This was interesting because I believe most volunteers are running 24/7 and their GPU's are lasting more than a year. With utilization approaching 100 percent.
But ours don't draw 700 watts per card. And run in a less noisy environment. So maybe data centers count as a hostile work environment?
Tom M
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!
This was interesting because I believe most volunteers are running 24/7 and their GPU's are lasting more than a year. With utilization approaching 100 percent.
But ours don't draw 700 watts per card. And run in a less noisy environment. So maybe data centers count as a hostile work environment?
Tom M
An article on Tech Spot a while ago said that most AI gpu's are not used 24/7 or even at 100% 100% of the time and they were suggesting some new software could be coming to move things around so that more gpu's are used fully instead of 100 gpu's running at half speed as it would be alot more efficient!!
This was interesting because I believe most volunteers are running 24/7 and their GPU's are lasting more than a year. With utilization approaching 100 percent.
But ours don't draw 700 watts per card. And run in a less noisy environment. So maybe data centers count as a hostile work environment?
Tom M
Probably, as it runs on 100% all the time...but that is why I use Tthrottle & a fix rate of 85°C top for GPU...so avg. for my Tesla is around 55,5% (around ~77% top), while my RTX 4000 runs on ~22% with no Tthrottling back.
As they do not use Ttthrottling (probably) & they are in hostile environment - with no space for cooling...I suspect that those are bad ways for card to work in!
The 5000 model MOBILE Nvidia gpu's have been listed in the UNofficial pci database!!
Recent entries in the unofficial PCI ID Repository have revealed the existence of at least five mobile variants of Nvidia's upcoming GeForce RTX 50 series graphics cards. The identified models include the RTX 5050, 5060, 5070, 5080, and 5090. The higher-end versions are reported to incorporate a mobile adaptation of the GB203 GPU. Nvidia has not yet provided an official response to these findings.
The PCI ID database entries were discovered by Harukaze5719 and suggest that Nvidia plans to announce the RTX 50 series in early 2025. Specifically, the RTX 5080 and 5090 models list the GPU type as GB203M. It remains unclear whether GB203M is a mobile variant of the GB203 GPU and whether all the information in the listings is accurate.
Nvidia has not formally announced the RTX 50 series yet, but the company is scheduled to present at the Consumer Electronics Show on January 7, 2025, where it may introduce the next generation of graphics cards. Early leaks suggest that the RTX 5080 and RTX 5090 desktop versions will feature 10,752 and 21,760 CUDA cores, respectively, along with significant memory capacities and bandwidth, all built on the Blackwell architecture.
I hope it will drive the rtx 40xx prices down out of the stratosphere. I don't think it will actually shrink any of those cards that take up 4 slots though.
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!
If you want to run a respectable e@h GPU RAC but save a little money on electricity you can power limit your EVGA rtx 3080 ti to as low as 200 watts and still get about 2.0M Rac. I have jumped mine back up to 250 watts which looks like about 2.5M Rac.
If you want to run at 90-95% speeds but save some money on electricity you can power limit it to 300 watts.
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!
I hope it will drive the rtx 40xx prices down out of the stratosphere. I don't think it will actually shrink any of those cards that take up 4 slots though.
Probably won't lower the price much (sadly). There is one factory 2 slot 4090 (GeForce RTX™ 4090 SUPRIM LIQUID X 24G from MSI) that is great- the others are all huge.
Getting back into it using my EVGA 3080TI
For those 2-3m ppd numbers. Are yall running multiple work units at a time?
Do you run cpu workloads along side that or does it work better to do GPU compute only.
Been crunching for about a week now but the numbers are not that impressive yet.
Getting back into it using my EVGA 3080TI
For those 2-3m ppd numbers. Are yall running multiple work units at a time?
Do you run cpu workloads along side that or does it work better to do GPU compute only.
Been crunching for about a week now but the numbers are not that impressive yet.
I am guessing you are running multiple tasks of the All-Sky gravity wave? Start at 1x for your baseline. Use no more than half your CPU cores under boinc manager. Then add a task.
If the increase in the run time doesn't double then you may have room for another task.
Don't exceed 90 percent of the GPU memory. You probably can't run more than 5x tasks for that application.
For the BRP 7/meerkat you are probably limited to 1x. The meerkat app is less sensitive to a more than 50 percent CPU load than the All-Sky GW app.
You should only compare your results to other Windows systems. Linux has a Nvidia MPS which is not available under Windows.
Hope This Helps.
Tom M
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!
https://www.extremetech.com/c
)
https://www.extremetech.com/computing/data-center-ai-gpus-may-have-extremely-short-lifespans
This was interesting because I believe most volunteers are running 24/7 and their GPU's are lasting more than a year. With utilization approaching 100 percent.
But ours don't draw 700 watts per card. And run in a less noisy environment. So maybe data centers count as a hostile work environment?
Tom M
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!
Tom M
)
An article on Tech Spot a while ago said that most AI gpu's are not used 24/7 or even at 100% 100% of the time and they were suggesting some new software could be coming to move things around so that more gpu's are used fully instead of 100 gpu's running at half speed as it would be alot more efficient!!
Tom M
)
Probably, as it runs on 100% all the time...but that is why I use Tthrottle & a fix rate of 85°C top for GPU...so avg. for my Tesla is around 55,5% (around ~77% top), while my RTX 4000 runs on ~22% with no Tthrottling back.
As they do not use Ttthrottling (probably) & they are in hostile environment - with no space for cooling...I suspect that those are bad ways for card to work in!
non-profit org. Play4Life in Zagreb, Croatia, EU
The 5000 model Nvidia gpu's
)
The 5000 model MOBILE Nvidia gpu's have been listed in the UNofficial pci database!!
Recent entries in the unofficial PCI ID Repository have revealed the existence of at least five mobile variants of Nvidia's upcoming GeForce RTX 50 series graphics cards. The identified models include the RTX 5050, 5060, 5070, 5080, and 5090. The higher-end versions are reported to incorporate a mobile adaptation of the GB203 GPU. Nvidia has not yet provided an official response to these findings.
The PCI ID database entries were discovered by Harukaze5719 and suggest that Nvidia plans to announce the RTX 50 series in early 2025. Specifically, the RTX 5080 and 5090 models list the GPU type as GB203M. It remains unclear whether GB203M is a mobile variant of the GB203 GPU and whether all the information in the listings is accurate.
Nvidia has not formally announced the RTX 50 series yet, but the company is scheduled to present at the Consumer Electronics Show on January 7, 2025, where it may introduce the next generation of graphics cards. Early leaks suggest that the RTX 5080 and RTX 5090 desktop versions will feature 10,752 and 21,760 CUDA cores, respectively, along with significant memory capacities and bandwidth, all built on the Blackwell architecture.
https://www.guru3d.com/story/nvidia-rtx-50-mobile-graphics-cards-identified-in-unofficial-pci-database/
Let the resales begin!!! WOO HOO!!!
I hope it will drive the rtx
)
I hope it will drive the rtx 40xx prices down out of the stratosphere. I don't think it will actually shrink any of those cards that take up 4 slots though.
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!
My results are in. If you
)
My results are in.
If you want to run a respectable e@h GPU RAC but save a little money on electricity you can power limit your EVGA rtx 3080 ti to as low as 200 watts and still get about 2.0M Rac. I have jumped mine back up to 250 watts which looks like about 2.5M Rac.
If you want to run at 90-95% speeds but save some money on electricity you can power limit it to 300 watts.
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!
I always ran mine at 300W for
)
I always ran mine at 300W for about 3.0M ppd per 3080Ti.
_________________________________________________________________________
Tom M wrote: I hope it will
)
Probably won't lower the price much (sadly). There is one factory 2 slot 4090 (GeForce RTX™ 4090 SUPRIM LIQUID X 24G from MSI) that is great- the others are all huge.
Have been away from the
)
Have been away from the project for a time.
Getting back into it using my EVGA 3080TI
For those 2-3m ppd numbers. Are yall running multiple work units at a time?
Do you run cpu workloads along side that or does it work better to do GPU compute only.
Been crunching for about a week now but the numbers are not that impressive yet.
yummycheese wrote: Have been
)
I am guessing you are running multiple tasks of the All-Sky gravity wave? Start at 1x for your baseline. Use no more than half your CPU cores under boinc manager. Then add a task.
If the increase in the run time doesn't double then you may have room for another task.
Don't exceed 90 percent of the GPU memory. You probably can't run more than 5x tasks for that application.
For the BRP 7/meerkat you are probably limited to 1x. The meerkat app is less sensitive to a more than 50 percent CPU load than the All-Sky GW app.
You should only compare your results to other Windows systems. Linux has a Nvidia MPS which is not available under Windows.
Hope This Helps.
Tom M
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!