GPU FLOPS: Theory vs Reality

Message boards : Number crunching : GPU FLOPS: Theory vs Reality
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 5 · 6 · 7 · 8 · 9 · 10 · 11 . . . 20 · Next

AuthorMessage
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1958961 - Posted: 7 Oct 2018, 0:36:56 UTC - in response to Message 1958956.  

Curious Grant, do you have VAT like most Commonwealth countries?
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1958961 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 1958965 - Posted: 7 Oct 2018, 0:51:20 UTC

High taxes makes the most sense. But you’d think it wouldn’t be so bad, Aus is much closer to China/Taiwan where this stuff comes from than the US.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 1958965 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1958966 - Posted: 7 Oct 2018, 0:54:34 UTC - in response to Message 1958961.  

Curious Grant, do you have VAT like most Commonwealth countries?

Yep, although here it's called GST (Goods and Services Tax, different name, same thing).
The price gouging is above & beyond taxes, it's been going on for years.

One classic example was an AC/DC boxed set of CDs. Australian band, and many of their albums (all of the early ones) were recorded & pressed here in Australia. When the boxed set was released it was cheaper to buy it from overseas & pay exorbitant postage rates, exchange rates, and import duty, than to buy it locally. From memory it was something like less than half the price.

And the fact the $US has been climbing for a while now & our $ falling just makes things worse than usual (but even when allowing for taxes & exchange rates, the prices are still out of all proportion).
Grant
Darwin NT
ID: 1958966 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1958970 - Posted: 7 Oct 2018, 1:01:58 UTC - in response to Message 1958966.  

Thanks for the info. And the typical example of how GST is applied. I figured it wasn't due to higher transportation costs since most of the boards are made in China or Taiwan so closer to you guys than to us here in the States.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1958970 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1958973 - Posted: 7 Oct 2018, 1:12:12 UTC - in response to Message 1958970.  

I couldn't remember what typical things cost there. Been too many years since I visited back in the 80's. I always say you can get an idea of what the cost of goods is in any country by the price of a loaf of bread as your benchmark.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1958973 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1958977 - Posted: 7 Oct 2018, 1:13:29 UTC - in response to Message 1958965.  

High taxes makes the most sense.

The GST is only 10%
There are company taxes, but most of the larger companies don't pay then due to creative accounting, only the smaller companies tend to pay them.
There are other taxes, but they are mostly on Excise, customs, Fringe benefits (goods in lieu of pay for high level staff), Luxury vehicles
The only state taxes are Payroll tax, stamp duty on conveyances, some land & gambling taxes. Local taxes are Municipal rates & that's it.

So the only taxes that really affect the pricing of most goods are the GTS, some payroll tax (minimal impact), some company taxes (maybe, still minimal impact), Municipal rates.
Grant
Darwin NT
ID: 1958977 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 1959012 - Posted: 7 Oct 2018, 4:37:06 UTC - in response to Message 1958849.  
Last modified: 7 Oct 2018, 4:40:43 UTC

I'm sure that Vega cards are NOT running "opencl_ati5_nocal" applications. They would be running the same exact SoG r3584 code that the Nvidia cards run. Same source code. Just compiled for ATI hardware versus Nvidia hardware. OpenCL is platform neutral.


I know there are a bunch of variations on the AMD gpu app names, so I grabbed the closest one from an AMD laptop I am running.

Which one is the "best" one again? I think I found a thread where I asked questions about this so the Vega cards should be running "opencl_ati5_SoG" tasks?

Thank you.

Tom
A proud member of the OFA (Old Farts Association).
ID: 1959012 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34253
Credit: 79,922,639
RAC: 80
Germany
Message 1959037 - Posted: 7 Oct 2018, 8:28:55 UTC - in response to Message 1959012.  

I'm sure that Vega cards are NOT running "opencl_ati5_nocal" applications. They would be running the same exact SoG r3584 code that the Nvidia cards run. Same source code. Just compiled for ATI hardware versus Nvidia hardware. OpenCL is platform neutral.


I know there are a bunch of variations on the AMD gpu app names, so I grabbed the closest one from an AMD laptop I am running.

Which one is the "best" one again? I think I found a thread where I asked questions about this so the Vega cards should be running "opencl_ati5_SoG" tasks?

Thank you.

Tom


For AMD cards it doesn`t really matter which version you are using.
The last revisions only has changes for nvidia cards.


With each crime and every kindness we birth our future.
ID: 1959037 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1959098 - Posted: 7 Oct 2018, 17:55:49 UTC - in response to Message 1959012.  

I just dropped over to Mike's World for his Seti application download page and saw that there was a link for Seti ATI r3584_SoG and know that version is the latest developed by Raistmer.
http://mikesworld.eu/download.html
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1959098 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 1959156 - Posted: 7 Oct 2018, 23:18:32 UTC - in response to Message 1959098.  

I just dropped over to Mike's World for his Seti application download page and saw that there was a link for Seti ATI r3584_SoG and know that version is the latest developed by Raistmer.
http://mikesworld.eu/download.html


Thank you both for the additional information.

Tom
A proud member of the OFA (Old Farts Association).
ID: 1959156 · Report as offensive
Profile StFreddy
Avatar

Send message
Joined: 4 Feb 01
Posts: 35
Credit: 14,080,356
RAC: 26
Hungary
Message 1963123 - Posted: 3 Nov 2018, 12:49:06 UTC
Last modified: 3 Nov 2018, 12:50:17 UTC

Hi,
just wanted to share my experience with AMD vega56.
Got mine few days ago, it is a PowerColor Red Dragon 56. Undervolted the GPU by 0,1V @ P6 and P7 states. (1.2V -> 1.1V @ full load)
OS: Windows10
CPU: Ryzen 1800x @ 3.9Ghz (Core Voltage 1.25V, ~115W peak power draw during seti while crunching on 12 threads)
1 CPU core is dedicated to the GPU in app_info.xml.
Screenshot during GPU crunching, while the computer is not crunching CPU tasks, just 1 GPU task (blc01):



https://imgur.com/EvJ0PLj

As you can see, 1 cpu core is utilized 50% max. but it fluctuates, so does GPU core utilization which rarely reaches 100%.
GPU power consumption max. ~158W, but usually lower, it averages at about 135-140W I think.
1 blc01 task takes 300-320 seconds to complete.
I is not bad, but the CPU is more efficient - at least in windows10.

I wonder if vega would be more efficient under linux? Are the special apps available for Linux are better optimized?
ID: 1963123 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 1963137 - Posted: 3 Nov 2018, 14:45:21 UTC - in response to Message 1963123.  


I wonder if vega would be more efficient under linux? Are the special apps available for Linux are better optimized?



I am told you won't see any gpu speed up by going to Linux. The CUDA91 gpu app is specific to the Nvidia hardware. There is an AMD specific cpu app for the SSE 4.1 but it might be nearly the same as the Windows in Lunatics distro.

There is a new beta test on an windows AMD cpu app at "Mikes World" that you could download. I am also told that one of the Lunatics cpu app AMD runs the AVX(?) instruction set better than the SSE instruction set so you could re-run the Lunatics installer and see if that improves things.

I am told the Vega 56/Vega 64 competes in the Gtx 1070 range in terms of performance so you would be looking to compare with gtx 1070's on the LeaderBoard. For comparison purposes you will have to search lower than the very top since most of the very top are running Linux.

Are you letting the Video card self turbo boost or are the voltages you are reporting the result of tweaking? What kind of command line parameters are you using for the gpu tasks?

HTH,
Tom
A proud member of the OFA (Old Farts Association).
ID: 1963137 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1963157 - Posted: 3 Nov 2018, 16:31:44 UTC

The Lunatics installer does not do what it is supposed to do when you choose the suggestion about running SSE41 or SSE42 on AMD cpus. There is no SSE41 or SSE42 app in the installer. The choice is a remnant from the past with earlier apps. If you choose to install either of the SSE41/2 apps, what is really installed is the SSE3 app. The SSE3 app is the stock app like what the normal BOINC scheduler dishes out. The only better performing choice in the Lunatics installer is to choose the AVX application for cpu.

For Linux OS, the Lunatics website has both SSE41 and SSE42 apps available for manual installation. On Ryzen AMD cpus, I find the SSE41 app to be faster than AVX.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1963157 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34253
Credit: 79,922,639
RAC: 80
Germany
Message 1963167 - Posted: 3 Nov 2018, 17:34:08 UTC

Correction.

All apps included in the Lunatics Installer are optimized apps.
When you choose sse4.1 or 4.2 you get MB8_win_x64_SSE3_VS2008_r3330.exe which is much faster then stock.


With each crime and every kindness we birth our future.
ID: 1963167 · Report as offensive
Profile StFreddy
Avatar

Send message
Joined: 4 Feb 01
Posts: 35
Credit: 14,080,356
RAC: 26
Hungary
Message 1963279 - Posted: 4 Nov 2018, 16:34:16 UTC - in response to Message 1963137.  


I wonder if vega would be more efficient under linux? Are the special apps available for Linux are better optimized?



I am told you won't see any gpu speed up by going to Linux. The CUDA91 gpu app is specific to the Nvidia hardware. There is an AMD specific cpu app for the SSE 4.1 but it might be nearly the same as the Windows in Lunatics distro.

There is a new beta test on an windows AMD cpu app at "Mikes World" that you could download. I am also told that one of the Lunatics cpu app AMD runs the AVX(?) instruction set better than the SSE instruction set so you could re-run the Lunatics installer and see if that improves things.

I am told the Vega 56/Vega 64 competes in the Gtx 1070 range in terms of performance so you would be looking to compare with gtx 1070's on the LeaderBoard. For comparison purposes you will have to search lower than the very top since most of the very top are running Linux.

Are you letting the Video card self turbo boost or are the voltages you are reporting the result of tweaking? What kind of command line parameters are you using for the gpu tasks?

HTH,
Tom


Hi Tom,

thanks for your answer. I am using the latest optimized apps. I downloaded them from "Mikes World". So my Ryzen CPU is using: MB8_win_x64_AVX_VS2017_r3714_Vlock
Vega will boost automatically if the actual temperature and power conditions allow it to do so. So if I undervolt the GPU, I give the card bigger headroom to turbo itself because of lower power consumption and lower temperatures. The factory clock speed of the GPU is 1580MHz, but it boosts up to to 1630. More tweaking would be possible, but I don't have too much time for this at the moment.
According to the command line parameters, I didn't change anything. It must be some default.
ID: 1963279 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 1963292 - Posted: 4 Nov 2018, 17:42:03 UTC - in response to Message 1963279.  

According to the command line parameters, I didn't change anything. It must be some default.


There are some command line parameters mentioned in the "readme" for various gpus that has some offerings for high end AMD video cards that MIGHT make a difference.
Some of the parameters would possibly crossover from what helps with NVidia's. Others appear to be specific to the AMD cards.

Since the community is so lopsidedly NVidia it is hard to get advice from the AMD card users.

Tom
A proud member of the OFA (Old Farts Association).
ID: 1963292 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34253
Credit: 79,922,639
RAC: 80
Germany
Message 1963330 - Posted: 4 Nov 2018, 22:24:19 UTC
Last modified: 4 Nov 2018, 22:25:52 UTC

Since the community is so lopsidedly NVidia it is hard to get advice from the AMD card users.


I`m not really sure how you are coming to this conclusion.
Since i made most of the testing related to the comandline values it`s not very hard to get information at all.
All app params are tested on AMD cards first.
There are sections for AMD and nvidia cards included in the read me files of both vendors.

For further optimization you can still ask me.


With each crime and every kindness we birth our future.
ID: 1963330 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 1963352 - Posted: 5 Nov 2018, 1:51:35 UTC - in response to Message 1963330.  
Last modified: 5 Nov 2018, 1:55:32 UTC

Since the community is so lopsidedly NVidia it is hard to get advice from the AMD card users.


I`m not really sure how you are coming to this conclusion.
Since I made most of the testing related to the command line values it`s not very hard to get information at all.
All app params are tested on AMD cards first.
There are sections for AMD and Nvidia cards included in the readme files of both vendors.

For further optimization, you can still ask me.


Whoops, I stand corrected. I should have said something like "I am not sure who to ask" and would have been more accurate and you still would have had to say "Here I am." Sorry, Mike.

Tom
ps. What would be a good starting place for a command line for a Vega 56?
A proud member of the OFA (Old Farts Association).
ID: 1963352 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34253
Credit: 79,922,639
RAC: 80
Germany
Message 1963448 - Posted: 5 Nov 2018, 15:14:04 UTC

A good start for vega would be.

-sbs 2048 -period_iterations_num 1 -spike_fft_thresh 4096 -high_perf -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64


With each crime and every kindness we birth our future.
ID: 1963448 · Report as offensive
Profile StFreddy
Avatar

Send message
Joined: 4 Feb 01
Posts: 35
Credit: 14,080,356
RAC: 26
Hungary
Message 1963461 - Posted: 5 Nov 2018, 16:51:08 UTC - in response to Message 1963448.  
Last modified: 5 Nov 2018, 17:46:31 UTC

Thanks Mike,

Applied these settings, now blc01 tasks are completed in about 210sec.
Without this tweak, completion times were around 300sec. Great improvement.

Edit: the completion time above is true only if 1 GPU task is running and there are no CPU tasks.
If the CPU starts to crunch, GPU completion time increases and overall GPU utilization decreases. It looks like the other threads are also using cpu core0 which is used by the GPU.

Regards
ID: 1963461 · Report as offensive
Previous · 1 . . . 5 · 6 · 7 · 8 · 9 · 10 · 11 . . . 20 · Next

Message boards : Number crunching : GPU FLOPS: Theory vs Reality


 
©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.