Folding Forum

ok, this is clearly a serious problem since our changes aren't working. We've been internally thinking of a plan. Here's where we are:
1) core17 update: we've come up with a workaround for core17 so it runs on Maxwell. It won't run as fast as it could (due to the driver issue), but it will at least run. It would run at full speed on non-Maxwell. We're pushing to get that new core17 out.

2) AS update: If we can't resolve the AS issues, we're thinking of reverting back to giving Core17 to everyone, including Maxwell GPUs, even before we get Core17 out, since this is a major issue for non-Maxwell donors.

I'm sorry we don't have better news at the moment, but this is where we are right now. We are actively pushing on both the core update and AS update fronts to fix this.

Thank you for the info.

Ben

Thanks for that, appreciated.

VijayPande wrote: 1) core17 update: we've come up with a workaround for core17 so it runs on Maxwell. It won't run as fast as it could (due to the driver issue), but it will at least run. It would run at full speed on non-Maxwell. We're pushing to get that new core17 out.

Any feel for the speed penalty eg as a % of what it should be? I've been thinking of getting a Maxwell card (for a Linux box) but will stick with Kepler if it's too slow.

I have just picked up 2 13000 wu, on 2 different computers both with no flag on 780 ti's.

Ben

Windows 8
V7.4.4
client-type is normal (the default)
GTX 660 Ti (Kepler GK104) -> Got 13001 (FahCore_17) @ 2014-04-10T20:34:26Z
GTX 650 (Kepler GK107) -> Got 13000 (FahCore_17) @ 2014-04-10T20:35:37Z

I guess the Servers are playing nice so those having issues getting FahCore_17 WUs on their Kepler GPUs might want to remove any client-type set and see if they get FahCore_17 or not.

billford wrote:Thanks for that, appreciated.

VijayPande wrote: 1) core17 update: we've come up with a workaround for core17 so it runs on Maxwell. It won't run as fast as it could (due to the driver issue), but it will at least run. It would run at full speed on non-Maxwell. We're pushing to get that new core17 out.

Any feel for the speed penalty eg as a % of what it should be? I've been thinking of getting a Maxwell card (for a Linux box) but will stick with Kepler if it's too slow.

Any speed penalty from a temporary work around would then also be temporary, and then the card would be back to normal speed. I am NOT asking for you to answer, but just consider how slow is too slow? How long before the permanent driver fix arrives is too long? Does that change your buying decision?

Nobody can predict when NV will provide a permanent driver fix.

Right now Maxwell should be getting Core_15 assignments which, compared to core_17, are slow (at least in terms of PPD). The choice between one version of slow and another version of slow won't really be apparent until the new version of Core_17 for Maxwell is tested, and I'm sure those who will be testing it will have an opinion.

7im wrote: but just consider how slow is too slow? How long before the permanent driver fix arrives is too long? Does that change your buying decision?

Possibly to your surprise, I've considered all those. They don't have definitive answers, but they'll be taken into account as best I can when I make the decision and I'll take the consequences.

If I get it right then I get it right, if I get it wrong then it's about $50 I could have saved (or should have spent) by buying the other card. Hardly life changing either way.

billford wrote:

7im wrote: but just consider how slow is too slow? How long before the permanent driver fix arrives is too long? Does that change your buying decision?

Possibly to your surprise, I've considered all those. They don't have definitive answers, but they'll be taken into account as best I can when I make the decision and I'll take the consequences.

If I get it right then I get it right, if I get it wrong then it's about $50 I could have saved (or should have spent) by buying the other card. Hardly life changing either way.

No, not surprised you'd say that. Not surprised you cut my quote short either.

7im wrote:Not surprised you cut my quote short either.

I (usually) only quote the bits I'm responding to, it makes for brevity

Bruce had adequately covered the first point (in effect- "how temporary is temporary?") and I saw no point in repeating it!

Anyway, I've had a good look around and tried (not entirely successfully) to understand how GPU specs relate to folding performance, and am leaning towards going for broke and getting a GTX 780 Ti. Though I suspect I'll have to drop the Ti

But either will severely limit my interest in Maxwell for some time to come

So, it sounds like the assign issue is now resolved?

It seem like my keplers on adv or beta ar at least getting sent to p1300x instead of 8018 but wont grab adv/beta work. Improved but still a little wonky.

Consistently getting 1300x wu's with no flag on 3 780 ti's.

Ben

I just picked up a 7626 wu with no flag on 780 ti, I guess I'll switch to advanced.

Ben

Changed to advanced and picked up a core 17 wu.

Ben