All things Amd GPU

Skip Da Shu
Skip Da Shu
Joined: 18 Jan 05
Posts: 151
Credit: 1036926627
RAC: 845100

Installing ( 'dual boot'

Installing ( 'dual boot' optional) and using Linux Mint V22 cinnamon is quite easy in most situations although I'll admit to not using dual boot in a couple years.  Especially with Nvidia cards and using the Costamagnaianfranco/Boinc alternate source (PPA) for BOINC install.

Skip

mikey
mikey
Joined: 22 Jan 05
Posts: 12661
Credit: 1839059161
RAC: 4353

Skip Da Shu

Skip Da Shu wrote:

Installing ( 'dual boot' optional) and using Linux Mint V22 cinnamon is quite easy in most situations although I'll admit to not using dual boot in a couple years.  Especially with Nvidia cards and using the Costamagnaianfranco/Boinc alternate source (PPA) for BOINC install.

Skip 

I think it's Costamagnagianfranco/boinc

I added a "g" for gian and made the large 'B' in boinc a small letter, I don't know if the capital or small letter makes a difference but I do know that you DO need the 'g' in 'gian' to make it work

Skip Da Shu
Skip Da Shu
Joined: 18 Jan 05
Posts: 151
Credit: 1036926627
RAC: 845100

no doubt

no doubt on the "g"... much doubt on the case

Filipe
Filipe
Joined: 10 Mar 05
Posts: 186
Credit: 402626726
RAC: 364301

tictoc wrote: I have a

tictoc wrote:

I have a 7900xtx on my bench, so I can run some tasks later as a sample.

The XTX in addition to the increased clock speed over the XT, also has more shaders, TMUs, L2 and a wider memory bus.  It won't be a direct comparison, but my guess is the XTX is probably 15-20% faster than the XT on O3AS.

 

Your 7900xtx is running O3AS tasks in +/- 13min.

Are you running only 1 task at a time? or more concurrently?

tictoc
tictoc
Joined: 1 Jan 13
Posts: 44
Credit: 7142085465
RAC: 8643519

Filipe wrote:Your 7900xtx

Filipe wrote:

Your 7900xtx is running O3AS tasks in +/- 13min.

Are you running only 1 task at a time? or more concurrently?

7900xtx is running 3x tasks. Which would put it at roughly 3.4M ppd running 24/7. 

Anything over 3x tasks would crash and reset the driver.  I haven't had time to look into what was actually going on, so I just settled on running 3x tasks.  Overall throughput might be bit  higher at 4x or 5x, but I don't think it would have been all that much better than 3x. The GPU is running stock core/boost and memory clocks, with the power cap set at 460W.  GPU runs Einstein O3AS tasks at 3025 MHz core, 2500 MHz memory, and average package power is ~315W, with peak power usage of 355W.

System Specs - CPU: AMD 7960X @ 5 GHz | OS: Arch Linux | Kernel 6.12-rc2 | ROCm 6.2.2

Unrelated, but somewhat interesting, is that I saw the same driver crashing issues when I was running BRP7 tasks on a 6900xt.  It will run all day at 2x tasks, but falls apart at 3x.  Meanwhile a Radeon VII will happily crunch 3x BRP7 and 6x O3AS.  No idea if this is just a Linux/ROCm OpenCL issue, or if the same thing happens on Windows.  

Filipe
Filipe
Joined: 10 Mar 05
Posts: 186
Credit: 402626726
RAC: 364301

Which one is faster at

Which one is faster at Einstein@home? (Windows only!)

The 7800xt or the 4070Super? Tey are about the same price here

 

 

B.I.G
B.I.G
Joined: 26 Oct 07
Posts: 117
Credit: 1165534421
RAC: 1003321

tictoc wrote:7900xtx is

tictoc wrote:

7900xtx is running 3x tasks. Which would put it at roughly 3.4M ppd running 24/7. 

and average package power is ~315W, with peak power usage of 355W.

That is impressive, granted my system also downloads BRP7 tasks which are not as credit efficient as GW tasks but with the W7600 I get 900k RAC running GW tasks at 3x and the card is using 115 watts average, I thought the high performance cards are much less energy efficient than their low performance counterparts.

tictoc
tictoc
Joined: 1 Jan 13
Posts: 44
Credit: 7142085465
RAC: 8643519

The 7900xtx could definitely

The 7900xtx could definitely be tuned to be more efficient.  I just let it run for the last week to get a baseline of the stock performance. 

When left alone RDNA2 GPUs will boost as high as they can, using the full available power budget.  An undervolt + setting a lower max power cap would undoubtedly increase the efficency without losing much performance. 

pututu
pututu
Joined: 6 Apr 17
Posts: 63
Credit: 653417392
RAC: 15

tictoc wrote: 7900xtx is

tictoc wrote:

7900xtx is running 3x tasks. Which would put it at roughly 3.4M ppd running 24/7. 

Anything over 3x tasks would crash and reset the driver.  I haven't had time to look into what was actually going on, so I just settled on running 3x tasks.  Overall throughput might be bit  higher at 4x or 5x, but I don't think it would have been all that much better than 3x. The GPU is running stock core/boost and memory clocks, with the power cap set at 460W.  GPU runs Einstein O3AS tasks at 3025 MHz core, 2500 MHz memory, and average package power is ~315W, with peak power usage of 355W.

System Specs - CPU: AMD 7960X @ 5 GHz | OS: Arch Linux | Kernel 6.12-rc2 | ROCm 6.2.2

Unrelated, but somewhat interesting, is that I saw the same driver crashing issues when I was running BRP7 tasks on a 6900xt.  It will run all day at 2x tasks, but falls apart at 3x.  Meanwhile a Radeon VII will happily crunch 3x BRP7 and 6x O3AS.  No idea if this is just a Linux/ROCm OpenCL issue, or if the same thing happens on Windows.  

Looks great. I will also suggest testing with higher memory clock (2700 MHz?) and lower the core clock a bit for the same power envelope. IIRC in my VII testing, I set the core clock to around 1450MHz and memory clock around 1100 MHz, running 5 to 6 tasks per gpu.

 

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6412
Credit: 9546253389
RAC: 14968826

I recently re-started my

I recently re-started my Windows/Ryzen 5700G system.  And started experimenting with the "dinky" iGpu.

I am running the All-Sky GW application on it.  The GPU load was running maybe 70% with 1 task.  I have bumped it to two tasks which pretty much loads the GPU up.  It looks like I will get a bump in the RAC for e@h with no issues.

However:

I am pretty sure that the CPU crunching is running slower now. 

I believe I had the same results with Ryzen 2200G cpu's and certainly with the more elderly Intel iGpu's I have played with.

It probably has to do with using non-gpu ram on the gpus.  (my understanding is that iGpu's share the regular MB ram rather than their own separate ram like a discreet GPU does.)

Just mumbling along.

 

 

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)  I want some more patience. RIGHT NOW!

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.