Folding Forum

I was reading up on Sony's PS4 and MS's Xbox 720, and it appears that the hardware of both platforms are quite similar, but also very powerful.

For example, from http://blogs.amd.com/fusion/2013/02/21/amd/, which is cited by the PS4 Wikipedia article

AMD “Jaguar” CPU cores with next-generation AMD Radeon™ graphics delivering nearly 2 TFLOPS of compute performance! This unique APU architecture enables game developers to easily harness the power of parallel processing to fundamentally change the console gaming experience. Not only creating the opportunity for new possibilities in software design, but also faster and more fluid graphics.

The graphics processing unit (GPU) consists of 18 compute units to produce a theoretical peak performance of 1.84 TFLOP/s. This computational power can be used for graphics, compute-heavy workloads, realistic physics simulation, or a combination of the three.

I don't know if "now that we can do just about any calculation on any piece of hardware" statement in Dr. Pande's recent blog can be taken literally, but the PG does use abstraction libraries like OpenMM to make porting much easier. GPUs in PCs obviously have high-performance capabilities, and it seems like the PS4's hardware is the same way. From the specs, it seems like there's a lot of computational performance that could be had. FAH already runs on Radeons.

From http://appleinsider.com/articles/13/02/ ... pabilities

Also pushed was the "immediacy" built into the PS4, including background downloading, instant play of new purchases and social network sharing.

"social network sharing" buttons have been added to folding.stanford.edu, the stats pages, the blog post, and Web Control. Perhaps the PG could take advantage of the PS4's sharing capabilities as well.

The PlayStation 4 will feature a chip based on the x86 architecture, on-board HDD storage, and 8GB of GDDR5 RAM.

It seems that the difficult but powerful Cell processor has been replaced by a 64-bit x86 AMD one, which makes that aspect very similar to a PC. The RAM is also exceptionally fast, and the PS4 uses unified memory, so there's no delay in moving data from RAM to the GPU. This latter aspect may not be as important for general-purpose work, but it seems like it would still be an advantage.

The Cell Processor in the PS3, according to many developers, was powerful but difficult to develop for, which led to the Xbox 360 becoming something of a developer standard. Consequently, many games developed for both systems were unable to take full advantage of the PS3's capabilities, and many games ported from the Xbox 360 to the PS3 underperformed versus their Xbox counterparts.

Again, both platforms are now much easier to develop for.

Back in 2007, FAH released its client for the PS3. Considering the PS3's radical architecture which made it exceptionally powerful at the time, it was quite impressive that they managed to port their MD simulation code to it. Many users were asking why FAH wasn't on the Xbox 360, and the answer that was always given was "it simply doesn't have the same power". This answer was on the FAQs for a long time and still is in some places. The PS3 simply had power that was unmatched by the PC or any other console. Considering the PS4's specs, it seems like a much easier to develop for, and if the PG were interested in doing that, the process should be much easier this time around.

Not much is known about the Xbox 720 at this time, mostly because Microsoft has not announced things. Still, some bits and pieces floating around seem promising:
http://www.expertreviews.co.uk/games/12 ... nd-rumours

Rumours of a 7000-series AMD GPU would certainly live up to this, and also place the Xbox 720 and PS4 on a very close footing in terms of graphical oomph, something that publishers and developers will appreciate. ... Original rumours had this being developed by IBM, using PowerPC cores, but it now seems as though AMD has stepped in and will be using x86 architecture. Sebastien has also stated that the Xbox 720 will use x86 hardware. "Our checks suggest that next-generation console hardware will be largely built from 'off the shelf' high-end PC components", he wrote.

So it seems that both the 720 and the PS4 are using hardware already in place in PCs, and that both have similar computational capabilities.

From http://bostinno.com/2013/03/04/xbox-720 ... 1816_0__ss

8 CPU cores at 1.6GHz, with a custom 800MHz graphics processor in the GPU. There are 8GB of RAM on board, with 32MB of fast embedded SRAM at double the throughput speed, and a 50GB-capable 6x Blu-ray drive.

They also show an architecture diagram that seems very analogous to a PC. The 8-core CPU is clearly capable of some good parallel computations, and if the 720's CPU uses out-of-order and branch-prediction optimizations as the PS4 supposedly does, then it may have some serious computational power behind it.

I'm sorry for the really long post, but I want to put some facts and figures out there for consideration. The gaming situation seems to have changed from 2007. Both the PS4 and the 720 appear to have a similar computational performance, and both use architectures that are analogous to the well-known PC's. My point is that if the PG has an interest and the resources, both of these platforms seem like a really good idea. In other threads people have asked about other platforms, like the Raspberry Pi, but those platforms are limited by a small market, battery power, and lack of serious computational capabilities. From the information above, it appears that neither the PS4 and the 720 have this problem. As Bruce implied here, there may be business objectives associated with the client, but I really think the PS4 and 720 are likely worth pursuing if FAH wants even more cutting-edge computational power. I suppose the FAH client for these platforms could just be a regular program like any other game, but that's a decision to be made further down the road I suppose.

Jesse_V wrote:Many users were asking why FAH wasn't on the Xbox 360, and the answer that was always given was "it simply doesn't have the same power".

The other, and perhaps more important, answer was that Microsoft wouldn't agree to help create the client/core. Considering the truly awful overheating issues the early models had, I don't think they were particularly motivated to have something running the hardware near 100% for extended periods of time.

I believe PG would have to get permission from either vendor in order to distribute a client, but lack of vendor support on coding the client may not be a deal-breaker this generation.

First, a bit of history:

The PS3 client could not have been developed without a lot of help from Sony. They were willing to invest in the development effort to help spur sales of the PS3. In comparison, the XBox was significantly under-powered and according to the benchmark sites would not have made a powerful enough geometry engine although they did make up some of that by the number of clients that might have been available.

It's difficult to predict the future. Stanford probably could develop reasonable clients for either/both of them with the right kind of support from Sony/Microsoft but without their strong support I don't think it will happen (though I'd be happy to be wrong). At the present time, Stanford currently has quite a number of development efforts underway and they'd have to take development resources away from other important things. From the information that you and I know, it's very hard to guess how many new clients would actually fold on either platform, compared to some other equal investment in something else.

It does appear that both may use off-the-shelf components which would facilitate hardware support. The next question is the OS. The OS on the PS3 was VERY tightly controlled by Sony, making it impossible to port something like FAH or OpenMM to it using the game developer toolkit, which meant that only Sony could develop the PS3 client. Will that be true for the PS4/Xbox 720 or will OS be open enough for easy development at a very basic level?

Another shorcoming was that FAH really couldn't run as a background task in the PS3. The PS4 is adding some capabilities to download in the background but it's not clear that those capabilities will similar to modern PC operating systems or more like trying to do two things at once on MSDOS. As a general rule, nobody wants to be able to play two games simultaneously so the OS doesn't necessarily need to support true multitasking.

At the present time, FAH is reducing the number of different platforms they support, not increasing that number. Instead of putting additional support into unique FahCores for NVidia and for ATI/AMD GPUs, they're moving toward a common FahCore that supports only OpenCL, leaving the responsibility of providing good OpenCL support to those who write drivers for GPUs. That's not a good sign for expanding into new platforms but anticipating the rapid growth in the number of them sold is a positive factor. Only time will tell.

bruce wrote:From the information that you and I know, it's very hard to guess how many new clients would actually fold on either platform, compared to some other equal investment in something else.

The cost-benefit benenfit ratios are obviously important to consider. We both know that they are working on finishing V7 and moving their GPU cores towards OpenCL, but once they are "done" with those, then what? Gamers pay more attention to hardware/software specs and I believe that they are much more likely to be interested in downloading F@h if they hear about it than the newcomer using a PC. So then at least part of the question becomes: will there be enough PS4/720 consoles out there to justify the effort?

The estimate for Jan 2013 is that 77 million PS3s have been sold worldwide. According to this post by Sony, more than 15 million PS3 users have participated in F@h. Assuming that those correspond to unique PS3s, dividing out these numbers gives 19.4% (This seems high to me, but even if you attempt to counter the assumption by dividing the numbers in half, it's still a ton of actual PS3s). As of February 11, 2013, 76 million Xbox 360s have been sold worldwide. As you said, it's difficult to predict the future, but it seems very likely that both of these new platforms are going to be very successful. (Sony and MS obviously think and hope so). Even if F@h was run on even 1% of either these new platforms, (that number depends on PR) that's tapping into a huge amount of "unusued" computational power. The PS3, with it's very low-end hardware by current standards, gave 100 million hours of computation to F@h over five years and 7 months.

bruce wrote:It does appear that both may use off-the-shelf components which would facilitate hardware support. The next question is the OS. The OS on the PS3 was VERY tightly controlled by Sony, making it impossible to port something like FAH or OpenMM to it using the game developer toolkit, which meant that only Sony could develop the PS3 client. Will that be true for the PS4/Xbox 720 or will OS be open enough for easy development at a very basic level?

Question: what important services or utilities did F@h need from the OS that the Sony-provided development toolkits could not provide?

I don't know the details of the APIs either company provides for their console, but given their new graphics capabilities (have you seen those demos!) they obviously support some really advanced functions that may be useful to OpenMM, analagous to the capabilities of OpenGL 4.3 (2012) vs those from OpenGL 2.1 (2006) for rendering graphics.

since hardware info is out now
i wonder if Pande approached sony or microsoft about using FAH on the consoles

owned66 wrote:since hardware info is out now
i wonder if Pande approached sony or microsoft about using FAH on the consoles

Perhaps. https://www.youtube.com/watch?v=KbWgUO-Rqcw are the "highlights" of Xbox One Reveal 2013 event.

PS4 is supposedly more 3rd party developer friendly:
http://hardware.slashdot.org/story/13/0 ... s-xbox-one

owned66 wrote:since hardware info is out now
i wonder if Pande approached sony or microsoft about using FAH on the consoles

I really hope they do. The PS4 with it's 8GB DDR5 will be ridiculously good at folding. I will buy a few extra systems just for the purpose if they have a folding client. If anyone knows how to contact those involved in the PS3 folding project and encourage them to get a jumpstart on a PS4 version, please do so.

bdwilson wrote:The PS4 with it's 8GB DDR5 will be ridiculously good at folding.

I'm not so sure. The amount of ram isn't anything to be focusing on, as FAH is not particularly memory-hungry, nor does the memory speed usually make much of a difference. An 8-core AMD CPU isn't anything to write home about, and the GPU is roughly equivalent to a 7870, which is pretty good but not great.

I don't mean to sound like I'm ragging on the PS4 specifically, as the Xbox1 has roughly the same CPU and a weaker GPU (on par with a 7790). It's just that this generation of consoles doesn't seem like a big leap forward over PCs as the last one was. The PS3's quasi-multi-core Cell processor was a radically advanced piece of hardware at launch when true dual-core processors were just starting to be sold, while the Jaguar APU will be an average part when it comes out.

If you really mean you'd drop over $1k on PS4s, I'd say you'd be better off folding-wise investing in one or two very high-end GPUs. With the QRB, a pair of GTX 780's may significantly outproduce three PS4's and use less power to boot.

owned66 wrote:...i wonder if Pande approached sony or microsoft about using FAH on the consoles

. Whether they have or not is not known so all we can do is speculate. It's unlikely that there will be no mention of it UNLESS / UNTIL they've reached an agreement about a partnership. I'm not hopeful.

I agree with Zagen30. At the time the PS3 was released, it contained a CPU that was quite a bit more advanced than the PCs of that day. Time passed and and the PS3 was left in the dust as the PC hardware continued to improve in accordance with Moore's Law while the PS3 which made no advances whatsoever. The software development for the PS3 was expensive because it was a unique platform.

The PS4 uses contemporary hardware that's comparable to current generation PCs. With Stanford's development money currently going into a major expansion of software for contemporary GPUs, I do not feel optimistic about them investing also in a custom development effort for the PS4. The fundamental question is whether it would require a custom development or if there's a way to run software that has been developed for the PC. Sony might fund the development, but that seems unlikely, too.

Though RAM is not particularly significant, there are many, many different GPU models that have more DDR5 that FAH needs and they generally have 32- to several thousand-shaders (cores) compared to the PS4 processor with only 8, so adding even a low-end GPU to an existing PC will soon be much more productive than adding a PS4 to your farm -- and there are no additional development costs.

bruce wrote:Though RAM is not particularly significant, there are many, many different GPU models that have more DDR5 that FAH needs and they generally have 32- to several thousand-shaders (cores) compared to the PS4 processor with only 8, so adding even a low-end GPU to an existing PC will soon be much more productive than adding a PS4 to your farm -- and there are no additional development costs.

I think one of the keys is how easily they could get a GPU core running on it, as that's the growth market. Since it's based on a 7870 it in theory should be a decent performer for the near future and in theory shouldn't require a massive amount of new code, but who knows.

Open standards like OpenCL, CUDA, and AVX are much easier fruit to pick than proprietary game consoles. Don't hold your breath as MS and Sony won't be providing a DOS prompt to run fah any time soon.

50 new petaflops is nothing to sneeze about remember the ps3 client added 40,000 active users. And at roughly 2 teraflops a piece depending if they are similar to desktop GPUs(there maybe parts torn out to keep the price down). Also the shared ram could help a lot the CPU would not have to copy the data back and forth between ram.

We're still 5 months from hardware launch with a past client something might be worked out with Sony but Microsoft I think might be tough to get access to the hardware. If there were a xboxone and ps4 client it would not be to far out to think 100 petaflops total for fah. But I can only see a client for a ps4 not because better hardware I just think Microsoft scared for security for their system.

It maybe better for fah if people bought $1000 titans but with or without fah millions are going to buy a PS4/xbox. And pande group always say don't go out of your way to buy hardware by what you need for other things. Do I think people should buy multiples no way the price per preformance is close but not ovewelming and at launch GPU may have a price drop. I just hope something can be worked out for a client I don't like to things go idle.

bdwilson wrote: I will buy a few extra systems just for the purpose if they have a folding client. If anyone knows how to contact those involved in the PS3 folding project and encourage them to get a jumpstart on a PS4 version, please do so.

Sony probably isn't keen on people buying extra PS4s just for folding as they will make a big loss on those, at least until unit cost falls back to the sale price.. Although X million units looks good (especially c/w XBOX, being a very very similar product) profit comes from games sold per console.

(Merged with existing topic)

I was hoping that F@H would be on the new PS4, is that a possibility?