Intel OSTC's kernel patches reveal initial details about Skylake

[meloz]

Intel's Open-Source Technology Center pushed some changes towards the kernel. From these patches Larabel has compiled following information:

The initially added Skylake PCI IDs / product support come in several groups and include:

Skylake ULT GT1: 0x1906
Skylake ULT GT2: 0x1916
Skylake ULT GT2F: 0x1921
Skylake ULT GT3: 0x1926
Skylake ULX GT1: 0x190E
Skylake ULX GT2: 0x191E
Skylake DT GT1: 0x1902
Skylake DT GT2: 0x1912
Skylake Halo GT1: 0x190B
Skylake Halo GT2: 0x191B
Skylake Halo GT3: 0x192B
Skylake SRV GT1: 0x190A
Skylake SRV GT2: 0x191A
Skylake SRV GT3: 0x192A
Skylake WKS GT2: 0x191D

ULT, ULX and DT are self-explanatory. But what about Halo? How would a Halo GT1 differ from DT GT1? SRV is presumably reserved for the small-socket Xeons? WKS is a weird one.

Bit crazy how many variant of iGPU they will have, when combined with different CPUs will result in large number of SKU. SKU inflation and segmentation continues unabated at Intel.

Another thing that stood out:

The HD Graphics Gen9 for SKL/Skylake has a third display plane compared to Haswell/Broadwell and features five HDMI/DP/eDP display ports.

Yes, at last! I hope motherboard manufacturers give us at least one decent motherboard with three DP or HDMI outputs.

You can read the entire article at Phoronix.

 

[Khato]

Only guess I have as to what the difference between DT and Halo might be is with respect to the CPU core config. Which I'd like to hope would be a difference between quad and hex core, but it's also quite possible that it'd be dual vs quad. Regardless, interesting find.

 

[erunion]

Only guess I have as to what the difference between DT and Halo might be is with respect to the CPU core config. Which I'd like to hope would be a difference between quad and hex core, but it's also quite possible that it'd be dual vs quad. Regardless, interesting find.

maybe. The lack of different core counts is strange. But if Halo does mean the quad/hex dies, does it make sense to make a gt1 version?

My thinking is that Halo is the successor to devil's canyon and unlocked broadwell Iris Pro.(and maybe halo gt1 is an unlocked Pentium)

 

[IntelUser2000]

ULT, ULX and DT are self-explanatory. But what about Halo? How would a Halo GT1 differ from DT GT1? SRV is presumably reserved for the small-socket Xeons? WKS is a weird one.

WKS/SRV: Workstation/Server. Why is that weird? They already sell iGPU enabled Xeons to go against entry-level workstation/server graphics cards.

Skylake Halo GT1:
Skylake Halo GT2:
Skylake Halo GT3:

Halo is probably the H series and the 25W "U" parts.

 

[erunion]

Halo is probably the H series and the 25W "U" parts.

oh yeah! The H series will definitely have gt3, I think you nailed it.

 

[meloz]

WKS/SRV: Workstation/Server. Why is that weird? They already sell iGPU enabled Xeons to go against entry-level workstation/server graphics cards.

Why do they need to designate it WKS and SRV, though? Mere SRV or WKS would have been enough to differently identify all the SKUs that will end up as baby Xeons. Unless there is a core count difference between SRV and WKS...

Halo is probably the H series and the 25W "U" parts.

So 25 Watt TDP parts get GT3 graphics, but not DT? Won't this create a situation where a GT2 of higher TDP SKU will outperform GT3 of restricted TDP cousin?

I suspect Halo could be unlocked -or at least somewhat overclockable- CPUs. That's what I want to believe, anyway.

Anyway, this is the third data point that Skylake is on schedule and we should see it launch in June / July 2015 at the latest.

 

[IntelUser2000]

Why do they need to designate it WKS and SRV, though? Mere SRV or WKS would have been enough to differently identify all the SKUs that will end up as baby Xeons. Unless there is a core count difference between SRV and WKS...

I don't know. But then again Broadwell doesn't seem to be getting "GT4" part that was in earlier kernel patches either. The easiest way to compare is to see if there is an equivalent for Haswell. Then you can see what WKS and SRV really are.

So 25 Watt TDP parts get GT3 graphics, but not DT? Won't this create a situation where a GT2 of higher TDP SKU will outperform GT3 of restricted TDP cousin?

Actually for Haswell GT3 28W parts perform about on par with GT2 Desktop ones, while GT3e 45W parts outperform GT2 Desktop significantly.

GT3 isn't purely about performance, but performance/watt too. And being in a position to make much money as possible, better perf/watt with similar specs otherwise equals more money for them.

 

[cbn]

The HD Graphics Gen9 for SKL/Skylake has a third display plane compared to Haswell/Broadwell and features five HDMI/DP/eDP display ports.

:thumbsup:

 

[witeken]

Anyway, this is the third data point that Skylake is on schedule and we should see it launch in June / July 2015 at the latest.

How do you come to that conclusion when Intel stated H2'15, which starts in July and ends at the end of December.

 

[witeken]

But then again Broadwell doesn't seem to be getting "GT4" part that was in earlier kernel patches either.

The GT4 part you're referring to is Skylake, not Broadwell.