Folding Community Board

[GOTSpectrum]

1 hour ago, LazyDev said:

Whelp! There goes my RTX 2070. 2070 on the left, 2080 on the right.

 

Can't you RMA it?

[Juggy]

Does anyone know if you can fold on this?

 

https://www.amazon.com/Nvidia-Tesla-GDDR5-Cores-Graphic/dp/B00Q7O7PQA/ref=sr_1_12?dchild=1&keywords=RTX+3090&qid=1598704175&sr=8-12

[chaozbandit]

6 minutes ago, Juggy said:

Does anyone know if you can fold on this?

 

https://www.amazon.com/Nvidia-Tesla-GDDR5-Cores-Graphic/dp/B00Q7O7PQA/ref=sr_1_12?dchild=1&keywords=RTX+3090&qid=1598704175&sr=8-12

It's a kepler card so theoretically should be fine. Found a few threads of it working on VM instances.

 

[Juggy]

2 minutes ago, chaozbandit said:

It's a kepler card so theoretically should be fine. Found a few threads of it working on VM instances.

 

Thanks, I wonder if it makes sense in price/performance. $350 for 4992 Cuda cores seems a pretty decent deal unless I am missing something.

[LAR_Systems]

1 hour ago, Juggy said:

Thanks, I wonder if it makes sense in price/performance. $350 for 4992 Cuda cores seems a pretty decent deal unless I am missing something.

This likely does not make sense unless you're getting it for free to tinker with.

This is a data center card so you need to keep in mind that if you are not putting it in to a forced air server case, you will need to rig up a cooling fan and duct to force air through the card to keep it cool in a normal PC case.

When you factor the time / cost of doing that and that it's older generation so the power usage will be high as well you are probably better off getting a RTX 2060 Super in terms of purchase price, PPD and electricity cost to operate.

Have a look at the PPD DB https://folding.lar.systems/folding_data/gpu_ppd_overall  to see various NVIDIA cards to get an idea of performance and electricity cost.  Which may help your decision.  

I personally looked in to this kinda of stuff for about 4 months seeing used server cards on Ebay but eventually decided on consumer cards to avoid running a server case for data center cards with fans that sound like jet engines while folding.

Another thing to keep in mind, is for many of the data center cards you're paying for all the DRAM which for folding won't be needed at the sizes often found server tier cards.   Making it more affordable to find a comparable GPU processor model with far less RAM on the consumer side.

[marknd59]

1 hour ago, Juggy said:

Thanks, I wonder if it makes sense in price/performance. $350 for 4992 Cuda cores seems a pretty decent deal unless I am missing something.

I did the same kind of analysis as LAR_Systems and came to the same conclusion the RTX 2060 super is the sweet spot and bought 2 of them to upgrade my folding system. If you power limit them to 80% you drop the power draw down to ~135W with no real effect on PPD and you are getting ~519136 points per kWh. Here is a sheet I put togeather for the analysis I did of the different cards.

 

https://docs.google.com/spreadsheets/d/1iauimDyaUS0jnpeYqRNbM2uub-ASvS1OEz2yEUNJgls/edit#gid=0

[Juggy]

3 hours ago, LAR_Systems said:

This likely does not make sense unless you're getting it for free to tinker with.

This is a data center card so you need to keep in mind that if you are not putting it in to a forced air server case, you will need to rig up a cooling fan and duct to force air through the card to keep it cool in a normal PC case.

When you factor the time / cost of doing that and that it's older generation so the power usage will be high as well you are probably better off getting a RTX 2060 Super in terms of purchase price, PPD and electricity cost to operate.

Have a look at the PPD DB https://folding.lar.systems/folding_data/gpu_ppd_overall  to see various NVIDIA cards to get an idea of performance and electricity cost.  Which may help your decision.  

I personally looked in to this kinda of stuff for about 4 months seeing used server cards on Ebay but eventually decided on consumer cards to avoid running a server case for data center cards with fans that sound like jet engines while folding.

Another thing to keep in mind, is for many of the data center cards you're paying for all the DRAM which for folding won't be needed at the sizes often found server tier cards.   Making it more affordable to find a comparable GPU processor model with far less RAM on the consumer side.

Thank you for the great response, I come from a datacentre virtualisation background running large VM clusters on Dell blade and NetApp technology however my Tesla, Cuda, GPU compute knowledge is not that great. Having said that I do have a few R7xx and R9xx boxes laying about but it sounds to me that the power vs reward question makes it a futile exercise.

 

[Juggy]

3 hours ago, marknd59 said:

I did the same kind of analysis as LAR_Systems and came to the same conclusion the RTX 2060 super is the sweet spot and bought 2 of them to upgrade my folding system. If you power limit them to 80% you drop the power draw down to ~135W with no real effect on PPD and you are getting ~519136 points per kWh. Here is a sheet I put togeather for the analysis I did of the different cards.

 

https://docs.google.com/spreadsheets/d/1iauimDyaUS0jnpeYqRNbM2uub-ASvS1OEz2yEUNJgls/edit#gid=0

I would agree, I ran a Gigabyte RTX 2060 Super Gaming OC and price vs performance it really cannot be topped.

 

Hopefully when RTX3xxx is launched I will find enough spare daneros to keep my 2080 Super and build a folding only rig with a few spare parts I have lying around.

 

2080 Super plus 3090 should help the cause and the team a bit.

[LazyDev]

8 hours ago, GOTSpectrum said:

Can't you RMA it?

In the process of getting this done. It's a Zotac Mini 8GB.

[Brookler]

So close to Silver Badge

[Eviac2370]

Is there any kind of way, that folding home will support, using all cores but like 60-70% of the CPU? So instead of choosing how many threads you want going, then instead choosing how many percentage the entire CPU uses?

[justpoet]

When you set up the non-web client, you can go into Configure > Slots

In here, click on the CPU option and the Edit button.

 

Where it says CPUs at the top, the default is -1 for "use as many as this project likes", which is usually all of them.  You can set that to any number of threads you want, for example 10 if you have a 12 thread CPU and want to leave 2 free for GPU folding and general use.  I often set up 2 different CPU slots, with them totaling together the number of threads I want to use max, so I can tell one of the slots to finish ahead of expected heavier use of the system, while still letting the other slot keep chugging.  At least on older high core count multi-CPU server machines, this doesn't negatively affect PPD for me as long as they're both running.

 

There isn't a way to tell it to only load each thread at a certain percentage.  But it will adjust the priority of the process, both with the general slider and "Folding Core Priority" under Advanced.  With it low, it'll still try to not let there be idle time on the CPU, but anything else you start running will get priority over F@H.  I find best practice for a normally used machine is to leave the priority low and to keep 2 threads unused, and that basically makes the performance hit not noticeable unless you're running something else that really wants to do a sustained 100% stress load on the CPU (like a long software encode or similar).

[marknd59]

Is the Discord server down or have I just been kicked off for some reason ?

 

Edited -- forget I've figured it out.

[GOTSpectrum]

Folding month 2020 is live!!!!!!

 

 

 

[LAR_Systems]

4 hours ago, GOTSpectrum said:

Folding month 2020 is live!!!!!!

 

 

 

I feel my office heating up already.

[GOTSpectrum]

14 minutes ago, yaboistar said:

due to solicitors doing what solicitors do best, i am unable to confirm for sure my attendance

Just sigh up then you are covered either way!

[Eviac2370]

20 hours ago, justpoet said:

When you set up the non-web client, you can go into Configure > Slots

In here, click on the CPU option and the Edit button.

 

Where it says CPUs at the top, the default is -1 for "use as many as this project likes", which is usually all of them.  You can set that to any number of threads you want, for example 10 if you have a 12 thread CPU and want to leave 2 free for GPU folding and general use.  I often set up 2 different CPU slots, with them totaling together the number of threads I want to use max, so I can tell one of the slots to finish ahead of expected heavier use of the system, while still letting the other slot keep chugging.  At least on older high core count multi-CPU server machines, this doesn't negatively affect PPD for me as long as they're both running.

 

There isn't a way to tell it to only load each thread at a certain percentage.  But it will adjust the priority of the process, both with the general slider and "Folding Core Priority" under Advanced.  With it low, it'll still try to not let there be idle time on the CPU, but anything else you start running will get priority over F@H.  I find best practice for a normally used machine is to leave the priority low and to keep 2 threads unused, and that basically makes the performance hit not noticeable unless you're running something else that really wants to do a sustained 100% stress load on the CPU (like a long software encode or similar).

So you're saying. 2 threads is more than enough for an RTX 2070 super? (More specific: i9-9900KF 5ghz "boost mode", RTX 2070S asus strix 8gb) cooled with a corsair H115i "280mm". I normally run 6 CPU, as it runs about 80C avg

[Eviac2370]

On 8/30/2020 at 10:08 AM, yaboistar said:

that's how it works iirc, i don't think fah has control over individual threads but it puts a load on representative to, say, 10/12ths of the cpu's max load

 

i'd frankly prefer to be able to tell it to use individual threads so i could then leave one or two threads spare to set nvidia drivers to use so i can run cpu folding without the gpu penalty that outweighs the benefit of adding the cpu in

I get you, but i don't do anything else on my computer, as i have it running while sleeping. So to me that's irrelevant

[marknd59]

12 minutes ago, Eviac2370 said:

So you're saying. 2 threads is more than enough for an RTX 2070 super? (More specific: i9-9900KF 5ghz "boost mode", RTX 2070S asus strix 8gb) cooled with a corsair H115i "280mm". I normally run 6 CPU, as it runs about 80C avg

You should leave 1 thread for each Nvidia card you have in your system.

[Juggy]

20 hours ago, GOTSpectrum said:

Folding month 2020 is live!!!!!!

 

 

 

Thank you for this, I have signed up. Who knows, maybe the RTX3090 will arrive by then

[NetoriusNick]

both of the above list have been helpful for getting a new card thanks

[justpoet]

Just in time for folding month signups….

 

The perpetual "Ready".

 

Latest server to do it (right now) is assigning me, but not responding.  Temple's server: 155.247.164.213

Pings work fine, just can't get assignment connections.

[GOTSpectrum]

20 minutes ago, yaboistar said:

aye captain, set phasers to fruit salad

The most tasty of phasers! 

[GrockleTD]

7 minutes ago, GOTSpectrum said:

Sorry GOTS my phone is acting up I spent 20 minutes trying to get rid of this quote box

[Juggy]

I watched the whole launch but is this post from GamneSpot real? That would make the 3090 twice as powerful as the 2080Ti with newer gen Cuda cores

 

https://www.gamespot.com/articles/nvidia-reveals-rtx-3090-3080-3070-video-cards-pric/1100-6481673/