New FMA application released


Message boards : News : New FMA application released

Message board moderation

To post messages, you must log in.
AuthorMessage
Profile Kyong
Project administrator
Project developer
Project tester
Project scientist
Avatar

Send message
Joined: 9 Jun 12
Posts: 584
Credit: 52,667,664
RAC: 0
Message 7385 - Posted: 20 Nov 2022, 20:04:38 UTC
We are happy to present you another set of applications that will utilize those CPUs, which possess the FMA instruction set!

It was built to support both Linux and Windows OS with 64bit architecture.
What must be taken into account with these applications is that depending on the CPU architecture, generation, model, version, speed and number of utilized cores, in some cases the FMA applications may run slower than the corresponding AVX ones because of how those instructions are handled on different architectures, details that I'll not going further here as there is a lot of information on Internet. Also, there could be another case using even the same FMA application. For instance, there could be a situation where on the same particular system, having your preferences restrict the use of just a single core of, let's say, an Intel(R) Xeon(R) W Processor Xeon W-2195 of the Skylake-W Architecture, that single core will run much faster, close to processors' Turbo frequency of 4.3GHz, than if your configuration allows the client to utilize let's say 12 or more cores. In the second case those cores will run close to the processors' base frequency of 2.3GHz, depending on their actual number, which will result drastically in lowering of the application performance. Take a look at this article where under the "Per Core Turbo Data" chapter you will find explanation about how the Turbo ratio limits works.

Still, while the Boinc server is capable of finding the best performed application for every particular system, taking into account multiple factors, and after a while it will start sending the right one for every particular system. Which means that even if your CPU is equipped with the FMA engine it still might receive an AVX tasks and there is nothing to be concerned.

So happy crunching and thank you for your support!

Radim Vančo (FoxKyong)
ID: 7385 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 1 Jan 14
Posts: 300
Credit: 32,204,049
RAC: 8,643
Message 7404 - Posted: 21 Nov 2022, 15:16:39 UTC
Thank you for doing this
ID: 7404 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Drago75

Send message
Joined: 4 May 20
Posts: 7
Credit: 5,110,082
RAC: 18,556
Message 7529 - Posted: 30 Nov 2022, 20:36:37 UTC
What is the advantage of the FMA tasks? Are they faster than avx? Or more accurate?
ID: 7529 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Drago75

Send message
Joined: 4 May 20
Posts: 7
Credit: 5,110,082
RAC: 18,556
Message 7530 - Posted: 30 Nov 2022, 20:36:40 UTC

Last modified: 30 Nov 2022, 20:41:02 UTC
sorry, double posting
ID: 7530 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Keith Myers
Avatar

Send message
Joined: 16 Nov 22
Posts: 98
Credit: 52,380,237
RAC: 355,885
Message 7531 - Posted: 30 Nov 2022, 20:44:29 UTC - in response to Message 7530.  
Depends on the cpu architecture and features set. Can be more efficient than AVX2.

A proud member of the OFA (Old Farts Association)
ID: 7531 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
magic_sam

Send message
Joined: 16 Nov 22
Posts: 18
Credit: 6,962,168
RAC: 2,876
Message 7577 - Posted: 16 Dec 2022, 17:49:24 UTC

Last modified: 16 Dec 2022, 17:52:13 UTC
Hi all,

From this project home page:

Still, while the Boinc server is capable of finding the best performed application for every particular system, taking into account multiple factors, and after a while it will start sending the right one for every particular system. Which means that even if your CPU is equipped with the FMA engine it still might receive an AVX tasks and there is nothing to be concerned.


How long does it take for the server to take the CPU flags on the computing nodes into account ? I've been crunching for several weeks now and I have yet to run a single FMA task. CPU is some Ryzen 7 4800H (Zen 2), and I ran FMA tasks in the past for a different BOINC project (TN-Grid).

Cheers, Sam
ID: 7577 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
magic_sam

Send message
Joined: 16 Nov 22
Posts: 18
Credit: 6,962,168
RAC: 2,876
Message 7578 - Posted: 16 Dec 2022, 17:51:23 UTC

Last modified: 16 Dec 2022, 17:51:56 UTC
Sorry double post
ID: 7578 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
stfn

Send message
Joined: 24 May 21
Posts: 15
Credit: 3,156,937
RAC: 12,417
Message 7592 - Posted: 22 Dec 2022, 8:55:02 UTC

Last modified: 22 Dec 2022, 8:55:15 UTC
I'm also curious about the choosing of the instruction set, because I have a Ryzen 3700X, and the only app version I am getting is sse2. Shouldn't I be getting sse3 or fma?
my blog about raspberry pis and astrophotography
ID: 7592 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Georgi Vidinski
Volunteer moderator
Project administrator
Project developer
Project tester
Avatar

Send message
Joined: 22 Nov 17
Posts: 159
Credit: 13,180,466
RAC: 58
Message 7604 - Posted: 23 Dec 2022, 5:24:01 UTC - in response to Message 7592.  
It has been discussed here already: https://asteroidsathome.net/boinc/forum_thread.php?id=793&postid=6541
“The good thing about science is that it's true whether or not you believe in it.” ― Neil deGrasse Tyson
ID: 7604 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Aurum

Send message
Joined: 11 Jul 17
Posts: 22
Credit: 333,302,879
RAC: 21
Message 7710 - Posted: 16 Jan 2023, 7:18:17 UTC - in response to Message 7385.  
...finding the best performed application for every particular system...
Radim Vančo (FoxKyong)
How do you actually define what's the best instruction set? I think it should be the most energy efficient. See, for example:
Thermal design power and vectorized instructions behavior, Amina Guermouche & Anne-Cécile Orgerie, CONCURRENCY & COMPUTATION: PRACTICE & EXPERIENCE, Feb 2021.
https://hal.archives-ouvertes.fr/hal-03185821/document
ID: 7710 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Aurum

Send message
Joined: 11 Jul 17
Posts: 22
Credit: 333,302,879
RAC: 21
Message 7723 - Posted: 27 Jan 2023, 14:11:03 UTC - in response to Message 7604.  
It has been discussed here already: https://asteroidsathome.net/boinc/forum_thread.php?id=793&postid=6541
Says nothing about energy efficiency. Just a single WU race mentality. But then you're not paying for the electricity.
ID: 7723 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Aurum

Send message
Joined: 11 Jul 17
Posts: 22
Credit: 333,302,879
RAC: 21
Message 7724 - Posted: 27 Jan 2023, 14:12:17 UTC - in response to Message 7531.  
Depends on the cpu architecture and features set. Can be more efficient than AVX2.
Energy efficiency or race speed?
ID: 7724 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : News : New FMA application released