Folding Community Board

[Pezui]

I too admit that I have a problem. I have been trying to get under 300th place for a while. I could turn on CPU folding but the point to heat ratio is just bad. I have clawed up to 306th and with some solid work units I think I can catch 300th by mid march. On the other hand in team... uh... catching anyone is a long slog, and lord knows @Igor will smash me if I get too close for too long. 

[justpoet]

Welp, it got that 1 WU done, but it was failing hard on the next couple.  Got it going again, will see if it manages to hang on to complete another.

I can only guess there must be some expected instructions in some of the WU that the pentium can't really handle.  It shouldn't be a RAM thing, as it is 8gig running a stripped down ubuntu test install with nothing else running.

[justpoet]

17 hours ago, justpoet said:

Welp, it got that 1 WU done, but it was failing hard on the next couple.  Got it going again, will see if it manages to hang on to complete another.

I can only guess there must be some expected instructions in some of the WU that the pentium can't really handle.  It shouldn't be a RAM thing, as it is 8gig running a stripped down ubuntu test install with nothing else running.

Couldn't do it.  Kept failing.  Now running prime95 to see if that uncovers something requiring a return vs a random instruction set issue.

[Gorgon]

7 hours ago, justpoet said:

Couldn't do it.  Kept failing.  Now running prime95 to see if that uncovers something requiring a return vs a random instruction set issue.

Don’t you mean mprime (e.g. the Linux version)?

 

Have you run memtest on it yet? Could be RAM

 

I was running BOINC on an Asus Chromebox with an even slower processor recently and 2GB DDR2 for shits ‘n giggles. Slow as molasses but entertaining. My i3 Skylake NUC was appropriated by the kids recently for a Minecraft server and I finally powered down my Athlon x64 II system likely for good this time.

[justpoet]

1 hour ago, Gorgon said:

Don’t you mean mprime (e.g. the Linux version)?

 

Have you run memtest on it yet? Could be RAM

 

I was running BOINC on an Asus Chromebox with an even slower processor recently and 2GB DDR2 for shits ‘n giggles. Slow as molasses but entertaining. My i3 Skylake NUC was appropriated by the kids recently for a Minecraft server and I finally powered down my Athlon x64 II system likely for good this time.

Yeah, mprime.  I didn't memtest it, but that's a good idea.  It handled the mprime torture workload for more than 24 hours with no issues or errors.

[TVwazhere]

On 2/18/2022 at 10:26 PM, CT854 said:

Gotta be honest, though, I'm 100% ripping off @vsteel

On 2/20/2022 at 7:08 AM, vsteel said:

That meme was generated by @TVwazhere  I can't get any credit for it.

The original was this one:

[rkv_2401]

Linus and Jake run Folding@home on an nVidia A100 :

 (Around the 16:45 mark).

 

I guess that counts as a shoutout to F@h? They're using LARSystem's client too.

6 million PPD seems kind of low for a card of this size. Aren't 3080s able to match that?

[CT854]

2 hours ago, rkv_2401 said:

I guess that counts as a shoutout to F@h? They're using LARSystem's client too.

6 million PPD seems kind of low for a card of this size. Aren't 3080s able to match that?

I do get ~6M PPD on my 3080.

 

I wonder whether there's anything about how jobs are written that stumble particularly hard on cards like the A100.

[RollinLower]

2 hours ago, CT854 said:

I do get ~6M PPD on my 3080.

 

I wonder whether there's anything about how jobs are written that stumble particularly hard on cards like the A100.

i mean, most of the F@H GPU units are optimised for CUDA these days, and the A100 has about half of the CUDA cores of a 3090. 

 

it makes sense too, most people who run F@H will not be buying 10K GPU's so the program is really optimised for consumer grade hardware, not datacenter beasts like this.

[rkv_2401]

53 minutes ago, RollinLower said:

i mean, most of the F@H GPU units are optimised for CUDA these days, and the A100 has about half of the CUDA cores of a 3090. 

 

it makes sense too, most people who run F@H will not be buying 10K GPU's so the program is really optimised for consumer grade hardware, not datacenter beasts like this.

Agreed, last I heard they hadn't created optimized WUs even for the 3080Ti/ 3090.The LOW core and boost clocks(765Mhz and 1400Mhz), plus the low power limit compared to a 3080 probably don't help either.

[GOTSpectrum]

 

 

 

[NetoriusNick]

likely the wrong place but any give a guidance on ssd providers to stay away form, need a bunch (10) ssd for work for critical and non critical. All systems are backed up separately but I dont want to to have to replace drives frequently but they need to be cost effective as £50-£60  for a 120gb sandisk will make my boss kill me

[Gorgon]

2 hours ago, NetoriusNick said:

likely the wrong place but any give a guidance on ssd providers to stay away form, need a bunch (10) ssd for work for critical and non critical. All systems are backed up separately but I dont want to to have to replace drives frequently but they need to be cost effective as £50-£60  for a 120gb sandisk will make my boss kill me

Honestly if your just replacing spinning rust with SSDs then you'd likely not notice the difference between 2.5" SATA and m.2 NVMe as the difference between just a HDD and a SSD is so huge.

 

These Kioxia 240GB look like a good deal at £25.98 but the old adage of "No one ever got fired for buying IBM" holds true, substituting Samsung for IBM of course, and the Samsung come with a good migration utility for Windows whereas most other vendors leave Data Migration up to you.

 

If you really MUST have m.2 NVMe then I'd avoid the GigaByte and look at something like the 250GB WD Blue's at £38.99

[BondiBlue]

8 hours ago, NetoriusNick said:

likely the wrong place but any give a guidance on ssd providers to stay away form, need a bunch (10) ssd for work for critical and non critical. All systems are backed up separately but I dont want to to have to replace drives frequently but they need to be cost effective as £50-£60  for a 120gb sandisk will make my boss kill me

In that kind of a situation where cost is a big factor I'd choose between Samsung or Crucial. I have several drives from each, and I've had nothing but good luck from them so far. My previous employer had a fleet of Lenovo ThinkCentre desktops that used WD SSDs, and I can't recall any of those having issues either. 

[NetoriusNick]

16 hours ago, Gorgon said:

Honestly if your just replacing spinning rust with SSDs then you'd likely not notice the difference between 2.5" SATA and m.2 NVMe as the difference between just a HDD and a SSD is so huge.

 

These Kioxia 240GB look like a good deal at £25.98 but the old adage of "No one ever got fired for buying IBM" holds true, substituting Samsung for IBM of course, and the Samsung come with a good migration utility for Windows whereas most other vendors leave Data Migration up to you.

 

If you really MUST have m.2 NVMe then I'd avoid the GigaByte and look at something like the 250GB WD Blue's at £38.99

sorry sata only it was more of who to avoide as i see some that are 11 a go but they have the same % of reviews that say the drive arrived dead as some of the bigger well know manufatures

[Gorgon]

5 hours ago, NetoriusNick said:

sorry sata only it was more of who to avoide as i see some that are 11 a go but they have the same % of reviews that say the drive arrived dead as some of the bigger well know manufatures

I've had good results with Samsung, Crucial and SanDisk over the years. Had one Adata 6000-series NVMe fail and one Intel 6-series NVMe fail but the Intel was in my ESXi server and was 90% full and hosting an app with tons of writes.

 

For work I'd stick Big Brands that have been in the space for a while such as Samsung, Crucial, WD etc,

[LAR_Systems]

On 2/17/2022 at 3:27 AM, RollinLower said:

whoa be carefull now, you sure the wiring in your house can handle such loads?

Please don't burn your place down, monitor your PSUs at all times.

[TVwazhere]

32 minutes ago, LAR_Systems said:

Please don't burn your place down, monitor your PSUs at all times.

And dont use sketchy wires/adapters

[LAR_Systems]

On 2/24/2022 at 12:28 AM, rkv_2401 said:

Linus and Jake run Folding@home on an nVidia A100 :

 (Around the 16:45 mark).

 

I guess that counts as a shoutout to F@h? They're using LARSystem's client too.

6 million PPD seems kind of low for a card of this size. Aren't 3080s able to match that?

I have been busy working and meanwhile this dropped when I was not looking.

My cousin just flipped it over to me with an LOL.

 

So the A100 https://folding.lar.systems/gpu_ppd/brands/nvidia/folding_profile/ga100_a100_pcie_40gb work unit that they got was for a Cancer project, so that's going to be lower on the PPD and they only ran the test for one WU so that's all we have a sample.

 

But for comparing the A100 to the 3090 that Cancer Wu as in the DB the site has the data https://folding.lar.systems/projects/folding_profile/18021 based on others submissions on the same project.

 

Average PPD 3090 = 6,333,449

Average PPD A100 = 5,886,775

 

Given the power usage difference pointed out in the LTT video that is a massive PPD performance for the power used on F@H.  Edit: In a form factor you can jam in to a rack case!

The A100 if left to run a full spectrum of different work unit types over the course of a week would crush the 3090/3080 in terms of PPD per watt I'm guessing.

Anyone know how to get @jakkuh_t to leave it on for us

Edited March 3, 2022 by LAR_Systems
Adding thought.

[NetoriusNick]

7 hours ago, LAR_Systems said:

I have been busy working and meanwhile this dropped when I was not looking.

My cousin just flipped it over to me with an LOL.

 

So the A100 https://folding.lar.systems/gpu_ppd/brands/nvidia/folding_profile/ga100_a100_pcie_40gb work unit that they got was for a Cancer project, so that's going to be lower on the PPD and they only ran the test for one WU so that's all we have a sample.

 

But for comparing the A100 to the 3090 that Cancer Wu as in the DB the site has the data https://folding.lar.systems/projects/folding_profile/18021 based on others submissions on the same project.

 

Average PPD 3090 = 6,333,449

Average PPD A100 = 5,886,775

 

Given the power usage difference pointed out in the LTT video that is a massive PPD performance for the power used on F@H.  Edit: In a form factor you can jam in to a rack case!

The A100 if left to run a full spectrum of different work unit types over the course of a week would crush the 3090/3080 in terms of PPD per watt I'm guessing.

Anyone know how to get @jakkuh_t to leave it on for us

an now all I can do is dream

[rkv_2401]

11 hours ago, LAR_Systems said:

I have been busy working and meanwhile this dropped when I was not looking.

My cousin just flipped it over to me with an LOL.

 

So the A100 https://folding.lar.systems/gpu_ppd/brands/nvidia/folding_profile/ga100_a100_pcie_40gb work unit that they got was for a Cancer project, so that's going to be lower on the PPD and they only ran the test for one WU so that's all we have a sample.

 

But for comparing the A100 to the 3090 that Cancer Wu as in the DB the site has the data https://folding.lar.systems/projects/folding_profile/18021 based on others submissions on the same project.

 

Average PPD 3090 = 6,333,449

Average PPD A100 = 5,886,775

 

Given the power usage difference pointed out in the LTT video that is a massive PPD performance for the power used on F@H.  Edit: In a form factor you can jam in to a rack case!

The A100 if left to run a full spectrum of different work unit types over the course of a week would crush the 3090/3080 in terms of PPD per watt I'm guessing.

Anyone know how to get @jakkuh_t to leave it on for us

Thanks for the comparison! The difference is much smaller than I expected, and the A100 only draws ~60 to 70% of the power of a 3090. That's actually pretty interesting tbh.

[rkv_2401]

I've been seeing a lot of YT commenters recently asking about ways to support LTT/LMG, such as on the recent LMG clip titled "Why LMG needs more money".

Just thinking out loud here; do you guys think folding for the LTT folding team could be a potential a way of showing support and engaging with LMG's content? We're already the 2nd most productive Folding team after banano.cc, and it costs very little in electricity to run Folding@home to achieve what I believe is a big contribution to healthcare research. Children and students strapped for cash (such as myself) may be more inclined to show support this way - I know I do.

I continue to fold with the LTT team because of the community here, but I wouldn't have found them if I weren't a fan of LMG's work!

 

[LazyDev]

Well, my computer has developed an intermittent fault that I cannot identify.

 

The system would crash to a black screen. The CPU Ez led would be lit, and the debug code would be 00. After a minute or so, the system would boot as normal automatically. Sometimes, it would throw a BSOD, but it would be so fast, that the screen would go black as it was in a gradient to said blue screen. I've never been able to identify the error code, as windows doesn't have time to log it.

 

Other than that, it works fine. Just "fine".

 

Edit

 

I'm going to be looking for a more suitable chassis. It'll need to have space for an ATX motherboard, two power supplies, and at least 9 GPU's.

 

4 PSU's if I choose to go get some 4xxx GPU's.

[TVwazhere]

9 hours ago, LazyDev said:

Edit

 

I'm going to be looking for a more suitable chassis. It'll need to have space for an ATX motherboard, two power supplies, and at least 9 GPU's.

 

4 PSU's if I choose to go get some 4xxx GPU's.

A mining Case would likely be your best bet. 

https://www.amazon.com/Bitcoin-Mining-Convection-Expert-Accessories/dp/B09J2JQ5LB/ref=dp_fod_2?pd_rd_i=B09J2JQ5LB&psc=1

https://www.amazon.com/dp/B07H41S74S/ref=emc_b_5_t`

[Pezui]

Anyone else been getting some sweet work units recently? I am pushing past 18M ppd right now which is fantastic!