MilkyWay@H - Benchmark thread Winter 2016 on (updated 1-2021) - GPU & CPU times wanted for new WUs

[IEC]

Mmmm, would love to have the card for MW or Einstein

IEC
Did you ever get to water cooling your VII?

Nope, never did. Just undervolt and underclock it a bit and it runs cool and quiet on the stock air cooler.

 

[Assimilator1]

Taking a visual average from 2 pages of 227.5x credit WUs my RX 580 (underclocked to 1266 MHz) is averaging 96s, running single WUs at a time. Will be interesting to see if that's consistent.
[edit]And my HD 7870 XT DS seems to be doing 227.5x WUs in 66-69s, so an average of about 68s.

What are you guys seeing?

 

[Assimilator1]

And with running 2 tasks at once, I'm getting about 165-170s with my RX 580 and 136-140s on my HD 7870XT.

Interesting, their seems to be no real improvement on my 7870XT, anyone else find that with a Tahiti GPU?

 

[Fardringle]

Now that I (sort of) have the overheating issues resolved, I grabbed the last 200 or so results from my R9 280X. The GPU clock speed is set at 900MHz (stock is 1000MHz) and the RAM is running at 1500MHz. I have the 20.8.3 Adrenalin 2020 drivers installed.

The CPU is a Ryzen 9 3900X, limited to 105 watts PPU so it usually sits around 3.9MHz.
32GB of DDR4 3600MHz CL16 RAM.

OS is Windows 10 Pro.

I'm running two tasks at a time with .05 CPU assigned per task.

The 227.5 point tasks are averaging 77.2 seconds each.
The 244 point tasks are averaging 94.4 seconds each.

I can switch back to running 1 at a time for a while if you want those numbers as well.

 

[Assimilator1]

Wow! Your 280X is doing 2 WUs in nearly the time it takes my HD7870 XT to do one!
Weird, shouldn't be that bigger gap (HD 7870XT is nearly the same performance as an HD 7950, and my DS version is practically the same)...... wait, I remember seeing in the drivers it was clocked at 850 MHz I think, it's supposed to be 925 MHz!
[edit] Numbers by slider show 925 MHz, but actual speed is 850 MHz (confirmed by GPU-Z) and even after moving the slider and hitting apply, I don't know what's going on here. Funny the numbers match the HD 7950 boost......

 

[Fardringle]

Just for fun, and to show how pathetic they are before this machine gets retired, these are the stats for my old Quadro 4000 cards in my dual Xeon 5680 computer.

Dual Xeon 5680 processors.
24 GB DDR3 1600MHz RAM

Running 1 task at a time per GPU (2 total).

The 227.5 point tasks average 481.1 seconds.
The 244 point tasks average 537.3 seconds.

That's pretty hideous, especially considering that these graphics cards DO support double precision...

I had been running them on MilkyWay because they SHOULD do better than that with the double precision, and because no other GPU project will even talk to them. But the whole machine is going away soon. Even though the CPUs aren't completely horrible, the whole system just uses way too much electricity compared to the results that it produces..

Edit for 2X run times:
Running a total of 4 tasks at a time (2 per GPU) with .25 CPU assigned.

The 227.5 point tasks average 799.2 seconds. That's 81.5 seconds faster per task than running singles.
The 244 point tasks average 859.4 seconds. That's 107.6 seconds faster per task than running singles.

It's still terrible PPD, but a pretty significant increase when running two tasks at a time!

 

[Icecold]

Edit - wrong thread.

 

[Assimilator1]

Nvidia has never done very well with MW, although you'd probably find they'd do better if you ran multiple WUs on them, your experiment isn't over (unless the machine has already gone).

 

[Fardringle]

I tried running two tasks each on the Quadro cards, but for some reason BOINC on this computer refuses to recognize that any config files exist. When I tell it to manually update the config files, it says it can't find a cc_config.xml file, even though it exists, and I used it at some point in the past to tell BOINC to use all GPUs.

And it doesn't even seem to be trying to look for app_config.xml files in the project folders. The file is in the right place with the right formatting for MilkyWay but BOINC doesn't see it. I tried putting the wrong app name in the file to make it give an error message but it doesn't error out either.

I just completely uninstalled and reinstalled BOINC to see if that would fix the problem, but it didn't. It's really not a big deal, but it is definitely annoying when things don't work right!

 

[emoga]

@Fardringle Are you on windows? Are you certain that this specific computer isn't hiding known file extensions?

Spoiler: How to Hide or Show File Name Extensions in Windows 10

 

[Fardringle]

Wow.

I always turn that "feature" off because I hate it. But apparently a Windows update or something turned it back on so I was seeing cc_config.xml and app_config.xml in the folders, but they were really cc_config.xml.txt and app_config.xml.txt...

Thanks for the suggestion! I never would have thought of checking that because I KNOW it was turned off. (Yay Windows).

It's running four GPU tasks now (2 on each Quadro 2000) with .25 CPU assigned. I'll wait to see how they run with that before I lower the CPU allocation any more.

edit: I edited my previous benchmark post with the 2X run times for the Quadro GPUs.

 

[Assimilator1]

Did you manage to grab some single WU times on your 280X?

That's weird, my HD 7870XT is now doing 2 WUs at a time in about 90s (~36% faster), even getting it to run at 925 MHz again (from 850) ~9% faster, doesn't account for it.
Faster WUs maybe? Let's see what my RX 580's doing now, about 164s, so about the same then, it's not the WUs! I don't get it....

 

[Fardringle]

I did grab numbers for single tasks, but apparently didn't actually update my answer. I don't recall exactly and prefer not to switch back to singles again, but I think it was around 46-48 seconds.

 

[Assimilator1]

Updated the op, will be adding the few times we have here for the 227.5x WUs soon, feel free to add more!

@Fardringle Your old dual xeon rig, how many WUs was those averages calculated over? (22/10)

 

[Fardringle]

Updated the op, will be adding the few times we have here for the 227.5x WUs soon, feel free to add more!

@Fardringle Your old dual xeon rig, how WUs was those averages calculated over? (22/10)

I don't recall exactly, but I'm pretty sure it was at least 100 tasks on each of the tests.

 

[Assimilator1]

Ah, way more than 6, that's cool

 

[Assimilator1]

I've just seen that I forgot to ask you for single WU times for your 280X, is it too late now?

 

[Fardringle]

I did grab numbers for single tasks, but apparently didn't actually update my answer. I don't recall exactly and prefer not to switch back to singles again, but I think it was around 46-48 seconds.

Not exact, but is that good enough?

 

[pututu]

One of our members have a Radeon VII Pro. Running time is 30 secs for 4 tasks or about 7.5 secs per task. His rig is here.

Also posted in our [H] forum.

Cards are available if you have $1900 (ea)

 

[Assimilator1]

Sweet!

Fardringle
Yea that'll do, I'll count it as 47s
Doh! re my memory! lol

 

[Assimilator1]

I just set my Firepro S9150 back to run one task at a time instead of 4 for the purpose of this benchmark thread and ran some WU's:

Run time (sec)​	Credit​	Application​
33.16​	227.53​	Milkyway@home Separation v1.46 (opencl_ati_101) windows_x86_64​
31.13​	227.12​	Milkyway@home Separation v1.46 (opencl_ati_101) windows_x86_64​
31.14​	227.12​	Milkyway@home Separation v1.46 (opencl_ati_101) windows_x86_64​
30.17​	227.51​	Milkyway@home Separation v1.46 (opencl_ati_101) windows_x86_64​
31.26​	227.51​	Milkyway@home Separation v1.46 (opencl_ati_101) windows_x86_64​
31.14​	227.12​	Milkyway@home Separation v1.46 (opencl_ati_101) windows_x86_64​
31.19​	227.51​	Milkyway@home Separation v1.46 (opencl_ati_101) windows_x86_64​
34.16​	227.53​	Milkyway@home Separation v1.46 (opencl_ati_101) windows_x86_64​
27.14​	227.51​	Milkyway@home Separation v1.46 (opencl_ati_101) windows_x86_64​
31.14​	227.12​	Milkyway@home Separation v1.46 (opencl_ati_101) windows_x86_64​
25.32​	227.12​	Milkyway@home Separation v1.46 (opencl_ati_101) windows_x86_64​

The following are while running 4 WU's at a time:

Run time (sec)​	Credit​	Application​
96.54​	227.52​	Milkyway@home Separation v1.46 (opencl_ati_101) windows_x86_64​
83.47​	227.51​	Milkyway@home Separation v1.46 (opencl_ati_101) windows_x86_64​
87.9​	227.51​	Milkyway@home Separation v1.46 (opencl_ati_101) windows_x86_64​
95.56​	227.53​	Milkyway@home Separation v1.46 (opencl_ati_101) windows_x86_64​

If my math is correct - running 1 WU at a time comes out to around 641,000 PPD. 4 WU's at a time comes out to around 865,000 PPD.

What CPU was that with?
Only just getting around to adding times, and realised you hadn't said (recently anyway).

 

[Icecold]

What CPU was that with?
Only just getting around to adding times, and realised you hadn't said (recently anyway).

It would have been an i3-4160 with all cores/threads open feeding the GPU's with no CPU tasks. As far as I've observed it is not CPU limited/bottlenecked in this configuration.

 

[Assimilator1]

Fast times btw! Although not as fast as Holdolin's VII Pro!

One of our members have a Radeon VII Pro. Running time is 30 secs for 4 tasks or about 7.5 secs per task. His rig is here.

Also posted in our [H] forum.

Cards are available if you have $1900 (ea)

Thanks for that , I've added his 4WU time to the table, does he fancy running at least 5x 227.5 credit single WUs? (so I can add a time to the single WU table).

 

[Icecold]

Fast times btw! Although not as fast as Holdolin's VII Pro!

That is true. I could buy 18 S9150's for what one VII Pro costs, though

 

[Assimilator1]

Lol, wow! What ya waiting for??