New RTX3xxx cards

[scott@bjorn3d]

EVGA boards with the 6 cap pictures where taken before they found issue, all shipping cards are right.

[ajm]

"Grapevine says that there are reports of instabilities on ASUS TUF and Strix cards as well. So 6x MLCC does not make you immune."

reddit - Ampere POSCAP/MLCC Counts : hardware

Well, all the more reason for FAH to test these cards for their specific use case. Because the problem appears without manual overclocking, when just using the card as stock. It might be a catastrophe when those cards will start to fold...

[ap1978]

"Grapevine says that there are reports of instabilities on ASUS TUF and Strix cards as well. So 6x MLCC does not make you immune."

reddit - Ampere POSCAP/MLCC Counts : hardware

Well, all the more reason for FAH to test these cards for their specific use case. Because the problem appears without manual overclocking, when just using the card as stock. It might be a catastrophe when those cards will start to fold...

Sure

[MeeLee]

For most people folding on the 3080 gpus, it wouldn't matter much.
Since that gpu has like a ~320W tdp limit, I would presume that most people would just lower the power slider, and run it at a much lower power consumption, and a slightly lower boost speed.

[ajm]

Slightly? Someone on reddit claims that he can avoid the crashes at 1850 MHz and 0.9 V. For FAH, I guess he'd have to go even lower than that...
I canceled my order.

[foldy]

They for sure will fix the HW in 2nd edition or provide BIOS updates. One reason to wait before buy is the AMD release soon which maybe lowers Nvidia prices more. And the Nvidia 2nd edition HW may have more VRAM 16GB+ if AMD also provides high VRAM on its upcoming GPUs. Then the low VRAM nvidia 1st edition GPUs may get even cheaper and still are good enough for FAH.

[HaloJones]

A BIOS update can only reduce the GPU Boost settings which can just as easily be achieved by underclocking with Afterburner

[MeeLee]

A BIOS update can only reduce the GPU Boost settings which can just as easily be achieved by underclocking with Afterburner

We're not sure where the issue is exactly.
They may lower power on certain shaders which may not affect FAH at all.

Vram size is no issue for FAH. It uses <2GB for large atom WUs, and usually closer to 500MB of VRAM.

I suspect that once they move over to much larger atoms, they might need to increase the window resolution in which the atom is simulated.
Then we might need a bit more VRAM.

[MeeLee]

Slightly? Someone on reddit claims that he can avoid the crashes at 1850 MHz and 0.9 V. For FAH, I guess he'd have to go even lower than that...
I canceled my order.

It's very common for Nvidia GPUs to lower the boost clock depending on the load.
With FAH, or many other OpenCL computing tasks, making more than 90% use of the GPU, the GPU frequencies tend to dip, even at stock power settings.
However, the strange thing of it is, that a -200Mhz frequency drop of a ~2Ghz core clock, doesn't directly translate in a 10% PPD drop.
It'll more than likely result in <5% (<2% in some cases).
Because the memory and other parts of the GPU are still running at the same speed (including the PCIE data to the GPU).

[ap1978]

Seems like the latest drivers fixes this for people. Without having to use Afterburner or update the vBIOS.

[bruce]

Vram size is no issue for FAH. It uses <2GB for large atom WUs, and usually closer to 500MB of VRAM.

I suspect that once they move over to much larger atoms, they might need to increase the window resolution in which the atom is simulated.

"Simulated'? I think you're talking about the viewer, so maybe you mean "viewed"

The simulation runs in floating point numbers and has no window.

[MeeLee]

Vram size is no issue for FAH. It uses <2GB for large atom WUs, and usually closer to 500MB of VRAM.

I suspect that once they move over to much larger atoms, they might need to increase the window resolution in which the atom is simulated.

"Simulated'? I think you're talking about the viewer, so maybe you mean "viewed"

The simulation runs in floating point numbers and has no window.

Makes sense!
That's why I said 'I suspect'.
But it's good that that comment is corrected.
Essentially the VRAM is dependent on atom count.

In Boinc they do have 'resolutions' they work with on some projects.

[bruce]

Yes, VRAM depends on atom-count. If the protein has N atoms, then you enough VRAM to store the coordinates and other characteristics of each of those atoms. If the entire protein fits in a box that can be X by Y by Z and the viewer should display everything that is in that box, but the viewer is optional since the computed location of each atom is the same whether or not the viewer is used.

[Foxter]

Some RTX 3090 PPD/Temperatures/Power consumption from my team leader:

https://www.mygarage.ro/componente/2600 ... ost4358177

I would add the pictures here but I cannot find the add attachment button, maybe this is restricted because I am a new member?

As for the above post text:

3090 Msi Ventus. (no need for translation here)
In [email protected]% power, tmax 75 ca si setari. (When folding at home. Settings: Power Limit set to 70%, Maximum temperature allowed set to 75 degrees Celcius)

[ap1978]

Some RTX 3090 PPD/Temperatures/Power consumption from my team leader:

https://www.mygarage.ro/componente/2600 ... ost4358177

I would add the pictures here but I cannot find the add attachment button, maybe this is restricted because I am a new member?

As for the above post text:

3090 Msi Ventus. (no need for translation here)
In [email protected]% power, tmax 75 ca si setari. (When folding at home. Settings: Power Limit set to 70%, Maximum temperature allowed set to 75 degrees Celcius)

Thanks for sharing!