathome.cern.ch/lhcathome/cpu_list.phpthis list seems to be incomplete, though. Just out of curiosity, I looked for my 2 AMD Turion - they did not show up, for what reason ever. BTW: exactly the 2 machines with the Turions are producing these "missing hearbeat" errors once in while. All other PCs with Intel processors don't. Maybe it's a coincidence, or not. |
Send message Joined: 15 Nov 14 Posts: 602 Credit: 24,371,321 RAC: 0 |
Very nice. The numbers seem to be more accurate than most such comparisons that I have seen. They track well with what I see on my Ivy Bridge, Haswell, Coffee Lake, and Ryzen machines. Thanks. |
Send message Joined: 15 Jun 08 Posts: 2402 Credit: 225,594,838 RAC: 121,381 |
CPU Comparisons by the LHC Project : Hot Topic AMD Ryzen Threadripper 2990WX 32-Core Processor [Family 23 Model 8 Stepping 2] Avg. cores/computer: 205.71 There must be somebody out there running AMD Threadripper samples from next year. Number of computers? Should better be called "Number of BOINC clients". I guess I'm not the only one running more than 1 client on the same computer. AMD Ryzen Threadripper 2950X 16-Core Processor [Family 23 Model 8 Stepping 2] Avg. cores/computer: 18.17 18.17 > 16 looks like HT on/off, but only at first sight as "Avg. cores/computer" should better be called "Avg. # reported cores/BOINC client". This can be tweaked by at least 2 different client settings. GFLOPS/core, GFLOPs/computer If anybody believes in a 2 digit accuracy I suggest to run the BOINC benchmarks - while the computer is idle/under full load - while the BOINC client is set to use 1, 2, 3, ..., all available cores - on different BOINC client versions - on different OSs - while we have full moon, half ... (ok, this is not serious) |
©2024 CERN