AMD Bulldozer FMA4 app


Message boards : Number crunching : AMD Bulldozer FMA4 app

Message board moderation

To post messages, you must log in.
1 · 2 · 3 · Next
AuthorMessage
Profile Crunch3r
Avatar

Send message
Joined: 19 Jun 12
Posts: 21
Credit: 107,293,560
RAC: 0
Message 2809 - Posted: 17 Mar 2014, 19:38:50 UTC
Is someone interested in experimenting with a app that uses the AMD fma4 instruction set ?

FWIW, i've been running it for a couple of weeks now at it works (valid results).

If someone is interested i could upload it.

Let me know.

Join BOINC United now!
ID: 2809 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
frankhagen

Send message
Joined: 18 Jun 12
Posts: 15
Credit: 5,027,400
RAC: 0
Message 2812 - Posted: 17 Mar 2014, 20:51:33 UTC - in response to Message 2809.  
Is someone interested in experimenting with a app that uses the AMD fma4 instruction set ?

FWIW, i've been running it for a couple of weeks now at it works (valid results).

If someone is interested i could upload it.

Let me know.


AAAH - noob alert! ;)

currently i got no AMD-FX at hand, but would you throw in some performance-numbers?
ID: 2812 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Crunch3r
Avatar

Send message
Joined: 19 Jun 12
Posts: 21
Credit: 107,293,560
RAC: 0
Message 2813 - Posted: 17 Mar 2014, 21:01:41 UTC - in response to Message 2812.  


AAAH - noob alert! ;)

currently i got no AMD-FX at hand, but would you throw in some performance-numbers?


I wish i could throw in some numbers but unfortuantely, my cooling is not efficient enought.. (meaning my fx6100 keeps throttling down to 1.4 GHz)... So it's a guessing game.

So, the only available option is testing it on some bullozers that can handle the heat ;)

Join BOINC United now!
ID: 2813 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile HA-SOFT, s.r.o.
Project developer
Project tester

Send message
Joined: 21 Dec 12
Posts: 176
Credit: 134,883,341
RAC: 2,339
Message 2828 - Posted: 18 Mar 2014, 14:11:26 UTC - in response to Message 2809.  
Is someone interested in experimenting with a app that uses the AMD fma4 instruction set ?

FWIW, i've been running it for a couple of weeks now at it works (valid results).

If someone is interested i could upload it.

Let me know.


Did you recode the app or change compiler settings only? I have FMA2 app and not difference between AVX and FMA2.
ID: 2828 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Crunch3r
Avatar

Send message
Joined: 19 Jun 12
Posts: 21
Credit: 107,293,560
RAC: 0
Message 2829 - Posted: 18 Mar 2014, 16:43:47 UTC - in response to Message 2828.  
Is someone interested in experimenting with a app that uses the AMD fma4 instruction set ?

FWIW, i've been running it for a couple of weeks now at it works (valid results).

If someone is interested i could upload it.

Let me know.


Did you recode the app or change compiler settings only? I have FMA2 app and not difference between AVX and FMA2.


I rewrote the code.

Join BOINC United now!
ID: 2829 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile HA-SOFT, s.r.o.
Project developer
Project tester

Send message
Joined: 21 Dec 12
Posts: 176
Credit: 134,883,341
RAC: 2,339
Message 2830 - Posted: 18 Mar 2014, 18:15:32 UTC - in response to Message 2829.  
Is there any speedup over avx?
ID: 2830 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Crunch3r
Avatar

Send message
Joined: 19 Jun 12
Posts: 21
Credit: 107,293,560
RAC: 0
Message 2861 - Posted: 21 Mar 2014, 19:18:41 UTC - in response to Message 2830.  
Is there any speedup over avx?


AVX was always slower than the sse2 or sse3 app. Not even using the intel compiler could fix that.

So i've switched to gcc 4.8.2 and recoded the parts of the app that could make use of amds fma4.

The real question is, if the app is faster than the sse2 one (fastest of them all), however , as i've mentioned earlier, my cooling sucks and my fx 6100 keeps trottling so there's no way of me telling what the differece is.

Join BOINC United now!
ID: 2861 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
cyrusNGC_224@P3D

Send message
Joined: 1 Apr 13
Posts: 37
Credit: 153,496,537
RAC: 0
Message 2862 - Posted: 21 Mar 2014, 22:39:31 UTC - in response to Message 2861.  
I have a AMD processor, which supported FMA4 and does not clock down.
ID: 2862 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
[TA]Assimilator1
Avatar

Send message
Joined: 24 Aug 13
Posts: 111
Credit: 30,963,035
RAC: 9,479
Message 2863 - Posted: 22 Mar 2014, 0:20:48 UTC - in response to Message 2861.  

Last modified: 22 Mar 2014, 0:21:19 UTC
Is there any speedup over avx?


AVX was always slower than the sse2 or sse3 app. Not even using the intel compiler could fix that.

So i've switched to gcc 4.8.2 and recoded the parts of the app that could make use of amds fma4.

The real question is, if the app is faster than the sse2 one (fastest of them all), however , as i've mentioned earlier, my cooling sucks and my fx 6100 keeps trottling so there's no way of me telling what the differece is.

Couldn't you underclock it to 1.4 GHz & run the SSE2 app to compare times?
Team AnandTech - SETI@H, Muon1 DPAD, Folding@H, MilkyWay@H, Asteroids@H, LHC@H, POGS, Rosetta@H, Einstein@H,DHPE & CPDN

Main rig - Ryzen 3600, 32GB DDR4 3200, RX 580 8GB, Win10
2nd rig - i7 4930k @4.1 GHz, 16GB DDR3 1866, HD 7870 XT 3GB(DS), Win7
ID: 2863 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Falconet

Send message
Joined: 23 Oct 12
Posts: 18
Credit: 59,850
RAC: 15
Message 2924 - Posted: 21 Apr 2014, 20:34:03 UTC
Have you tested the app any further?
ID: 2924 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
alexander

Send message
Joined: 28 Apr 13
Posts: 87
Credit: 26,716,176
RAC: 151
Message 2989 - Posted: 7 May 2014, 18:55:55 UTC - in response to Message 2809.  

Last modified: 7 May 2014, 18:56:25 UTC
Is someone interested in experimenting with a app that uses the AMD fma4 instruction set ?

FWIW, i've been running it for a couple of weeks now at it works (valid results).

If someone is interested i could upload it.

Let me know.


Hi,
I've added my A10-7700
https://asteroidsathome.net/boinc//show_host_detail.php?hostid=88982
which is fma able. Unfortunately BOINC does not say fma4, but since it is a brand new device I suggest it is.
One info is still missing; your application, is it for Windows or Linux? If Windows is good I would be glad to test it!

Alexander
ID: 2989 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile BilBg
Avatar

Send message
Joined: 19 Jun 12
Posts: 221
Credit: 623,640
RAC: 0
Message 2991 - Posted: 8 May 2014, 1:18:48 UTC - in response to Message 2989.  

Last modified: 8 May 2014, 1:25:45 UTC

The link is this:
http://asteroidsathome.net/boinc/show_host_detail.php?hostid=88984

Yes, your AMD A10-7700K have FMA3 and FMA4
http://www.cpu-world.com/CPUs/Bulldozer/AMD-A10-Series%20A10-7700K.html

Maximum operating temperature 72.4°C
You can use TThrottle to control the temperature.

You can also use SIV - System Information Viewer to find CPU Features for certain:
http://rh-software.com/
http://rh-software.com/downloads/siv.zip

Click on the right-bottom the button that say [MMX+ .....]







- ALF - "Find out what you don't do well ..... then don't do it!" :)
ID: 2991 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Falconet

Send message
Joined: 23 Oct 12
Posts: 18
Credit: 59,850
RAC: 15
Message 2998 - Posted: 9 May 2014, 19:01:42 UTC

Last modified: 9 May 2014, 19:05:17 UTC
I have tested the app but only ran 4 tasks with it.

http://asteroidsathome.net/boinc/results.php?hostid=13438&offset=0&show_names=0&state=4&appid=

Against an i7-4770, my A8-6500 3.5GHZ took 10,996.37 CPU time vs 9,167.34 CPU time on the i7.

Against an i7-3770, mine took 9,752.09 CPU time vs 7,808.02 CPU time on the i7.

Against an i5-4570, mine took 10,705.40 CPU time vs 7,551.07 CPU time on the i5.

The other task was ran against a GPU.

So, they take longer but the i7's and i5 have much larger Floating Point and Integer figures than mine. I don't even know what clock they are running.


My A8-6500:

Measured floating point speed 1523.05 million ops/sec
Measured integer speed 6243.76 million ops/sec

i5-4570:

Measured floating point speed 3496.34 million ops/sec
Measured integer speed 14860.04 million ops/sec

i7-3770:

Measured floating point speed 3688.48 million ops/sec
Measured integer speed 13060.77 million ops/sec

i7-4770:

Measured floating point speed 3563.17 million ops/sec
Measured integer speed 14800.09 million ops/sec


I think this needs more testing. Hope the project scientists are interested in testing this.
ID: 2998 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
alexander

Send message
Joined: 28 Apr 13
Posts: 87
Credit: 26,716,176
RAC: 151
Message 2999 - Posted: 9 May 2014, 20:44:32 UTC
Hi Falconet,

thx for the info, but you are running two A8, one with Linux and one with Windows. I still do not know: is it a linux or a windows app? And if windows, where can I get it? And do I need a.xml file?

Alexander
ID: 2999 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Falconet

Send message
Joined: 23 Oct 12
Posts: 18
Credit: 59,850
RAC: 15
Message 3000 - Posted: 9 May 2014, 20:58:55 UTC - in response to Message 2999.  
It is for Windows. I only have one A8-6500.
I have a link but I don't want give it without the author's permission.
An app_info file is already included and configured (only needs to be placed in the project's folder).
ID: 3000 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
alexander

Send message
Joined: 28 Apr 13
Posts: 87
Credit: 26,716,176
RAC: 151
Message 3001 - Posted: 9 May 2014, 21:20:29 UTC - in response to Message 3000.  
It is for Windows. I only have one A8-6500.
I have a link but I don't want give it without the author's permission.
An app_info file is already included and configured (only needs to be placed in the project's folder).


I've sent a PM to Cruncher, let's wait for the answer.
ID: 3001 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile HA-SOFT, s.r.o.
Project developer
Project tester

Send message
Joined: 21 Dec 12
Posts: 176
Credit: 134,883,341
RAC: 2,339
Message 3003 - Posted: 9 May 2014, 21:54:10 UTC - in response to Message 2998.  
am I right, that there is no speedup over our apps? I have tested FMA app long time ago on i5-4670 and there was no speedup over avx app.
ID: 3003 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Falconet

Send message
Joined: 23 Oct 12
Posts: 18
Credit: 59,850
RAC: 15
Message 3004 - Posted: 9 May 2014, 21:55:25 UTC - in response to Message 3003.  

Last modified: 9 May 2014, 21:55:35 UTC
I am not sure. It may reduce the gap between AMD and Intel but more testing is required.
ID: 3004 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 1 Jan 14
Posts: 300
Credit: 32,052,075
RAC: 14,628
Message 3007 - Posted: 10 May 2014, 10:56:45 UTC - in response to Message 3004.  
I am not sure. It may reduce the gap between AMD and Intel but more testing is required.


That would be NICE, reducing the gap I mean.
ID: 3007 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Falconet

Send message
Joined: 23 Oct 12
Posts: 18
Credit: 59,850
RAC: 15
Message 3009 - Posted: 10 May 2014, 11:02:28 UTC - in response to Message 3007.  
I am not sure. It may reduce the gap between AMD and Intel but more testing is required.


That would be NICE, reducing the gap I mean.


Indeed it would.
ID: 3009 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
1 · 2 · 3 · Next

Message boards : Number crunching : AMD Bulldozer FMA4 app