Intel is finally talking details about the first Broadwell-Y, now known as Core M for no particular reason. SemiAccurate told you a lot of the low-level details weeks ago, now it is time for launch #2.
What Intel is talking about this time around is not much, mostly self-congratulatory marketing terms for items that are detrimental to the owner. Of the 27 slides that were presented, three had information worth repeating, a die shot slide, a SKU table with partial info, and a curiously labelled block diagram. Due to time constraints, only some of the questions we asked could be answered, our bad, but all three important pics are below.
Note how much of the die is GPU…
There are three SKUs that are labeled in the usual Intel clear and explanatory fashion, 5Y70, 5Y10a, and 5Y10. Sorry to have to go over what each is, but some readers won’t figure it out from the names. They are actually all the same part and all cost $281 in 1000 unit quantities, quite reasonable for a chip targeting $10-30 tablet and phone SoCs. The main difference is that 5Y70 has, you guessed it, vPro and enterprise support, the 5Y10 twins don’t. These two are actually the part but the A is specced for 4.5W while the non-A is a 4W part.
All three SKUs, barely three but technically three
Now things get curious, all the parts cost the same price, and the top two are the same wattage but radically different speeds. The ’70 has more features, runs 300MHz faster base, 600MHz faster turbo, and has a 50Mhz faster GPU for the same TDP. Umm… sorry, can’t explain that one. Neither can I explain the difference between the 5Y10’s, one uses .5W less but performs exactly the same for the same price. You should buy the higher TDP version because… because… because all of a sudden Intel’s naming and fusing schemes makes absolute sense, at least comparatively.
The GPU that Intel is calling HD Graphics 5300 is the GT2 version sporting 24EUs. There is a higher and lower EU count GT1 and GT3 coming but those won’t be -Y/Core M level products. They support DX 11.2 and OpenGL 4.3, but no Crystalwell/GT3e type cache until the real models launch. The RAM types listed are DDR3L and LPDDR3, both at 1600MHz but with very different features. The rest of the specs not listed in the slides were specifically not disclosed until after launch, we asked and Intel said no.
A few more bullet points in some slides that may be of interest, wake on voice feature and the audio DSP are somewhat related. The wake part is done by an external hardware chunk which also doubles as a DSP. This allows Intel to do a few neat audio tricks while lowering CPU load, a nice touch. If you were waiting for a massive leap when then news of a DSP leaked, this is what you get. Trueaudio it is not.
Windows InstantGo is the new marketing term for connected standby, a nightmarish security hole that can brick machines hit by malware. If you are at all sane you will disable this in BIOS before you first boot, if they allow you to. Really, it is that bad but Microsoft wants it and Intel doesn’t have the spine to say no, your loss. Unless you write malware, then this is your golden ticket, enjoy. Platform Trust Gen 2 adds SPI flash monotonic counter protection, TPM 2.0, and has battery removal features too but it won’t do squat to help.
More questions than answer in this block diagram
Then there is the block diagram above, the technical heart of the presentation. We know it is a barely adequate glossing over of things but it is the best they gave us. We asked a bunch of questions that Intel honestly didn’t have time to answer before the launch. There were also a few that should have been trivial to respond to but weren’t answered.
One of those is the link that says 4x PCIe on the top right. Remember what we said about Broadwell cutting GPUs out of the picture, here, here, and here? Remember how people laughed and said we were insane for suggesting such a thing in December of 2012? Guess what? We asked Intel what PCIe level those lanes were, they didn’t answer. We asked what chip those PCIe lanes came from, CPU or chipset, they didn’t answer. While this could be because of time, both it and the refusal to give the specs to us are almost assuredly because they don’t want people to realize how crippled Broadwell is, and not just the -Y SKUs.
Possibly related but more likely really due to time constraints, Intel would also not say what the GLCI was or where the PCIe2.0 lanes it offers comes from. We don’t even know what the acronym means. Questions about what ME10 brings to the table over its predecessors was not answered either but there is unlikely to be much here. ME7 was Sandy, ME8 was Ivy, and ME9 was probably Haswell and now we are +1 better. Yay? C-Link is also a black hole for the moment.
In the end the triumphant second launch of the delayed and crippled Broadwell brings almost nothing new to the table. It is almost assured that Broadwell is indeed cutting external GPUs out of the market in the way SemiAccurate exclusively revealed almost two years ago but we can’t be certain until after publication. Please note that we do see the humor in light of the technical downgrade, ’tis to laugh but the joke is on Intel now. The rest is forgettable or won’t be revealed for at least 17 more launch events. Stay tuned, it will be exiting. Not.S|A
Latest posts by Charlie Demerjian (see all)
- More on Intel’s 10nm process problems - Sep 17, 2018
- Intel puts out another 14nm 2020 server platform - Sep 11, 2018
- Why Can’t Intel Supply Enough 14nm Xeons? - Sep 10, 2018
- Intel can’t supply 14nm Xeons, HPE directly recommends AMD Epyc - Sep 7, 2018
- AMD reintroduces the Athlon name with two CPUs - Sep 6, 2018