Just when you thought Intel’s Kaby Lake CPUs couldn’t get any better, you are right. With the launch of many new models SemiAccurate thinks the excitement around Intel’s “new” CPUs peaked with the name leak.
If you recall we were “excited” about the launch of Kaby Lake and said it would “fix the PC market”. For those of you whose browsers didn’t render the HTML <dripping sarcasm> tag, we were joking. The consumer PC market is coughing blood on an up sales quarter and nothing Intel has on the roadmap for the next ~3 years will change that.
Officially these new Kaby Lake CPUs will change your life by adding 4K content and encoding, barely, to the PC. All the things your phone did three generations ago can now be done, sort of, on the PC, wow! If you want 4K content you need Kaby, Windows, and have your pick of any of one services, Netflix. Three more are planned for the coming year with none of the big names present. Don’t expect the DRM to still work with the next generation of Windows and Intel updates, remember what happened the last few times?
Stick with the phone in your pocket, it does the job better, cheaper, more reliably, and with orders of magnitude greater content, 4K and otherwise. There were a lot of other contrived scenarios that Intel claims make Kaby worth considering but, well, we can’t repeat them with a straight face. Suffice it to say Kaby brings nothing useful to the table for consumer.
Normally we would mock the messaging more because there is no longer any real technical information released but in a shocking turn, Intel actually did. No not die sizes or talking about packaging, you know the things that actually give Intel a competitive advantage, those are not discuss-able topics any more. But in a ray of sunshine peeping under the door, they are starting to give a little out here and there. Lets take a look at the SKUs before we get into the “new” stuff.
The SKUs in detail
Once you catch your breath at the awe-inspiring almost incremental advances Kaby brings, we have the real news. Intel gave us the GPU decoder ring for numbers to shaders etc. First off anything with Iris Plus markings are now the Crystalwell SKUs all of which appear to be 64MB variants vs the older 128MB. The good news is they are much wider spread than before, essentially anything 640 or 650 now has a usefully sized L4 cache on package. On the down side the real advances, silicon and process optimizations and packaging tech are not discuss-able topics any more so we won’t. Isn’t it ironic that the only things Intel has a competitive advantage at are forbidden to talk to the press about. *SIGH*
Moving down the stack we have the 615, 620, 630, and P630 are GT2 models. The lone GT1 level part is the 610 but it is technically a GT2 die with half the units fused off. If you recall since Skylake GT2 is the lowest physical shader count die. Speaking of shader counts GT1 is at 12, GT2 24, GT3e at 48, the GT3 non-e and GT4s seem to be MIA for the moment, expect them in a few months like Intel did with Skylake when the appropriate OEM design refreshes happen. All this info may not be groundbreaking but it is refreshing that Intel is talking about it again.
Intel is also launching the 200-series chipsets starting out with the high-end Z270 and going down from there. The two highest end SKUs, Z270 and Q270 get the biggest update with four new PCIe lanes bringing the total to 24 from the 1xx’s 20. Why the change? That is the other big news, the 200 line now supports Xpoint memory aka Optane. Intel won’t disclose anything about how this works but SemiAccurate can say that it is just 4x PCIe lanes off the southbridge to plug a SSD into along with firmware support.
That is the good news, the bad news is that Xpoint still doesn’t actually work. OK things have now progressed enough to the point where endurance may allow a carefully scripted public demo or three, but the product line is still unquestionably broken. Endurance isn’t there, full stop. Intel is announcing that the 2xx chipsets are Optane Ready, not that it will actually ship in time to matter for this generation. In SSD form factors overprovisioning may allow a few press units here or there with golden sample chips but low endurance NVRAMs used as a high bandwidth cache are a really bad idea. Avoid this one like the plague for the next year or you’ll be sorry.
More interesting is that these new Optane Ready lanes are pulled off the chipset, not the CPU. Why is this a problem? Latency and the CPU to chipset connection. This link is already overburdened by what Intel hangs off of it, adding four more lanes will only make things worse from a latency and throughput perspective. Xpoint is all about low latency and outstanding queue depths both of which are utterly hobbled by the way Intel implemented this. In other words putting your most latency sensitive I/O device on your highest latency path is just dumb. Wait for Ice for this “advance”, it might be working by then.
So in a nutshell that is what Kaby Lake brings to the table, nothing really. It is at best an incremental advance with a few new video decode modes. The Optane Ready program is nothing more than a PR promise, woe betide any who actually try to use one in the real world, it won’t end well. About the only good thing we can say about the chip is Intel is now starting, just, to talk tech again. With luck this is a trend that will continue, the rest of the outlook is pretty bleak.S|A
Latest posts by Charlie Demerjian (see all)
- SemiAccurate has Skylake-SP die shots, sizes, and more - Jul 21, 2017
- Skylake-SP has a diverged core - Jul 19, 2017
- Intel’s Purley platform architecture is a step forward - Jul 13, 2017
- Intel launches Purley aka metal Xeons - Jul 11, 2017
- AMD’s Epyc has lots of connections - Jun 26, 2017