Any max-packet-size SMALL work units available for GPUs?

[Cassiopeia]

Good day to you all;

I have looked at the rest or the forum, but could not find an answer to my question.

I have a simple question:
Are there any max-packet-size SMALL work units available for GPUs?
Even if only from time to time?

I am asking this for this reason:
I try to make an old(er) computer work for Folding@Home.
It has a Core 2 Duo E8200 @ 2,66Ghz CPU and a Geforce GT710 with 2Gb of GDDR5.
I know this is older hardware.
For the CPU, I seem to be getting small work units which it can finish within the deadline of several days. So that's fine.
But for the GPU, I only seem to receive work units if I put the above parameter to at least "NORMAL" or "BIG".
But then I get work units for which the GPU will need 5 days or so. That is too long.

I know I could switch to the NACL client. I even have an Atom mini-pc which does this successfully.
But so I would like to get the GPU working for Folding@Home too.
So are there - even if just sometimes - small work units for GPUs?

If this is not possible according to you, then I will probably switch my GPU to GPUGRID:
I have a Geforce GT1030 doing small work units for that project, and it finishes them within 3-4 hours.
So the GT710 can probably do those too within 8-9 hours then. Which is still acceptable.

Thank you all and have a nice day;
Carl

[bruce]

The max-packet setting was originally designed to allow tiny packets which could be downloaded over dial-up modems in a reasonable amount of time and is not related to the time require to run the simulation. The feature isn't really supported any more.

The GT710 is a pretty slow GPU and finding WUs that can be completed by it is within the deadline may not be possible. The GT1030 has twice as many shaders, to it runs about twice as fast.

I'm guessing that both of your GPUs are installed in half-height cases which only have room for a single-slot GPU. Rarely, you may find a M/B with the slots reversed. allowing for a 2-slot GPU which will accommodate GPUs like the GTX 1050Ti LP.

If you don't want to upgrade your GT 710, GPUGRID is probably a good idea.

[JimboPalmer]

The GT 710 is several generations old. It uses a Kepler micro architecture design. Kepler was superseded by Maxwell, then Pascal, then Turing.

https://videocardz.net/nvidia-geforce-gt-710/

By comparison, the GTX 1030 is a Pascal design, so much newer (and more refined)

https://videocardz.net/nvidia-geforce-gt-1030-ddr4/

If I had to guess, the GTX 1030 is 2.5 times as fast as the GT 710. Due to the way F@H rewards speed, I suspect the GTX gets way more points per day. (5 times?)

[bruce]

I found an unused GT 710 in my junk department. I'll do some testing on it and if I confirm that it won't meet deadlines, I'll mark it as unsupported.

[Theodore]

@Bruce,
It would be interesting to know what PPD range this card has, just to know which similar cards will be unsupported in the near future.
I'm suspecting it probably performs close to a modern dual core 35Watt CoreI5 CPU ?

[Cassiopeia]

Hello again people;

Thanks all for your replies.

In the meantime, I am using the NACL client to help F@H in some way.
For those who are interested:
I downloaded some SMALL work units from GPUGRID for the GT710, and this it seems to be doing in between 12-18 hours, depending on the work unit. So that is doable... although not great. (My GT1030 does the same in about 5-8 hours.)
Mind you, my GT710 has GDDR5 memory. There are many models with slower DDR3 memory... don't know if that makes a big difference for CUDA applications just like with games... depends if CUDA is memory intensive, or not as much as games are.
And my GT710 is coupled with an old Core 2 Duo 8200 cpu... don't know if this slows down the video card. Not sure but I guess not: When I ran only a F@H GPU work unit, the CPU remained at +- 70% usage.
I THINK the GT710 can finish F@H work units within the deadline, but then you should let your computer run 24 hours a day. And I am not prepared to do this.

Yes indeed, it would be interesting to see what PPD the GT710 generates.

Yours sincerely;
Carl

[JimboPalmer]

For F@H, GPU RAM speed is not usually significant, I have no experience with GPUGRID.

Once you do not run 24/7 MANY GPUs will fail to complete in time. (and PPD diminish rapidly)

[bruce]

The GT710 I'm testing does run 24x7 and it does meet the deadline. I have not compared the PPD that it contributes with the loss in CPU performance when my Core2 Duo folds with 1 CPU rather than 2. (That would probably be different if I happened to have a CPU with hyperthreading.)

[Cassiopeia]

Thanks to you all for the replies people!
I love to read it.
And @Bruce: Thank you for testing your GT710!

But like I said, I am not prepared to let my pc run 24x7 to meet the deadline with my GT710.
But I want to continue to support F@H of course, just like the other projects I am helping: 'Mapping Cancer Markers' and 'Microbiome Immunity' (both from World Community Grid), Rosetta and SETI.
Instead of using 1 expensive pc, I decided to use several small mini-pcs (purchased on Aliexpress etc.) to work for these projects. Most are Apollo Lake N3450 or Gemini Lake N4100-based quad core Celerons.
I know that sounds like weak hardware, but you would be surprised about their speed:
My Core 2 Duo E8200 (4+ hours for 2 work units) cannot keep up with the Celeron 4-core N4100 ( about 7 hours for 4 work units) when it comes to parsing Mapping Cancer Markers work units.
And that Celerons' TDP is just 6 Watts versus the 65-watt TDP of the Core 2 Duo.

But so what I will do for F@H is this:
I will let the GT710 do small work units for GPUGRID. Their (small) work units are easier to handle with older hardware like the GT710.
As a compensation for F@H, I will let my Athlon GE200 do CPU units for F@H. That is a 3.2Ghz Ryzen-based cpu, so should be no problem.
And like I said, I have an Atom x5-8350 doing NACL client work units for F@H too. And that one, I let run 24x7 because it consumes almost nothing.
It usually does an nacl work unit every 2-3 hours, so that's just fine.
If, in the future I buy a better graphics card - a Turing-based GPU or so - I may consider letting it run F@H too.

Thank you everyone for your input, and have a nice day!
Carl

[bruce]

Sounds good

[JimboPalmer]

kinda on topic, Nvidia's numbering scheme

710

The ending 0 is a hint there is one GPU, if it ended in 5 there would be two on one card. (if the second digit is 0, the 5 just means half of 1 in the second slot, a 705 would be an even wimpier 710)

The second digit from the end indicates the power of the card in that generation. So the 1 indicates it has very little power. The 7 should indicate a generation, but Nvidia lies a lot here. The 710 was also a 630 and would become a 805. As Nvidia pretends it is a part of gen 7, they do admit it has little power.

Your 1030 is a Tenth Generation (for real!) low end part. But not bottom end. It is about 2.5 times the power of the 710.

[MeeLee]

Ramping up from a 710, you'd see a major speed boost with an RTX 2060 card; even a GTX 1660 ti is a great upgrade, and future proof for at least the next 4-5 years.

[bruce]

... and there are a lot of choices between a 710 and a 2060, depending on you budget.

The other numbering scheme that JimboPalmer left out is the suffix Ti which is kind of the opposite of the 5. It's an enhanced version of the GPU -- generally with more functional shaders. The GTX 750 is an intermediate speed gen7 GPU (which is still a viable contributor) and the GTX 750 Ti is an enhanced version of that card ... as with the GTX 660 Ti and the RTX 2080 Ti

[JimboPalmer]

The other numbering scheme that JimboPalmer left out is the suffix Ti which is kind of the opposite of the 5. It's an enhanced version of the GPU -- generally with more functional shaders.

I did leave that out! Which is odd as I use a GTX 1050 ti

[Theodore]

... and there are a lot of choices between a 710 and a 2060, depending on you budget.

True, but I'd agree that nothing folds as good for the money as a 2060.
The increase in initial purchase price over Pascal cards, is well worth it in the long run on the cost to run these cards (especially when ran 24/7).

I often wondered if the 'ti' cards were regular cards, with an additional GPU chip built in?
Seems like in the future we will see more cards with dual GPU cores built in the design.