8 ms·
Intel’s 6th Generation Skylake Processors Scheduled for 2H 2015
- vardump 12y agoWarning: link auto-plays video with sound. Edit/addition: AVX-512 in Skylake is pretty exciting. 32 FLOPS per clock cycle per core is amazing. Also good to see 64 or 128 MB of eDRAM being included in more configurations. It should also somewhat help with algorithms with larger working sets, not just with graphics. Integration is also interesting, latency between GPU and CPU should be low. This may open chances for better integration and sharing of work between those units. 1 Gbps ethernet starts to be a limitation, it's barely enough for current internet connections in some parts of the world. Nearly any SSD can do sustained 5 Gbps, most new mechanical disks push 2 Gbps. It's time for 10GBASE-T to be a standard, with power saving option to function at lower speeds. Nowadays even low end hardware such as $80 Asus etc. wifi-router with USB 3.0 can push 500 Mbps+ at file serving.
- bhouston 12y agoIt is hard to get mainstream usage of CPU extensions until they are on the large majority of chips because creating multiple code paths is a pain. Thus while AVX-512 is nice, I do not expect to see usage of it outside video codecs until at least 3 years from now, and it likely won't be widely used until 5-7 years from now. This problem is compounded by the slow upgrade cycles of PCs now that their performance doesn't drastically increase per generation.
- vardump 12y agoIt's worse than that. It's still problematic to even assume SSE4.2, not to mention AVX2, which would help significantly. Virtualization is making the problem worse, because hypervisors set CPUID bits to lowest common denominator within a vMotion / teleportation domain. I think (specialized) JITting is the answer for that. Generate code on the fly. I've been lately playing around with JITted [de]serializers, etc. I think it might be possible to have an order of magnitude gains in JSON, XML, msgpack, etc. processing. Even *printf can be much, much faster. If the format specifier is static, first run just generates code just for that specifier. Subsequent invocations can be about 10-20x faster versus normal style format string scanner. Edit: When I say JIT, I'm not talking about JVM or any other mainstream implementation specifically, but the concept in general. Which means just-in-time generation of (optimal) machine specific code.
- bhouston 12y agoI guess the other solution is to use OpenCL and write all kernels in that language. Then it should JIT into AVX-512 if it is available.
- vardump 12y agoIt might be acceptable for some subset of "embarrassingly parallel" problems. For other problems, writing OpenCL code that even approaches hardware potential is very hard. We might need other intermediary languages than it. Currently LLVM is a better fit than OpenCL in general. If you want the last 50% of performance, hand coding seems to be unfortunately the only way. Very little code needs to be that fast, though.
- tormeh 12y agoJIT and AOT on-device is the future, I agree. How good are the JVMs at taking advantage of new instructions?
- bhouston 12y agoMy understanding is that JVMs do not generate any advanced instructions from standard Java code. The only time they use these instructions is if they link in hand coded C++ libraries that make use of them. Edit: Stackoverflow answer says that in rare cases the JVM can vectorize simple loops: http://stackoverflow.com/questions/10784951/do-any-jvms-jit-compilers-generate-code-that-uses-vectorized-floating-point-ins http://stackoverflow.com/questions/10784951/do-any-jvms-jit-...
- vardump 12y agoJust take a look at the code they generate, it's easy. In short, they do generate SIMD code, but very rarely take advantage of vectorized execution.
- tormeh 12y agoWhy aren't the processor manufacturers making their own JITs and AOT compilers? Surely that would be a competitive advantage.
- nly 12y ago> creating multiple code paths is a pain GCC added some cool functionality to make this a lot easier recently: https://gcc.gnu.org/wiki/FunctionMultiVersioning https://gcc.gnu.org/wiki/FunctionMultiVersioning
- innocenat 12y agoNo it isn't. You are still going to need to maintain multiple code path. Especially with SIMD instruction set like AVX-512 -- to be performant you need to do things with direct assembly anyway.
- Keyframe 12y agoSeems nice, but isn't all that much different from checking support yourself and initializing function pointers to correct functions - more work, but portable.
- cmsj 12y ago+1 on the networking angle, although Intel does have Thunderbolt, which at least gets you 10Gb peer-to-peer networking if your hardware and OS support it. It does seem very weird that 10Gb copper networking has existed since 2006, but hasn't fallen down to prices so low that it naturally replaces gigE. It seems entirely possible at this point, that 802.11ax will be finished before 10GBASE-T becomes widespread outside of server applications.
- bhouston 12y agoI as well desire 10GBase-T that is affordable. It seems to be very slow coming. I think the issue is that while there is some business desire for this, the consumer only wants wireless -- thus that is where all the effort is going. Probably to get past 1GB in the next few years, one will deploy a wireless network (although wireless suffers horrible congestion as opposed to switched networks.)
- bhouston 12y agoIt is hard to get excited about this update if the core count stays the same (2/4) in the low end chips. The CPU throughput per core does not change that much per generation now: 2010-era Intel Core i7 980X (6 cores with HT) has a performance rating of ~9000: https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7+X+980+%40+3.33GHz&id=866 https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7+X+980... 2014-era Intel Core i7 5930K (6 cores with HT) has a performance rating of ~13500: https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-5930K+%40+3.50GHz&id=2336 https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-5930K... That works out to an CPU performance increase of ~1000/10% per year/generation. I guess that is impressive given no clock speed increase, and reduced power consumption, but that doesn't turn into more powerful results for consumers in terms of raw performance. If Intel does not increase the general core count available to consumers, we are basically stuck at the same level of performance we were 4 years ago -- this isn't the case with GPUs nor with mobile CPUs, which are increasing rapidly.
- vardump 12y agoCompilers are not quite up to the task, I think. I think those performance ratings are unfair in comparison to true potential between those CPUs. I think the latter is 2-3x faster given optimal instruction stream in compute heavy workloads. Branchy or code with random memory access patterns is likely to have nearly no performance gain.
- phaemon 12y agoIsn't the 2010-era Intel Core i7 980X about twice the price of the 2014-era Intel Core i7 5930K? In other words, aren't you comparing a high end CPU from 2010 to a medium range CPU from 2014?
- quarterto 12y agoA better comparison would be the 5960X, which scores ~16000: https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-5960X+%40+3.00GHz&id=2332 https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-5960X... There's no way you could call the $600 5930K "medium range" though :)
- 12y ago
- higherpurpose 12y agoJust like Broadwell was scheduled for the first half of 2014?
- sandGorgon 12y agoI dont know why you are being downvoted, but it is a valid questions. Broadwell was delayed for half a year and is still not widely available. Macbooks (which are a pretty important predictor of demand) are not going to have Broadwells till 2015 [1], so the question is if Intel is going to have Broadwell at all or skip directly to SkyLake. Plus, Broadwell was delayed because of manufacturing yield issues... so unless that is cleared up, SkyLake will be similarly affected. [1] http://www.macrumors.com/2014/07/09/broadwell-early-to-mid-2015/ http://www.macrumors.com/2014/07/09/broadwell-early-to-mid-2...
- gsnedders 12y agoAs I understand it, the issues are fundamental to the 14nm process — and one might expect once they can manufacture 14nm Broadwell parts with reasonable yields, they will be able to manufacture 14nm Skylake parts with reasonable yields. Last I heard the plan was for Broadwell to be available for something like six months only.
- TillE 12y agoThey announced a while ago that Broadwell's delay will not affect the scheduling of Skylake. They both use the same 14nm manufacturing process, so if Intel has fixed the issues with Broadwell, everything makes sense.
- femto 12y agoSome interesting numbers: In 1971, the 4004 microprocessor had 2300 transistors fabricated with a 10um process. If the 4004 was fabricated with the same 14nm process as the Skylake, over two hundred 4004s would fit in the area that a single transistor occupied in the original 4004. "Breakeven" occurred with the 68000: In 1979, the 68000 microprocessor had 68000 transistors fabricated with a 3.5um process. If the 68000 was fabricated with the same 14nm process as the Skylake, an entire 68000 would fit in the area that a single transistor occupied in the original 68000. Today's transistors are really small compared to what they used to be! The mind boggles at how many simple processor cores could be crammed into a current die. The numbers came from: https://en.wikipedia.org/wiki/Transistor_count https://en.wikipedia.org/wiki/Transistor_count
- swalsh 12y agoWhat blows my mind is that Intel continues to do it with what seems like really only one real competitor.
- higherpurpose 12y agoARM chip makers are competitors to Intel now. Whether you believe it or not, is less important. Intel, however, does believe it. Their entire chip strategy over the past few years has been a response to ARM - with each new generation they barely increase the performance of mainstream chips or even drop it, like they did with Core-M, as they try to lower their power consumption, to put them closer to ARM's range. I'm not talking about IPC here, but about the overall performance of the mainstream chips they're pushing in the market. First it was -M chips in laptops (30W-level performance), then it was -U chips (15W-level performance), and now it's -Y chips (5W - or so they say - level performance). Yes, a good chunk of that difference is from moving to new nodes and optimization, but not all of it. Another big part comes from lowering the performance and/or throttling their chips a lot more aggressively than before in order to achieve that drastic drop in power consumption. They don't care what AMD is doing these days, but they do care what ARM chip makers are doing. They even bought a 4G modem company, so they can integrate modems into chips the way Qualcomm does it. Speaking of Qualcomm, I believe it has just announced it's going into server chips now. Intel feels so threatened by ARM now, that it's losing $1 billion per quarter, to essentially give away its Atom chips for free to no-name tablet makers, or less important OEMs (who are the only ones desperate enough to take Intel's deal, because others know that as soon as Intel becomes a big player, they'll charge a much higher rate even than other ARM competition - Intel likes big profits on its chips and it's the only way for them to sustain the company). They even went into the business of IP licensing for crying out loud, just like ARM (they're licensing Atom IP to Rockchip and Spreadtrum). Intel has essentially admitted that it can't be competitive building its own mobile chips. They have to license the IP to others so other companies can build them - on TSMC's foundry, on an old process node, even by current ARM standards.