Change in BA requirements

[ChristianVirtual]

viewtopic.php?nomobile=1&f=24&t=25410

Thanks for the heads up on changes for BA;

and part of the change in bigadv threshold is because we would like to encourage moderately powerful machines to help boost the capabilities of non-bigadv SMP projects where we do a lot of this science.

Is there in addition any plan to extend QRB to regular CPU projects ? That would support the stated importance of science and could help to keep people motivated operating the current low-end BA systems. Just wonder.

Any new Core 17 in the pipeline ?

All comes down to: where to invest/spend next ...

[7im]

Um, CPU projects have had QRB since the A3 core was released several years ago. No new CPU projects have been released that didn't have QRB in many years.

And as stated before, yes, there are more Core 17 projects coming.

Also, it is ill advised to make purchasing decisions on minute by minute news updates. Better to watch the long term trends.

[Nathan_P]

QRB is already active on all SMP projects, my soon to be retired dual x5670 rig will crank out 75k PPD on SMP, not that i'll be running it as it sucks too much power now to make it worthwile

As for the announcement, the 1st one is irrelevant anyway. The realistic minimum has been 24 cores for some time now. As for the 2nd part - well it consigns all the 4p socket F machines, most G34 2p machines and all the Hex core intel xeon cpu's with HT - even the new ivy bridge ones, to smp only.

I'd like to see the conversion rate on those going to SMP only.....

As for where to invest next, well either 4p - preferably intel, 2p has to be intel and may as well be 10 or 12 core ivy bridge for BA units. Core 17 - well more projects would be nice,

otherwise scoop up a soon to be sold "slow" 2p machine from someone and crank out a load of SMP units

[ChristianVirtual]

QRB is already active on all SMP projects,

My bad, just looked again for k-factors. Some zeros, many above; I'm 90% GPU folder

[ChristianVirtual]

Better to watch the long term trends.

Please tell me what you see ? GPU (Core 17) > CPU/SMP ?
Any GPU-BA ? What requirements ... If I buy now GPU with 3GB will that be enough ? Need to be 3.5 GB ? I know, memory is not of big concern for folding, but again who knows what changes coming up.

I really think we should see some roadmap from PG on what's in the pipeline. How many Core 17 projects we can welcome in 2014ff. Even if some projects get stopped on the way, that's fine. But for me as regular donor I don't have any basics for planning.

[Nathan_P]

Here's an idea. Leave BA as it is, introduce new larger BA WU for the 4p machines, That way all the slower BA machines won't get shut down.

There is too much of gap between SMP and BA units, SMP runs the whole spread from 2 threads to 128 and BA runs from 16(24) to 128 but there are problem area's. Some machines are still potent enough to get a BA WU done in reasonable amount of time ( 3 days or less) but are currently forced to run stock SMP.

IMHO SMP should cover up to 12 threads, 12-24 Threads should have larger SMP units, Fast 24 up to 48 thread machines can run BA units, and 4p machines with 32 fast threads or more run Larger BA units

[7im]

Better to watch the long term trends.

Please tell me what you see ? GPU (Core 17) > CPU/SMP ?
Any GPU-BA ? What requirements ... If I buy now GPU with 3GB will that be enough ? Need to be 3.5 GB ? I know, memory is not of big concern for folding, but again who knows what changes coming up.

I really think we should see some roadmap from PG on what's in the pipeline. How many Core 17 projects we can welcome in 2014ff. Even if some projects get stopped on the way, that's fine. But for me as regular donor I don't have any basics for planning.

PG has lot's of things in their pipeline, but not all of them come to be released. They also don't give ETAs because they always change, so they are always inaccurate, so people always complain about it, so better not to disappoint. They give general info when appropriate, and then specific info only when close to release. Again, that way the info is more accurate, and less controversial.

If you had to buy hardware today, you go by today's standard. For a few thousand dollars, you can build either a 4P System and run BigAdv, or a 4 GPU system, and pull down around 800K PPD with either one.

6 months from now, obviously the roadmap will change. BA requirements go up. New GPUs will come along. CUDA JIT might come along, or not. New CPUs will come along with even more cores per chip. None of their performance will go down, but only go up. Longer term, look at how CPU performance gains are starting to plateau while GPU perform is on a steeper curve up.

A year from now is anyone's guess. The roadmap changes too quickly to predict anything accurately, because the hardware and software changes too quickly. That's simply a function of technological innovation, not anything FAH can control. FAH might build it's own Live Linux CD. FAH might be taken over by the CureCoin movement. AVX support is inevitable, if not sooner than surely later. SMP for GPU? BA for GPU? All possible, if there is a demand for it, if it helps the researchers get answers faster, they will persue it and develop it. If not, then no. Also note that researchers at PG come and go every few years, after their area of study is completed. So that changes the computing landscape as well. New fahcores come and go along with a specific area of study... Tinker, GromacsB, QMD, GPU1, Amber, SMP1, GPU2, etc.

Whether you purchase today, next week, or next year. Getting the fastest CPU with the most cores that you can afford, and the fastest GPU that you can afford, and that will always be the best choice on that day. CPUs and GPUs will be folding a long time.

[bruce]

Please tell me what you see ? GPU (Core 17) > CPU/SMP ?

The trends here are mostly dependent on hardware manufacturing trends. GPU manufacturers will continue to develop faster chips and enthusiasts will continue to buy them. CPU manufacturers will exploit new technologies that do more processing with less electricity. Motherboard manufacturers will continue to meet the needs that their customers present to them. ... and FAH will make use of all those trends as they develop.

Phones and tablets will continue to provide products with longer battery life and probably use chips which have limited processing power so even though there are lots and lots of them, they're probably not going to be a fruitful platform for FAH which needs lots of scientific processing power over long periods of time.

Any GPU-BA ? What requirements ... If I buy now GPU with 3GB will that be enough ? Need to be 3.5 GB ? I know, memory is not of big concern for folding, but again who knows what changes coming up.

That's difficult to predict. Traditionally, FAH has not needed much GPU memory -- just GFLOPS, which pretty much depends only uses the speed and quantity of the shaders.

During early testing of Core_17, there was some thought of harnessing multiple GPUs together to work on a single problem but two or more GPUs working together were reportedly significantly less productive than two or more GPUs working on separate problems. It seemed like a reasonably good idea but not using the technology that the public is buying. GPUs can work well together when creating beautiful screen images, not when trying to exchange data with remote shaders.

If my crystal ball isn't working well and something else happens, then (oh well) I was wrong.

I really think we should see some roadmap from PG on what's in the pipeline. How many Core 17 projects we can welcome in 2014ff. Even if some projects get stopped on the way, that's fine. But for me as regular donor I don't have any basics for planning.

As I said earlier, the PG roadmap is dependent on the hardware manufacturers roadmap plus the consumer trends that can be exploited by FAH. FAH has always recommended you base your purchasing decisions on equipment that you'll be using for something other than FAH and they'll work hard to use whatever equipment that you (massively plural) buy. If you do buy just for FAH, be aware of what the rest of the world will be buying for non-FAH activities and recognize the risks involved.

[bobc36]

Dr. Kasson,

Thank you for the announcement today. I really appreciate hearing from your group from time to time, even if there is nothing new to report.

With that being said, I cannot help but be saddened by this recent announcement. Due to the ability of many server grade computers being able to achieve the current deadlines, I am wondering why there will be a change in the requirements for -bigadv folding.

Are they deadlines for the current projects going to be tightened?
Are new, larger project going to be released?

I remember the core requirement increase 2011 when the core count restriction was changed and the outcry that followed. I understand the reason for that change was to eliminate the single chip -bigadv users as the client was never designed for them. ( I myself had a computer that was built for this sole purpose.) This time, however, I don't see the rationale behind the core increase.

What I'm trying to get at is I feel that I (and others) need more information about the roadmap of the project so that I (we) can plan our future donations.

Thank you for your time,

Erik (bobc36 of folding@evga)

[Nathan_P]

What we would appreciate more than a generic 24core/32 core announcement is an announcement on what will happen to preferred deadlines, that is what governs whether or not a system is BA capable

[powerarmour]

I'm just hoping it'll be 32 threads and not 32 cores, would be pretty sad if my dual Xeon E5-2650's couldn't make the deadlines.

[orion]

I'm just hoping it'll be 32 threads and not 32 cores, would be pretty sad if my dual Xeon E5-2650's couldn't make the deadlines.

It will be 32 threads, don't worry.

[Grandpa_01]

I'm just hoping it'll be 32 threads and not 32 cores, would be pretty sad if my dual Xeon E5-2650's couldn't make the deadlines.

You said the magic word (DEADLINE) we all know core count does not matter when it comes to speed of return. As Kasson said they currently do not have any better way to determine what is capable of running bigadv WU's. By the way fah has no way to determine the difference between a core count and thread count a 8 core 16 thread CPU is seen as a 16 core CPU.

A problem I do see though is that most of the rigs they are eliminating will not go to smp they will just start collecting dust in some obscure corner of a persons house because smp does not pay well enough to justify running them so I have a feeling this may backfire but I may be wrong.

Hopefully Peter will elaborate a little on the deadlines so we will know what wiil make it and what will not make it.

[PantherX]

Please note that two similar topics were merged together into a single one. Luckily, the order of posts from both threads are preserved.

[Viper97]

I'm just hoping it'll be 32 threads and not 32 cores, would be pretty sad if my dual Xeon E5-2650's couldn't make the deadlines.

You said the magic word (DEADLINE) we all know core count does not matter when it comes to speed of return. As Kasson said they currently do not have any better way to determine what is capable of running bigadv WU's. By the way fah has no way to determine the difference between a core count and thread count a 8 core 16 thread CPU is seen as a 16 core CPU.

A problem I do see though is that most of the rigs they are eliminating will not go to smp they will just start collecting dust in some obscure corner of a persons house because smp does not pay well enough to justify running them so I have a feeling this may backfire but I may be wrong.

Hopefully Peter will elaborate a little on the deadlines so we will know what wiil make it and what will not make it.

I think you are spot on here... smp will not receive any more attention other than what is out there now with the other folders.

What I do see is a move away from FAH into other DC projects as this is just a wrong decision all the way around and alienates those seeking to help.

I agree this will backfire completely. However, I'm just one voice.