Thinking about AMD GPU

Message boards : Number crunching : Thinking about AMD GPU
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile dcappello
Avatar

Send message
Joined: 3 Apr 99
Posts: 261
Credit: 170,969,320
RAC: 0
United States
Message 1090977 - Posted: 27 Mar 2011, 19:07:45 UTC

Of course I have been reading up on the AMD vs NVIDIA wars - all my rigs are running NVIDIA - but perhaps I need to try something from AMD - are the optimized apps ready to go? What is everybody thinking along these lines?

http://www.9to5mac.com/58014/amd-throws-down-the-gauntlet-to-nvidia-on-graphics-speed-claims/
ID: 1090977 · Report as offensive
Cruncher-American Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor

Send message
Joined: 25 Mar 02
Posts: 1513
Credit: 370,893,186
RAC: 340
United States
Message 1090981 - Posted: 27 Mar 2011, 19:15:56 UTC - in response to Message 1090977.  

If you want to run more than 2 GPUs, you are out of luck - at least, I could not get 4 working HD 5550s to run on the same MB. Any more than 2, and Win 7 (and/or Vista) 64bit hung during install of drivers. This, on same MB I can run 4 GT240s with no problems.
ID: 1090981 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1090986 - Posted: 27 Mar 2011, 19:34:30 UTC
Last modified: 27 Mar 2011, 19:34:59 UTC

Apps have been released for MB & AP on the Lunatics download site. Here are the local topics on the releases.
MultiBeam application for ATi GPUs released
New ATI OpenCL-based AstroPulse (rev516) released

Looking at the Hierarchy Chart from Tom's Hardware it looks like my HD4870 compares to a GTX 260 or GTX 460. Average tasks, 0.42 AR, run about 45-50 minutes for me on my HD4870. So you would want want to compare those cards I suppose. I see your machine with 2 GTX 260's runs around 15 minutes for average tasks.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1090986 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1090993 - Posted: 27 Mar 2011, 19:53:50 UTC

AMD came to GPGPU with very big delay. Their HD4xxx not actually comparable with NV GPUs in GP-GPU area.
If you will buy smth new, get at least HD5xxx. Even cheaper/low end HD5xxx can be better for SETI in long term. They have fast local memory available, further app optimization will be targeted to maximize local memory usage.

HD4870 is fast in raw computing power, but mostly stalled wayting for memory in current SETI apps...
ID: 1090993 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1090996 - Posted: 27 Mar 2011, 20:00:19 UTC - in response to Message 1090993.  

AMD came to GPGPU with very big delay. Their HD4xxx not actually comparable with NV GPUs in GP-GPU area.
If you will buy smth new, get at least HD5xxx. Even cheaper/low end HD5xxx can be better for SETI in long term. They have fast local memory available, further app optimization will be targeted to maximize local memory usage.

HD4870 is fast in raw computing power, but mostly stalled wayting for memory in current SETI apps...

I will fire up the HD5750 in my HTPC & see what kind of numbers it makes. It had a very small cooling setup so I would not want to leave it running Seti@Home constantly.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1090996 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1091075 - Posted: 28 Mar 2011, 2:45:33 UTC - in response to Message 1090996.  

AMD came to GPGPU with very big delay. Their HD4xxx not actually comparable with NV GPUs in GP-GPU area.
If you will buy smth new, get at least HD5xxx. Even cheaper/low end HD5xxx can be better for SETI in long term. They have fast local memory available, further app optimization will be targeted to maximize local memory usage.

HD4870 is fast in raw computing power, but mostly stalled wayting for memory in current SETI apps...

I will fire up the HD5750 in my HTPC & see what kind of numbers it makes. It had a very small cooling setup so I would not want to leave it running Seti@Home constantly.

My HD5750 has run though several .37 tasks in about 40-45 min. So for .42 tasks it might be closer to 30 min. I did a few 2 at once and it just seemed to run 2 twice as slow. So no advantage on this card for more then 1 it seems. It is actually running cooler then I expected. Only 59º-60ºC for 1 tasks and then around 63ºC for 2 at once. with the fan running around 40%. I'll be checking in on 5255585 later to see how it handles other AR tasks.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1091075 · Report as offensive
Profile skildude
Avatar

Send message
Joined: 4 Oct 00
Posts: 9541
Credit: 50,759,529
RAC: 60
Yemen
Message 1091086 - Posted: 28 Mar 2011, 3:15:28 UTC - in response to Message 1091075.  

Astropulse finishes in about 90 minutes on a 5850 using the latest Lunatics app


In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope
ID: 1091086 · Report as offensive
halfempty
Avatar

Send message
Joined: 2 Jun 99
Posts: 97
Credit: 35,236,901
RAC: 114
United States
Message 1091119 - Posted: 28 Mar 2011, 6:41:30 UTC

On the average MB task nVidia is faster. My HD5850 takes the same amount of time, about 22 minutes, than the cheaper lower power GTS450 it replaced. The HD5750 I sold not that long ago, which is priced in the GTS450 range, used to take about 34 minutes for the same type of task. On the plus ATi's don't choke on VLARs, and like skildude said it can crank out AP tasks in a very reasonable time.

MilkyWay is a different story. The GTS450 took about 20 minutes to crank out a separation task, the HD5850 does one in about 2 minutes. I do 90% Seti and 10% MW yet my RAC is twice as high on MW since I got the 5850.
ID: 1091119 · Report as offensive
Profile Gecko
Volunteer tester
Avatar

Send message
Joined: 17 Nov 99
Posts: 454
Credit: 6,946,910
RAC: 47
United States
Message 1091257 - Posted: 28 Mar 2011, 20:50:53 UTC - in response to Message 1091119.  
Last modified: 28 Mar 2011, 20:57:22 UTC


MilkyWay is a different story. The GTS450 took about 20 minutes to crank out a separation task, the HD5850 does one in about 2 minutes. I do 90% Seti and 10% MW yet my RAC is twice as high on MW since I got the 5850.


Besides the rather apples/oranges performance disparity between the categories of cards w/ HD5850 = 2088 SP Gflops vs. GTS450 = 601 SP Gflops, keep in mind that MW uses double-precision processing (DP) and that 5850 DP performance is 1/5th it's SP flop rating, or 418 Glops. On the GTS450, DP performance is 1/12 of SP = 50 Gflops.

So, your 5850 has 8x the DP processing ability of the 450 vs. the 3.5x SP advantage it has at Seti that can't be fully exploited yet at the present development stage of ATI apps. Your ATI card and MW are perfect crunching partners.

However, The workstation version of the GTS450, the Quadro 2000, ( @$450.00 ) does DP at 1/2 of SP, so is rated for 240 DP Gflops.

Rough numbers would imply this card crunching MW @ 8 1/2 mins using 192 shaders @ 625Mhz core, 1250 Mhz shader clock.

Oh, & it's a relative power-miser too w/ a TDP of only 62 watts. Doesn't even use a 6-pin connector.
ID: 1091257 · Report as offensive
halfempty
Avatar

Send message
Joined: 2 Jun 99
Posts: 97
Credit: 35,236,901
RAC: 114
United States
Message 1091673 - Posted: 30 Mar 2011, 16:37:04 UTC - in response to Message 1091257.  

Everything you say makes perfect sense. My 5850 was more of an emotional rather than a logical purchase. I've only recently learned about the wonders of GPU computing and have been using ATi's for many years, so when my old system died in October I put this one together and got a 5830. Much to my dismay it didn't fit in the case, so I returned that one.

Got a great deal on a 5750 and I went for that one without doing any research. That's how I learned which cards can't do DP, and at the time there was no MB ATi app. Ah, the joys of impulse purchases. Saved up and bought a GTS450, the best DP card I could afford and would fit, then sold the 5750. All was well in crunching land for a short while.

Not long after that Raistmer, through his much appreciated dedication, came out with the ATi MB app. Then I found a 5850 that would fit in my case without the use of a hacksaw. My fate was sealed.

So you see, while on a spec sheet the cards are dissimilar, in my warped and convoluted mind they are all "related".
ID: 1091673 · Report as offensive

Message boards : Number crunching : Thinking about AMD GPU


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.