FahCore17 for AMD is Here!!

[10e]

So far people with AMD 7870 to 7970 are seeing up to 40,000 PPD with the FahCore 17 units, so seemingly it's no joke these days. It's really about time.

If this is about science then utilize the hardware available to a proper level barring software not being up to snuff (like nVidia's OpenCL drivers)..

goo.gl/fYx2d

Edit: I'll be checking this out tonight on my main rig with my 7870 LE which is halfway between a 7870 and halfway between a 7950. This might be a decent PPD/Watt tradeoff with this card.

So far the regular 7870 Ghz editions are running nearly 40,000 PPD.

Too bad my 7950 died and needs RMA. Gosh darnit.

[Inik]

I will try to set-up my 7970 soon to see how it goes!

[lowfat]

'bout time we we get this.

[Hooded]

Does it help the older 6970's as well?

[great_big_abyss]

Originally Posted by muziqaz

Remember Pande Group's main objective is science. They write scientific papers not on how many points does Kepler lose but what has been done scientifically with the project.

While I agree with you on all your other points, I take small issue with this bit. Pande Group gets to write their scientific papers specifically because we, the users, are donating our extra clock cycles. Without us, the individual hardware owners, Pande Group and FAH would not exist.

Our reasons for donating are varied and ample. Many of us have known and lost friends and family members to various diseases. Some of us just want to see debilitating diseases eradicated from the face of the earth. We all have great reasons to contribute to FAH. One additional key reason is the points system. It promotes friendly competition, and makes donating 'fun'. Personally, many of the extra units I've brought online (at personal expense) have been to earn extra points, in order to better compete with other members of this forum. If points are taken away, reduced, or awarded 'unfairly', then some of the incentive disappears. People want to donate their extra clock cycles and electricity bills, but they want to have fun doing it. This is why you see some users migrating towards other avenues of distributed computing, such as BOINC.

Pande group needs to make sure to keep contributers happy with a predictable supply of points that is fair from platform to platform. It's not as if it costs Pande Group any money (aside from a few programmers) to award points.

______________________________________________

I have seen increases on the 6xxx series. The 6950 currently in my main rig is doing about 15K PPD, an increase of about 9K from core16.

[sswilson]

Originally Posted by muziqaz

HI there. As I was a part of the core pre beta testers group I would like to clear some thing up and give you guys some pointers about this new core.

This core uses OpenCL, not closed CUDA, but OpenCL. nVidia have their baby, which at this moment is not OpenCL, so their support for it is way lacking. This is one of the reasons 6 series are doing quite bad with this core. AMD is giving their best efforts to optimise their hardware for OpenCL, as this is becoming industries standard and it is open, that is why we see such improvement with their hardware on this core.
As you can see from release notes at proteneer's blog, titan cannot even run this core because nVidia lack of dedication to OpenCL.

As we all know 4 and 5 series (Fermi) cards were developed as compute beasts by nVidia and you should remember a lot of people ridiculed them and gave grief about it, as it was too hot, too big, too expensive and very late. Now nVidia released gamers card which is 6 series with most of the compute stuff cut out. Thus we have small, cold GPU core which is excellent for gamers, BUT is lacking in compute department, so again this is the second reason why 6 series is doing so 'badly'. It is capable to be as powerful as Fermi in compute perhaps, but this requires major dedication from nVidia to optimise drivers and API.

Now you might ask why proteneer went OpenCL way instead of CUDA with this core. Answer is simple: to make code as unified as possible and to make the code run on both nVidia and AMD gpus. It is pretty incredible that now we have two completely different architectures being able to run the same exact WU.
Also to note: this core at this development stage was being pushed to make it just work. Performance optimisations according to dev will come later. Now we sorted out most of the bugs and errors, so dev can dedicate his effort in performance optimisations. Now we get around 45k PPD on 7970s, but according to dev there is still plenty of optimisations for either nvidia and AMD. I cannot say any numbers, but the potential is quite huge.

So as this core is still in beta, I would urge everyone to calm down and give proteneer a chance to optimise this sucker, as he already made miracles with making this core run.

Most of you are moaning about 6 series running this core slow. In order to run this core on 6 series you have to use beta argument, and unless all of you who moan are in beta testing team, I don't see the reason why would you run beta argument if you are not in the beta tester team and moan. Running beta WUs while not on beta testing team is damaging to project, as problems are not being reported and in a way that is why we have this situation with core 17 and Keplers, because you guys are running what you are not really supposed to run in the 1st place. Those who run beta on 4 and 5 series and are not in beta team, just drop beta argument and enjoy fahcore 16.

In the end, I should give you guys a lot of credit, as you took this news way better than say EVGA folder forum people :D Remember Pande Group's main objective is science. They write scientific papers not on how many points does Kepler lose but what has been done scientifically with the project.

And you have to agree fahcore 17 is pretty amazing as it managed to bring AMD back into folding, and that on its own is absolutely great news

Sorry, but no... this isn't my issue with the current state of Nvidia folding, I stopped using beta / adv flags when they snuck the V2.25 core 15 in under the -advmethods setting.

The problem isn't -beta folding, it's with any WU other than the core 15 7XXX WUs. They aren't using kepler hardware to it's fullest extent as they only peg GPU power @ 80%. There's no doubt in my mind that the problem could very well be on Nvidia's end, but the continuing claims that these particular issues are related to lower compute ability seem to be misplaced if not outright dishonest.

[10e]

Originally Posted by Hooded

Does it help the older 6970's as well?

I think so. 6950s are getting over 13,000 PPD according to that google table I posted earlier.

The big deal is the lack of CPU usage now, which not only allows AMD GPU folding, but GPU folding in concert with CPU folding to maximize PPD. So, for example if you get a high speed I7 CPU using Sandy/Sandy Bridge-E or Ivy Bridge with a healthy overclock, you can still get more points from your AMD GPU.

It used to be that using an AMD GPU was counter-productive as it would take away enough of a core to slow down CPU folding, and thus cause a loss of overall PPD due to the Quick-return-bonus scheme.

Edit: For those on AMD 5/6000 series it's recommended you stick with Catalyst 13.1 drivers. For those on AMD 7000 series it's recommended you use Cat 13.2 beta.

So far on Catalyst 13.1 my 7870 "Tahiti" is getting over 20,000 PPD on a 7662 unit. I'll upgrade tonight to see if it improves. This is at 1025mhz in AMD Overdrive.

[MARSTG]

I added the -beta flag in client 7.2.9 for my GT430 and got a FaH Core 17 WU. However the PPD prediction plummeted dramatically from over 6k PPD to 2k PPD. Removed the flag and got my regular Core15 WUs.

[lowfat]

These are beta only? Guess I should remove my flags then.

[MARSTG]

I don't know if they are beta or not, I am just saying what happened in my case.