10 ms·
Intel Declares War on GPUs at Disputed HPC, AI Border
- vmarsy 10y agoInteresting article presenting a "war" between Intel CPU and Nvidia GPU. I'm curious where AMD falls in this war. While it's true GPU can provide really nice theoritical TFLOPS, the cost of moving data in and out of the GPU's memory is a well known issue. This renders GPU much less attractive for real applications. The reason I mentioned AMD is that I read some time ago about their Heterogeneous system architecture [1]. One of the HSA's goal is to get rid of the particular issue of moving that data. I wonder if this was adopted in any HPC cluster or anywhere else. [1] http://developer.amd.com/resources/heterogeneous-computing/what-is-heterogeneous-system-architecture-hsa/ http://developer.amd.com/resources/heterogeneous-computing/w...
- AlphaSite 10y agoI realise it's not the most reliable resource but it shows something interesting regardless. It's a server grade apu, combining a cpu and GPU onto a single socket. With HBM for huge amounts of bandwidth. They also have HIP for converting CUDA code into performance cross platform code. http://wccftech.com/amd-exascale-heterogeneous-processor-ehp-apu-32-zen-cores-hbm2/ http://wccftech.com/amd-exascale-heterogeneous-processor-ehp...
- magila 10y agoRight now AMD is fighting for its life, it's in no position to take on Intel and Nvidia in the HPC space. If their consumer Zen processors flop then AMD will likely be heading for a buyout and/or bankruptcy. Only if both Zen and Vega (their upcoming high-end GPU) are huge successes will they have the resources to mount a serious effort to capture some of the HPC market.
- rdtsc 10y ago> Right now AMD is fighting for its life, I thought so too, but looking at it stock recently it seems to be doing much better (especially in last year). Not sure how to interpret that. Wonder if it was mostly investors being happy they divested from that ARM microserver company, or there is genuine upswing and increased market share.
- photogrammetry 10y agoThe stock market can remain irrational longer than you can remain solvent.
- gscott 10y agoAMD is too important for China to let it fail.
- moyta 10y agoHuh? AMD is too important to Intel to let them fail, if AMD ever went broke, Intel would give them money just so they didn't have to deal with every country going at them for anti-trust law violations.
- erelde 10y agoI found myself thinking about that a few times these last years. Would states really sue Intel for anti-trust when it's not really their fault if there was no competition? I mean you can't expect a new CPU founder company to just open and compete directly...
- ansible 10y agoI mean you can't expect a new CPU founder company to just open and compete directly... That's exactly what the Mill CPU folks are trying. I have hopes for a RISC-V vendor to do this too.
- vidarh 10y agoThere's nothing illegal about being a monopolist. However, being a monopolist makes certain things illegal for you to do, and some of those things are things that are easy to suspect you of, but hard to prove you're not doing. As such, being a monopolist constrains you, as well as invites long, expensive, frustrating probes into your business that creates costs and risk (even if you're not intentionally violating anti-trust law, doesn't mean some business unit isn't inadvertently doing so). E.g. IBM was under an anti-trust probe from 1969 until 1982 over their mainframe business. The probe ended with the DOJ concluding the case was without merit, but it is widely considered to have had a big effect on IBMs decisionmaking for more than a decade.
- moyta 10y agoEh, AMD has been doing that and custom cores as their bread and butter, seeing as they design the chips that power all the >$100 consoles from Sony, Nintendo and Microsoft, they have a good market niche that provides them with a decent market to keep the company viable. I'm not sure of cracking a new market space would be a good idea either, just like how they aren't trying to go toe to toe with Allwinner offering H3 SOCs for $1ea with a quadcore and H.265 hardware decoding, it might not make sense to get into a 3 way war with Intel, Nvidia, etc. Retaking or getting into another market with a single vendor or none presently would likely be more profitable, just as VIA did with their point of sale motherboards with crypto accelerators and tons of serial ports. AMD only has so much money and talent, getting into a pissing contest is not something they are looking to do. Intel can piss harder and longer than them (just look at the contra-revenue they did with the HP Stream 7 and the $50 Walmart android tablet, Intel charged the manufacturer for the chip $X and paid them $X+Y for using said chip to compete with ARM).
- magila 10y agoThe console business has razor-thin margins, there's not enough money being made there to sustain a company the size of AMD. Where that business might save them is if say Microsoft decided to buy them out to secure their supply of chips for the Xbox. The fact that it would force Sony to look elsewhere for their next-gen console wouldn't hurt I'm sure.
- moyta 10y agoSure, I'm not saying its something with huge margins, just like how Allwinner isn't getting rich with their dirt cheap SOCs. That being said, they have a few markets like that, and that is what sustains a company. I highly doubt Microsoft or anyone will buy AMD, as it would not make financial sense, AMD already builds these chips cheaper than any of them could build them on their own, and any kind of merger would revoke AMD's x86 license, and Intel's x86-64 license, thus destroying the ecosystem. On another note, a razor thin margin might not be glamorous, but you can run a business like that, though it will not be glamorous.
- 10y ago
- goalieca 10y agoNintendo chose nvidia for their next generation console. Going to be arm of all things.
- formula1 10y agoThe article mentions that to write for Intels Xeon Phis, its x86. I imagine this is mostly accurate but is creating writing for this as simple as creating another thread? That seems naive Looked into it, [0] looks like you add some compiler instructures before code you want to outsource to the coprocessor. It doesnt look extreme, mostly naming which variables need to be sent and to where. Ive never worked with cuda so I dont know if its a pain or not. Looking into it, it seems instead of x86 code it is cuda prefixed functions that will offload. Granted, there are speed advantages to having direct accessbto memory, I dont think Api is the biggest selling point. Im quite interested in seeing if nvidia thinks this is a threat and how they will respond. New systems may stay with nvidia just because the experience and failures have been there. Though if the speed gains are even 25% I would bet there would be a migration [0] https://software.intel.com/sites/default/files/managed/ee/4e/intel-xeon-phi-coprocessor-quick-start-developers-guide.pdf https://software.intel.com/sites/default/files/managed/ee/4e...
- deckar01 10y ago> The basis of the Intel MIC architecture is to leverage x86 legacy by creating a x86-compatible multiprocessor architecture that can utilize existing parallelization software tools. Programming tools include OpenMP, OpenCL, Cilk/Cilk Plus and specialised versions of Intel's Fortran, C++ and math libraries. https://en.m.wikipedia.org/wiki/Xeon_Phi https://en.m.wikipedia.org/wiki/Xeon_Phi It's not threads, it's driver and compiler support for existing multiprocessing frameworks.
- frozenport 10y agoActually, there are several modes of operation. In one mode, it acts like a 60 core machine running your favorite x86 Linux applications, in another mode you can run OCL kernels on it. The former, appears to give poor performance, see: https://cug.org/proceedings/cug2015_proceedings/includes/files/tut105.pdf https://cug.org/proceedings/cug2015_proceedings/includes/fil...
- m_mueller 10y agoHere is the main problem IMO: To get reasonable performance (i.e. comparable to GPU), you need to make use of vector instructions as well. And it's still not easier to vectorize your x86 code rather than to just port it to CUDA. With CUDA you get two parallelisms done in one programming paradigm: Both vector and multiprocessor parallelism just gets mapped to the kernel scatter initialisation, and the kernel itself is just written as a scalar operations plus some index calculations. In terms of readability / ease of use it's [naive x86] > [CUDA] > [performance portable parallelized + vectorized x86]. An if you don't have the last version of x86, which almost noone has, then Xeon Phi will be just as much portation work as GPU, if not more (because Nvidia has a considerable lead in tooling support and has actually even gained more lead since Intel's Phi came out because Intel is just not very agile).
- llukas 10y ago"Second, with the above notes in mind, remember that Intel’s strength even with its own accelerator, is that there is no funky programming model—it’s CUDA versus good old X86." Now, please show us non-trivial use cases where recompiling "good old X86" to Knights Mill architecture yields satisfactory performance? To get good performance you need to ensure your code makes good use of Knights Mill vector instructions (AVX-512*).
- tlb 10y agoMost likely you call BLAS routines for matrix multiply, which use the right instructions. So any code using BLAS (which includes numpy) will run fast. Using a GPU is much more complex, because it has a separate memory space and you have to copy arrays back and forth. It's easy to lose all your performance gains if you get the array allocation wrong.
- llukas 10y agoIf you use KNL as an accelerator you got same problem as GPU. If you're using it in standalone mode then keep in mind that every part of code that is not using AVX runs on slow Atom-like CPU. You have to vectorize code yourself or find a library that did that for you. Neither of above is close to "recompile good old X86 and run".
- loeg 10y agoPresumably even on good old X86 Xeon you cared about performance and figured out how to vectorize your code, so there's no change there.
- CmdrSprinkles 10y agoThe argument is that you should be vectorizing your code anyway. You won't see as much of a boost on a skylake as you would on a KNL, but you are still going to see a pretty good boost. Obviously if your code is completely serial (and arguably bad), you have a lot of work ahead of you. Whereas, if you have been taking advantage of AVX already (or relying on libraries that do), you are basically just going to need to recompile. Are you going to get peak on a KNL? God no. But if it runs well on a "normal" processor, it is likely to run pretty well on the KNL. Which is a MUCH better foundation to start from than if you are trying to port a code to a GPU And those tweaks you make to better utilize that KNL? It will probably improve normal perf as well. It is obviously not as simple as the marketing led us to believe, but it never is. GPUs are a load of toss without a lot of work (and often some creative presentations). But the underlying principle of "Use the same code on both types of processor" does actually hold true as of KNL. As for performance numbers: Check the presentations from SC as they become available. Multiple outlets have begun to publish, or at least present, these numbers and their experiences using KNLs. I would personally argue that we still have a long way to go, but there is actually light at the end of the tunnel and it does look like there is an actual path now.
- SEJeff 10y agoI'm pretty sure Nvidia is cleaning house with Intel when you compare something like say a K80 vs a Phi. The number of people who can do amazing and wonderful things (and frameworks to support them) with an Nvidia GPU vs a Phi is also dramatically different.
- mattkrause 10y agonVidia has really hustled to make that happen. I expressed some vague interest in using CUDA for something and we were offered a demo account on one of their boxes, invited to apply for an academic hardware grant, and pointed at tons of training material (books, a MOOC, conferences, and even some 1-2 day sessions), libraries, and other resources. On the other hand, my institute apparently has some Phis. I discovered this because there's a small queue with that name on our cluster. I don't think I've ever been contacted by an Intel rep or offered training material--or even marketing material--from Intel, let alone a Phi of my own to experiment with.
- dman 10y agoIntel seems to be taking a long time getting things to market in terms of SKUs that the general developer can buy. This is critical for gaining mindshare for the programming model. Both Knights Landing and the new chips from Altera have failed to materialise into general availability in a SKU priced at <500$. Part of what has made Nvidia so successful here is that the CUDA ecosystem scales up and down their entire product line.
- moyta 10y agoThat and getting mindshare in the academic compute space. Where Supermicro is cleaning out all of Nvidia's higher end SKUs to build these academic compute clusters, Intel needs to insert itself and try to take the market. Currently they pair a mid-range Xeon with a couple GPUs, Intel needs to change that model if they want to win the market that has kept Nvidia in business.
- zamalek 10y ago> the general developer can buy Having a non-integrated GPU (i.e. NVIDIA or AMD) in a machine is not a stretch of the imagination these days. This means that any curious dabbler can grab Tensorflow and start learning ML. Over the past few months, I've been seeing exactly this on HN: many developers with ML Show HNs. In my opinion, it is precisely the general availability of ML that has caused this explosion. As you pointed out, the above scenario falls away entirely under an expensive SKU. I strongly believe that having to buy any specific SKU in order to approach this stuff will spell doom for the tech. If it's not on every SKU down to the Core 2, I really can't see it taking off.
- petra 10y agoWhy would na expensive SKU be a problem ? as long as they offered it as a service and they ROI would be significantly better, it would be fine.
- nl 10y agoBecause library developers want to have things available without paying by the hour. Once you have the software, a service is fine. But the ecosystem of tools around NVidia CUDA and CuDNN is there because you can buy a gaming card for a few hundred dollars and hack up libraries you find useful.
- nullnilvoid 10y agoNvidia's stock price has tripled from $30 to $93 in the last 12 months, while Intel's share price stayed at $34, exactly the same price a year ago. Although Intel is still a lot bigger, they want a piece of that pie too. That said, the chance for them to win a significant market share is slim. They have failed a few times in entering standalone GPU market. Will this time be different?
- moyta 10y agoNvidia has taken over a newer segment, cars, and they have a stronghold in the HPC market due to the sheer number of academic compute clusters that have based their setups on Nvidia hardware. Intel has forced Nvidia out of the mobile and lowend/midrange desktop space, eliminating what used to be their bread and butter, so this pivot by Nvidia to markets where they either control the whole stack with their custom ARM cores, or where Intel can't lock them out by getting rid of PCIE lanes like on laptops/desktops (AMD would get subbed in cause PCIE lanes are all that matter) has definitely put them in a stronger position, justifying that valuation.
- stonogo 10y agoIntel's roadmap is worthless, since they only seem to actually produce a given Xeon PHI SKU for about two years. Nobody -- and I mean nobody -- is happy with Intel's complete lack of availability for hardware replacements as the MIC addons fail. I can only hope the self-booting Knights Landing product is ... stabler. I am aware of about fifteen different sites that are enacting significant architecture changes as the 5110 supplies dry up and the 7120s are clearly next. It does not matter which product is technically superior when one of them is literally unsupportable.
- chunsj 10y agoThough I'm not that expert on this area, can Intel make H/W (FPGA) as S/W commodity by making FPGA design as open source or equivalent unlike other or existing vendors? If Intel can do this, I think in the long term, Intel can re-take their competence.
- jakozaur 10y agoIntel doesn't have that much of natural room to expand. They completely missed the mobile, Moore law is slowing down, so their edge slowly erodes. They been talking about killing GPU many years. They been doing it both technical and non-so technical point. E.g. in 2011 Intel paid $1.5Bln to NVIDIA, because the Intel killed NVIDIA chipset business through not so fair licensing practice. Instead NVIDIA invested so much in the GPU computing ecosystem CUDA. For many early years it was niche, but quickly growing. However, they hit jackpot with deep learning. I wouldn't be surprised that majority of self-driving cars will have NVIDIA chip in it.
- miloofcroton 10y agoThe only problem with Nvidia, if you look at the entire system, is that they rely on Intel's CPUs for most of the computer systems that they are a part of. Sure, they have their ARM core CPUs that have proven to be decent, but I don't see them blowing anyone out of the water with them. Furthermore, there is no interchanging GPUs, so the only people using Nvidia CPUs are the ones that buy into Nvidia's entire ecosystem, which is necessarily going to be fewer people/groups. AMD, on the other hand, while fighting Intel and Nvidia for the past decade and coming up short, may be at the horizon of a breakthrough. Only AMD seems ready to find the synergy between CPUs and GPUs. A Zen and Vega combination of architectures could essentially be an Intel i7 top of the line + an Nvidia GTX top of the line. Add some special sauce, and they could really do it. Nothing that Intel has done in a long time has been impressive, but Nvidia is on fire. AMD's CPUs may help them leap ahead instead of merely catching up (which Vega may do anyway).
- elcct 10y agoIt is like every year one reads that this time AMD beats Intel or NVIDIA, but when it comes to "showing the goods", AMD products are not much more than eccentric room heaters.
- geezerjay 10y ago> AMD products are not much more than eccentric room heaters. You can buy an "eccentric room heater" from AMD with 8 CPU cores for around 100€, which in some benchmarks runs equivalent to Intel's 900€ processors. Intel fanboys usually take jabs at AMD but they seldom back them up with evidence.
- nl 10y agoSo.. speaking as someone who does some deep learning (but quite a bit of ML more broadly). Intel are correct that the memory size on GPUs is increasingly becoming a problem in some complex deep learning models. Having access to cheap(ish) DDR4 RAM using the same programming model would be a big win. Interestingly, AMD is going after a related market (visualization) which is also very RAM hungry by building GPUs with M.2 SSDs onboard[1]. AFAIK no one has tried these for deep learning. But within the deep learning community there's a fair bit of cynicism about Intel's claims. Over the last few years they have frequently claimed parity with GPUs... any day now. Yet they keep falling further behind in any useful benchmark. [1] http://www.anandtech.com/show/10518/amd-announces-radeon-pro-ssg-fiji-with-m2-ssds-onboard http://www.anandtech.com/show/10518/amd-announces-radeon-pro...
- conjectures 10y agoMore memory would indeed be a bid deal for ML applications. Found this comment useful as it's not clear to me whether the 'Knights Landing' will actually be comparable to GPU in terms of pure grunt. Then I wonder if Nvidia will respond with more memory on their cards. From my totally uninformed perpective, that would seem easier than the transition Intel are trying to go for.
- nl 10y agoThe programming model for GPUs isn't great - since they aren't general purpose platforms you have to keep thinking about moving data around between the GPU memory and the main memory. Intel could bypass that and end up with an easier programming model. But.. believe it when you see it. Note that this is almost entirely about deep learning though. Non DL machine learning is mostly done on CPUs at the moment (with a few exceptions) and generally performs adequately. More speed is always nice, but you don't hear a lot of complaints about XGBoost's speed, or FastText, or VopalWabbit, or the various FTL regression algorithms implemented on CPUs.
- deepnotderp 10y agoYes, but deep learning is clearly the future. I'd love to fully unroll my rnns.
- wyldfire 10y ago> big plans to pull the mighty GPU down a few pegs in deep learning...will offer a 100X reduction in training time compared to GPUs...loosely laid down plans to eventually integrate...hesitant...on when "Please don't get too far into the OpenCL/SPIR/etc software ecosystem, guys. We will totally lap these GPUs Real Soon Now. And the best part is that it will all be x86!" I really wish Intel would produce a real GPU with fixed-function parts like the competition (and GDDR5 or whatever's next!). Their most attractive quality is their commitment to the open source world.
- madenine 10y ago> 100X reduction in training time compared to GPUs Sounds great, when will it.. >Intel loosely laid down plans to eventually integrate the unique Nervana architecture with Xeons this week, but was hesitant to put a year stamp on when that might happen. So their theoretical eventual integration is hypothetically 100x faster than current-gen GPUs? If they can deliver, great news for deep learning - but this seems like a lot of speculation.