F@H News: Core22 back at Beta (5700XT Navi runs!)

[Metallus97]

Just a quick update for all of you with (or planing to get) AMD 5700(xt)/5500(xt) etc.

Finally the beta testing for Core22 wich now fully supports the RDNA architecture is back at beta testing.

So if you want to help testing set your GPU Slot to beta and you should be assigned some work units form core22. The more of us do this, the faster they can develop the Core22.

BUT PLEASE: If your experience frequent crashes dont let it run, since that creates huge piles of unnecessary work/data to go through for the F@H team. IF you’re experience problems switch back to regular and let them know about it. 

 

First reports show 620k-700k PPD on the card wich is pretty ballin at the price. Other Cards also get a performance uplift from about 10%.

This is due to Core22 being built on a MUCH more modern version of OpenMM (thats the API used to compute. Supported by both team green and red).

 

I am REALLY happy to hear this, because I was hesitating on getting a 5700xt, because I want my card not only for gaming but also for F@H!

[Spotty]

1 hour ago, Metallus97 said:

First reports show 620-700PPD

620-700 PPD? Damn, my Core2Duo CPU does more than that. /s

But 600-700k for the RX5700 XT? That doesn't sound right. Am I missing something here? Is Core22 much harder to work and the payout lower?
Isn't that less than what you can get with a Vega56/GTX1070? Also much lower than the gaming performance equivalent Nvidia RTX cards. Much lower than the GTX1080Ti even.
 

RX 5700 XTs are $400-$500 depending on the model. You could get a used 1080Ti for the price of a new RX 5700 XT, and if those PPD numbers are anything to go by, the 1080Ti would be the much better option with around 1-1.3M PPD.

Vega 56 or a GTX1070ti can probably be had for around $250 or so 2nd hand, and the 1070Ti would be more power efficient.

[Metallus97]

21 minutes ago, Spotty said:

620-700 PPD? Damn, my Core2Duo CPU does more than that. /s

But 600-700k for the RX5700 XT? That doesn't sound right. Am I missing something here? Is Core22 much harder to work and the payout lower?
Isn't that less than what you can get with a Vega56/GTX1070? Also much lower than the gaming performance equivalent Nvidia RTX cards. Much lower than the GTX1080Ti even.
 

RX 5700 XTs are $400-$500 depending on the model. You could get a used 1080Ti for the price of a new RX 5700 XT, and if those PPD numbers are anything to go by, the 1080Ti would be the much better option with around 1-1.3M PPD.

Vega 56 or a GTX1070ti can probably be had for around $250 or so 2nd hand, and the 1070Ti would be more power efficient.

Holy crap I forgot the k 

its 620.000-700.000 PPD!!

sorry!! Edited it!

 

allso bare in mind, that Core22 is at version 0.0.2 right now and only doing validating work, not real since. Therefore PDD can change drastically. My VII is getting about the same PDD and my good old 290X got a boost of about 5-10%. Everything on Win10 latest version + AMD drivers latest version 

[Gorgon]

5 hours ago, Metallus97 said:

... BUT PLEASE: If your experience frequent crashes dont let it run, since that creates huge piles of unnecessary work/data to go through for the F@H team. IF you’re experience problems switch back to regular and let them know about it. 

Good advice but the Beta area to report any such issues is read-only and only Beta Team members can post there.

[Metallus97]

9 hours ago, Gorgon said:

Good advice but the Beta area to report any such issues is read-only and only Beta Team members can post there.

Yes, I applied there, since I am also active in this forum and will be getting a Navi on XMas

 

4 hours ago, yaboistar said:

it looks like one of the main differences between 21 and 22 is that low-atom-count WU's are actually processing at reasonable speeds on turing based now.

 

14196 - 64,514 atoms - runs at 2.8m on core21

11737 - 23,558 atoms - runs at 3.1m on core22

 

i'm interested to see how 11738 with it's 287,229 atom count will run. part of me thinks it;ll be just a few hundred thousand ppd difference, but there's also a part of me that hopes my 2080ti starts throwing out 4-5mppd lmao

Thanks for you testing! Very interesting!

at the Moment it seems my VII is a bit slower on 22 than in 21 but I’ll have to wait for more WUs to finish.

ill post an update this night 

[Metallus97]

UPDATE:

So far the combined error rate is 0,2%. Very nice! they will be pushing out more intensive projects over Christmas and are planing to go on advanced after the holidays! this is then just one step away from being used for real science. 
but sure the research groups first will have to create/redo projects for core 22

[Metallus97]

About my rig:

 

At stock my VII loses about 15% and 290x 20% of PPD (that’s also reflected in the TPF). Don’t know why and am now experimenting with OCing the cards.

power usage and utilization is the same. Although I have a about 30% increase in usage on the CPU thread that goes along with Core22 (but that ain’t a problem)

[Metallus97]

17 minutes ago, yaboistar said:

for us we're getting a nice ppd bump but it's also drawing a lot more power to do the computations. perhaps it's fully utilising all the cores available or all the cycles available now? let's face it, core21 was designed for, what, fermi?

Yes, 21 is old. But berry interesting to see my AMD stuff get hit so bad

[Spotty]

23 hours ago, yaboistar said:

my 1080Ti is now at 1.7m up from 1.2-1.4m which is another respectable bump

Wow! That's a decent increase. That should help me if we end up doing that competition with PC Games Hardware.

 

[Ithanul]

On 12/21/2019 at 2:00 AM, yaboistar said:

for us we're getting a nice ppd bump but it's also drawing a lot more power to do the computations. perhaps it's fully utilising all the cores available or all the cycles available now? let's face it, core21 was designed for, what, fermi?

Yep.  Core 21 came out around back than.  I recall a bit when it first came out.  There was a love/hate relationship back then as it was getting dailed-in on hardware.

I even remember Core 17s as it was a big deal as that allowed folding under Linux, previous Cores for GPUs where solely stuck to x86 Windows.  Sure don't miss though folding on a GTX580.

 

So, Core 22 is probably going to go through the dail-in phase for a bit.

 

I will see if my 1080Ti can grab one so I can see what it gets on PPD.  Probably be a good break from crunching GFN 21s over on PrimeGrid.

 

Got some Core 22s on my 1080Ti.

OS: W10 LSTB

Driver: 436.15

GPU Model: EVGA GTX1080Ti FTW3 on stock settings

 

Averaging around 1.66K or so.

[minibois]

I have F@H running part-time on my 2070 Super. Switched to BETA today and saw some nice improvements.

Keep in mind - again - I don't run F@H 24/7, so my output numbers don't reflect the PPD F@H estimates..

 

Anyways, C21 WU's I get like ~1.3m PPD, give it take depending on the WU (you know how it goes...)

C22 gives me between 1.4m and 1.5m PPD (it sometimes likes to report 1.6m to get my hopes up, but then settles down to usually 1.45m-1.5m)

 

Is it me though, or is everyone else also getting 'small' WU's? Like I used to get 200K WU's normally, but now I have gotten quite a few WU's that are around 80K-120K.

[sazrocks]

11739 mobile GTX 1660 @ 700k PPD. Holy cow.

11738 GTX 1080 @ 710K PPD. Terrible, this card should be getting 850k+.

[sanya567xxx]

I am literally chewing through this  (still my 1080ti with that modded bios and clocks at 2038 1.125v, mem 5164, fan 75, temp stabilized at under 65)

[RollinLower]

any newer news on RX 5700 performance for these WU's?

[Metallus97]

I think weather your perf is hit or boosted really is connected to how high or low atom count WUs you get.

I HOPPPPPPE that the production version of Core22 will have much high atom count ones  

[Metallus97]

UPDATE FROM THE FOLDING FORUM BETA THREAD: Core22 project 11737 was taken to advanced (over Christmas etc the failure rate was 0,3%. Thats pretty awesome!).

Navi users shall retain the beta flags. New core22 projects are 11738/39 and specifically for Navi testing. Navi on advanced wont get any Core22 atm.

 

So Imma switch my rig to ADV right now boizz!

[minibois]

18 hours ago, minibois said:

Anyways, C21 WU's I get like ~1.3m PPD, give it take depending on the WU (you know how it goes...)

C22 gives me between 1.4m and 1.5m PPD (it sometimes likes to report 1.6m to get my hopes up, but then settles down to usually 1.45m-1.5m)

Update on this, while my 2070 Super normally hits about 1.3m PPD, yesterday I got some 1.4-1.5m PPD WU's and now the current WU and the last WU I worked on has been close to 2m PPD on this single (not overclocked) card

This is on 11738.

 

I'm liking Core 22  

EDIT: next WU I went down to 1.75m-1.8m PPD. Still quite a bit higher!

[sazrocks]

Saw the 1080 system was up to 1.3m PPD and thought that it had finally sorted itself out. Nope, 1080 was still at 700k PPD. The CPU, on the other hand was at 600k+ PPD running on only 6 threads... what the heck?

[Metallus97]

1 hour ago, sazrocks said:

Saw the 1080 system was up to 1.3m PPD and thought that it had finally sorted itself out. Nope, 1080 was still at 700k PPD. The CPU, on the other hand was at 600k+ PPD running on only 6 threads... what the heck?

Whoa! Do you have info on wich project that happened?

[sazrocks]

1 hour ago, Metallus97 said:

Whoa! Do you have info on wich project that happened?

It's dropped back down to sane levels on the next WU, but the project is 14051.

[Metallus97]

6 hours ago, yaboistar said:

i'm gonna retain beta on the turing system for that sweet 4.7mppd but i'll switch everything pascal and below back onto advanced to take advantage of 11737 again. 0.3% is so low it could be some people dumping or crashing out due to unstable OC's or BSOD's tbh

You’re right. That’s it an therefore they more on.

damn am I exited to see what big Navi has to offer. I HOOOOOPE some power!

ill switch out my VII then for the main gaming card, but keep it for F@H 24/7

[RollinLower]

On 1/6/2020 at 7:28 PM, Metallus97 said:

You’re right. That’s it an therefore they more on.

damn am I exited to see what big Navi has to offer. I HOOOOOPE some power!

ill switch out my VII then for the main gaming card, but keep it for F@H 24/7

just curious: what's the powerdraw like on that VII when folding? 

[Zando_]

1 minute ago, RollinLower said:

just curious: what's the powerdraw like on that VII when folding? 

If tuned right it shouldn't be all that much. My RVII pulls about 180W under F@H load, running 1900MHz on the core (which results in 1770MHz or so on the core, the stock 1800Mhz target results in around 16-1650Mhz on the core under F@H load) and a -20% power target. 

[RollinLower]

3 minutes ago, Zando Bob said:

If tuned right it shouldn't be all that much. My RVII pulls about 180W under F@H load, running 1900MHz on the core (which results in 1770MHz or so on the core, the stock 1800Mhz target results in around 16-1650Mhz on the core under F@H load) and a -20% power target. 

that's not that bad actually. there's been some VII's going for pretty cheap on the used market over here, so i'm torn between upgrading my 1070's to VII's or going turing.

[Zando_]

2 minutes ago, RollinLower said:

that's not that bad actually. there's been some VII's going for pretty cheap on the used market over here, so i'm torn between upgrading my 1070's to VII's or going turing.

If you get good WUs and use the latest WQHL Adrenaline 2019 drivers, it should be solid. Around 1.3-1.9m PPD on my RVII, depending on WUs. Seems to do about that with either Core22 or Core21 units. 

Back on older drivers it would do around 700-800K, which is only slightly more/sometimes less than what my 1660 Ti did. And that was at full power consumption (around 250W, mebbe a bit higher) and an OC above stock clocks vs the 1660 Ti with a mild OC, and those cards can't really pull over 130W no matter how hard they try.