Folding Community Board

[Gorgon]

Looks like my timing was good to get those B-Stock EVGA 2080 XC Ultra's. There's hardly any stock showing available anymore on EVGA's website and the B-Stock is likewise showing fewer parts available and at higher prices.

[rkv_2401]

I need a way to reduce the power consumption drastically on my laptop's 1660Ti mobile. I have tried reducing the core clock by 400Mhz and memory by 502 (the furthest they'll go) and manually locked the voltage to 700mV (the lowest that will go) but it still hits 80C occasionally. Any suggestions? I'm already turning my remaining fan up to 100% and using a laptop cooler.

[Gorgon]

3 hours ago, rkv_2401 said:

I need a way to reduce the power consumption drastically on my laptop's 1660Ti mobile. I have tried reducing the core clock by 400Mhz and memory by 502 (the furthest they'll go) and manually locked the voltage to 700mV (the lowest that will go) but it still hits 80C occasionally. Any suggestions? I'm already turning my remaining fan up to 100% and using a laptop cooler.

nvidia-smi -i 0 -lgc 0,1440

works on my 1660ti and keeps it around 60C in a mid-tower with hardly any airflow.

 

Another thing you can try when not using it is to open it 45-60 degrees then rest it on the edges to form a upside-down "V" which should allow easier ingress of cooler air.

 

Edit - @rkv_2401

Here's the Efficiency vs. Graphics Clock for a 1660ti running p18202:

The peak efficiency was observed about 1350MHz but YMMV with a mobile variant. You should be able to safely go down to between 700-800MHz I suspect. Just keep lowing the Max Clock until your happy with the temperature and yield.

[rkv_2401]

I just switched my preference to Cancer WUs. Temps haven't gone above 70c all day. I get like 450k PPD instead of the 900k I got on 18xxx, but that's alright with me.

 

EDIT : Thanks for the update, Gorgon, the lowest mine will go is 930Mhz on Core before automatically locking the DRAM to 510Mhz. But yeah, now that I have this extra headroom to play with I think I'll probably increase it to 1150Mhz on core, that seems to be near the peak of the efficiency curve without drawing too much watts and overloading my thermals.

[Justaphf]

1 hour ago, Schnoz said:

Does anyone happen to know which type of WUs award the most PPD? I'm only aware that cancer WUs are not good with point yields, but I'm not sure of anything else.

You should check out @LAR_Systems's folding in the dark page.  You can see the best and worst 5 WU per GPU/CPU.  I don't think it has an overall best/worst ranking independent of platform/hardware as it is based on real systems reporting live results as they are folding.  Plenty of other cool stuff on there like efficiency, performance on Windows vs Linux, points per kWh, etc.  I'm a fan.

 

Example link to a 3080ti 5 best and worst recent WUs: https://folding.lar.systems/gpu_ppd/brands/nvidia/folding_profile/ga102_geforce_rtx_3080_ti#projects

I'm only seeing the 5 best and worst currently, it used to show the table of all reported WUs for a card but it's either gone or I'm forgetting how to get to it.

[Gorgon]

1 hour ago, Schnoz said:

Does anyone happen to know which type of WUs award the most PPD? I'm only aware that cancer WUs are not good with point yields, but I'm not sure of anything else.

This is very much a moving target, In generall it's best to stick with one Cause or set no preference.

 

Last time I analyzed my HfM stats I saw:

Average		Cause		
2,461,935	Cancer		-5%
2,529,325	Unspecified	-2%
2,873,266	Alzheimers	11%
2,534,796	Parkinsons	-2%

[justpoet]

As we are now nearing that time, being in October....I'll do the obligatory ask.

@GOTSpectrum, when should we expect to see details on the dates and details of your binge drinking group encouragement the annual winter folding month?

[CT854]

21 minutes ago, justpoet said:

when should we expect to see details on the dates and details of your binge drinking group encouragement the annual winter folding month?

Heating bill replacement event

[GOTSpectrum]

1 hour ago, justpoet said:

As we are now nearing that time, being in October....I'll do the obligatory ask.

@GOTSpectrum, when should we expect to see details on the dates and details of your binge drinking group encouragement the annual winter folding month?

I'd suggest following me to keep up to date 

 

 

Hopefully next week sometime if things work out 

 

[GOTSpectrum]

 

 

[justpoet]

4 hours ago, GOTSpectrum said:

I'd suggest following me to keep up to date 

...

Hopefully next week sometime if things work out 

Oof.  Sorry mate, had no idea.  I generally only look at this folding thread and occasionally the tech news section.  Sorry for the poorly timed funny.

[GOTSpectrum]

1 minute ago, justpoet said:

Oof.  Sorry mate, had no idea.  I generally only look at this folding thread and occasionally the tech news section.  Sorry for the poorly timed funny.

It's okay, I was planning to have a whole post introducing the event yesterday, talking about the history of folding events on the forum and such, but all you get is a logo and a date. Better than nothing I suppose 

[Shlouski]

Anyone heard any news on whether or not Fah will support arc gpu's? 

 

I was considering buying a couple and testing them out.

 

I think arc gpu's really need resizable bar to work properly, I have no idea if this is possible over bifurcated x4 risers, guess I will test it out. 

[CT854]

35 minutes ago, Shlouski said:

Anyone heard any news on whether or not Fah will support arc gpu's? 

 

I was considering buying a couple and testing them out.

 

I think arc gpu's really need resizable bar to work properly, I have no idea if this is possible over bifurcated x4 risers, guess I will test it out. 

I've been wondering about this too. I've been considering getting an A380 for AV1 encoding, but I'd also like to know if it's supported by F@H because I'd like that extra value too...

 

I might be wrong but isn't resizable BAR only important for ARC for gaming?

[Shlouski]

43 minutes ago, Shlouski said:

I think arc gpu's really need resizable bar to work properly, I have no idea if this is possible over bifurcated x4 risers, guess I will test it out. 

Resizable bar is working on 3 out of 5 cards, 2 out of the 4 cards on the 4x4x bifurcated pcie16_1 slot and the card on its own in the pci16_2 slot. 

 

I wonder why 2 cards work and the other 2 don't. 

[Gorgon]

48 minutes ago, Shlouski said:

Anyone heard any news on whether or not Fah will support arc gpu's? 

 

I was considering buying a couple and testing them out.

 

I think arc gpu's really need resizable bar to work properly, I have no idea if this is possible over bifurcated x4 risers, guess I will test it out. 

It Should work as long as they support OpenCL 1.2 and have FP64 support and the drivers to expose this. Looks like the hardware is capable.

 

They're not on the F@H Whitelist yet.

 

[miker07]

13 hours ago, Gorgon said:

It Should work as long as they support OpenCL 1.2 and have FP64 support and the drivers to expose this. Looks like the hardware is capable.

 

They're not on the F@H Whitelist yet.

 

Looks like FP64 support could be an issue from what I've seen. Tom' and extremetech's articles on the A380. https://www.tomshardware.com/news/intel-arc-will-not-support-fp64-hardware  https://www.extremetech.com/gaming/331935-intel-arc-a380-benchmarked-full-alchemist-gpu-specs-reportedly-revealed

[GOTSpectrum]

 

Event thread is live! 

[rkv_2401]

On 10/4/2022 at 12:50 AM, miker07 said:

Looks like FP64 support could be an issue from what I've seen. Tom' and extremetech's articles on the A380. https://www.tomshardware.com/news/intel-arc-will-not-support-fp64-hardware  https://www.extremetech.com/gaming/331935-intel-arc-a380-benchmarked-full-alchemist-gpu-specs-reportedly-revealed

Good find! That seems to be only for the Arc A380 though. Techpowerup's page (linked by Gorgon in his post) says that the A780 has 4.5TFLOPs of FP64 performance, almost 10x what even the 3090 offers.

[miker07]

43 minutes ago, rkv_2401 said:

Good find! That seems to be only for the Arc A380 though. Techpowerup's page (linked by Gorgon in his post) says that the A780 has 4.5TFLOPs of FP64 performance, almost 10x what even the 3090 offers.

Techpowerup also list the A380 as being 1:4 FP64, along with rest of Arc lineup, which isn't correct. It is just placeholder assumptions because they didn't know. Those articles also indicates all gaming consumer gpu's would be this way, so you have to get their professional/server cards much like Nvidia. Quote from Tom's article; "Keep in mind that this configuration only applies to Intel's gaming-centric Arc Alchemist GPUs." AMD cripples FP64 a lot less than Nvidia, so AMD gpu's are usually better if FP64 performance is important. With folding@home taking advantage of CUDA, Nvidia still has a folding advantage, but they at least have some support for FP64. Arc not having any FP64 hardware support could be an issue if required by Folding@home. Milkyway@home is 1 of the few Boinc projects that takes advantage of FP64. Radeon VII's and some other older AMD gpu's are beast at FP64 for consumer cards. Unless Intel worked to optimize folding@home specifically, which I doubt given all their software issues and delays, they will probably be behind Nvidia and AMD on performance per watt, and T-flop. Nvidia's performance shows how important optimizations are, as they consistently outperform AMD gpu's with comparable compute T-flops when it comes to folding.

[jctappel67]

2 Billy! Closing in on Diamond!

 

[rkv_2401]

3 hours ago, miker07 said:

Techpowerup also list the A380 as being 1:4 FP64, along with rest of Arc lineup, which isn't correct. It is just placeholder assumptions because they didn't know.

Ah, I did check that before posting but I missed the comma in the 1050 (I thought it was one singular gigaFLOP not 1050 lol). Yup they seem to be placeholder assumptions. I figured something was amiss because why would the A780 have 10x the FP64 performance of a 3090 if it's sibling card, the A380 was emulating FP64 support.

 

Agreed about the FP64 problems, if in the future they use it more liberally the emulated support would be a problem.  (https://foldingforum.org/viewtopic.php?f=38&t=36371 ) - As of now it should hopefully be fine, Jimbo says "For Folding@Home we can ignore Pixel Rate, Texture Rate, and FP16, we don't do them. The vast majority is FP32 with a sprinkling of FP64." but that was in 2020. Intel will, nevertheless, likely be behind nVidia and AMD on PPD/watt and maybe behind AMD on TFLOPS/price despite the A780 having a good FP32 TFLOPS compute number for it's price.

[NetoriusNick]

defiantly interested in the arc due to cost, cant wait to see the performance comparison, would love it if we can get LTT do a test with theirs?

[GOTSpectrum]

45 minutes ago, NetoriusNick said:

defiantly interested in the arc due to cost, cant wait to see the performance comparison, would love it if we can get LTT do a test with theirs?

That's something to bring up with @jakkuh_t

[IkeaGnome]

7 hours ago, NetoriusNick said:

defiantly interested in the arc due to cost, cant wait to see the performance comparison, would love it if we can get LTT do a test with theirs?

I'm not going to be setting alarms to get one, but I will more than likely end up with one. How long it'll take to get to Alaska and how much testing(if any) I'll be able to do before I go back to work has yet to be seen.