Skylake SKU

[NTMBK]

Gah, no 2+4e part. That would have been a perfect little Steambox.

 

[III-V]

Seems to confirm the 95W TDP SKU. Also confirms removal of FIVR (I'm still baffled by that).

Source?

FIVR is definitely out. It's temporary though, coming back with Icelake (tock) on 10nm.

95W, as has been stated many times, is a standard TDP rating for pre-production samples. Ivy Bridge and Haswell both had ES samples rated for 95W, but were eventually lowered to 77W and 84W, respectively. Heck, Ivy Bridge even had such a late adjustment that some retail boxes had a 95W TDP printed on them.

 

[witeken]

95W, as has been stated many times, is a standard TDP rating for pre-production samples. Ivy Bridge and Haswell both had ES samples rated for 95W, but were eventually lowered to 77W and 84W, respectively. Heck, Ivy Bridge even had such a late adjustment that some retail boxes had a 95W TDP printed on them.

You'd think Intel knows what the TDPs will be. Apparently they don't.

 

[ShintaiDK]

95W is a standard TDP for anything above 65W on the desktop(Excluding LGA2011-x.). Whatever the final TDP will be we have to see. It can be 95W it can be 80W or something entirely different.

 

[AtenRa]

Gah, no 2+4e part. That would have been a perfect little Steambox.

Core-M 2+2 = 82mm2

Almost half of that is the iGPU, 40mm2

GT4 = 3x larger = 120mm2

Add another 40mm2 for CPU cores etc and we arrive at 160mm2. Add another 85mm2 for eDRAM and we can start to imagine how expensive this thing will be at 14nm.

Ohh, we can add another 40mm2 for another pair of CPU cores for 4+4e and die size will be close to 200mm2 without eDRAM.

I dont want to see what the price will be

 

[mikk]

Core-M 2+2 = 82mm2

Almost half of that is the iGPU, 40mm2

GT4 = 3x larger = 120mm2

Uncore wasn't a doubling in HSW GT3, only the subslice doubled. It's not necessarily 3x over GT2. Die size of Gen9 could differ as well.

Add another 85mm2 for eDRAM and we can start to imagine how expensive this thing will be at 14nm.

Desktop Iris Pro gets 64MB of edram, your 85 mm² is wrong. I would expect that the edram is manufactured in 22nm and not 14nm.

 

[witeken]

Core-M 2+2 = 82mm2

Almost half of that is the iGPU, 40mm2

GT4 = 3x larger = 120mm2

Add another 40mm2 for CPU cores etc and we arrive at 160mm2. Add another 85mm2 for eDRAM and we can start to imagine how expensive this thing will be at 14nm.

Ohh, we can add another 40mm2 for another pair of CPU cores for 4+4e and die size will be close to 200mm2 without eDRAM.

I dont want to see what the price will be

200mm² really isn't that big. But you missed 2 important points. I don't know if Intel will do it, but if you reduce eDRAM to 64MB and put it on a 14nm process, you get a 4x reduction in area.

So you'd get 82mm² + 20mm² + 2*30mm² (if you take the portion of the graphics that actually gets duplicated) = 162mm² for a totally viable APU with good performance and outstanding power that is only 30mm² more than the GT2 22nm predecessor.

 

[ShintaiDK]

I dont want to see what the price will be

And thats coming from someone who champion 244-315mm2 chips?

 

[SteveGrabowski]

Is there any good information on Skylake's usage of hyperthreading? E.g., whether the desktop Core i5's stay 4 cores without hyperthreading, or move to 4 cores with hyperthreading? It would piss me off to buy a quad core i5 right now and then in a few months it's no longer comparable to current gen i5's.

 

[ShintaiDK]

Is there any good information on Skylake's usage of hyperthreading? E.g., whether the desktop Core i5's stay 4 cores without hyperthreading, or move to 4 cores with hyperthreading? It would piss me off to buy a quad core i5 right now and then by 2016 it's effectively an i3.

Its pretty safe to say its the exact same as today.

Celeron/Pentium dualcore-HT, i3 dualcore+HT, i5 quadcore-HT, i7 quadcore+HT.

 

[SteveGrabowski]

Its pretty safe to say its the exact same as today.

Celeron/Pentium dualcore-HT, i3 dualcore+HT, i5 quadcore-HT, i7 quadcore+HT.

Is there a good source on that? It just seems like such little change since Sandy Bridge in 2011, and I'd hate to be buying into 4C/4T at the very end. Then again, I guess Intel can keep going with incremental improvements every generation on the desktop since AMD hasn't done anything worthwhile there and since mobile is where the competition is.

 

[ShintaiDK]

Look at the slides. Quadcores and dualcores is what you get.

If you need mroe cores you can get it on the LGA2011 platform.

It has absolutely zero to do with AMD. But everything to do with the software thats out there.

 

[SteveGrabowski]

Look at the slides. Quadcores and dualcores is what you get.

If you need mroe cores you can get it on the LGA2011 platform.

It has absolutely zero to do with AMD. But everything to do with the software thats out there.

I guess it's unlikely they'd do something like i5-6xxx = 3GHz 4C/8T, i7-6xxx = 4GHz 4C/8T. If I buy a Haswell Refersh i5 I really want it to still be a high end gaming CPU in 2017. Especially with the Xeon E3-1231v3 such a tempting 4C/8T CPU at $55 more than an i5-4590.

 

[tential]

Look at the slides. Quadcores and dualcores is what you get.

If you need mroe cores you can get it on the LGA2011 platform.

It has absolutely zero to do with AMD. But everything to do with the software thats out there.

yup I see no reason for intel to add more cores right now. What's the point other than enthusiasts want more cores in cheaper products. If you're an enthusiast who simply wants more cores, you can pick up LGA2011. That's what I plan to do with Skylake.

Do I need them? Absolutely not. But Skylake will be when I move to 4K gaming and I want to have my first "Premium" Intel processor.

Just to have it, no actual need of the performance in most tasks I do daily. Probably would be served fine with my Haswell processor I have now (4770k).

 

[jpiniero]

something tells me the 4+4e H part is apple only

I expect Skylake-K (and really K models going forward) to have the best graphics.

Edit: I know you mean H model but my point still stands.

 

[Enigmoid]

One thing we can say is no 35W dualcores. All 35/45W quads or 28W U dualcores.

 

[III-V]

You'd think Intel knows what the TDPs will be. Apparently they don't.

We're about a year out, and Intel's yields still have some room for improvement.

 

[AtenRa]

Ok lets have some pics,

Haswell 4+2(GT2)

Haswell GT2 alone

Now, lets see Haswell GT3

Haswell 2+3(GT3)

Haswell GT3 is exactly double of GT2

Now, lets have a look at Core-M

Core-M

And here is Broadwell Core-M GT2 alone

Now if you measure each component we have,

Core-M 2+2 = 82mm2
GT2 = ~41mm2
Single CPU Core + L3 = ~10,5mm2
System Agent, Display Engine and Memory Controller = ~12mm2
Memory I/O = ~8mm2

For 2+3 (GT3) we have
GT3 = 2x41 = ~82mm2
CPU Cores = 2x10,5 = ~21mm2
System Agent etc = ~12mm2
Memory I/O = 8mm2 + Empty space 4mm2 = ~12mm2

= Total 2+3 die size = ~127mm2

For 4+3(Quad core + GT3)
GT3 = 2x41 = ~82mm2
CPU Cores = 4x10,5 = ~42mm2
System Agent etc = ~12mm2
Memory I/O = 8mm2

= Total 4+3 die size = ~144mm2

For 4+4 (GT4)
GT3 = 3x41 = ~123mm2
CPU Cores = 4x10,5 = ~42mm2
System Agent etc = ~12mm2
Memory I/O = 8mm2 + Empty space 8mm2 = ~16mm2 *

= Total 4+4 die size = ~193mm2

Here are some personal drownings estimations, could be completely wrong but what the hell

 

[AtenRa]

And thats coming from someone who champion 244-315mm2 chips?

240mm2 at 28nm is way cheaper today than 200mm2 at 14nm

 

[witeken]

Nice calculation, so I guess it's ~165mm² for 2+4e without eDRAM, which would add just a mere 20mm² at 14nm for 64MB to increase performance quite considerably.

 

[AtenRa]

Nice calculation, so I guess it's ~165mm² for 2+4e without eDRAM, which would add just a mere 20mm² at 14nm for 64MB to increase performance quite considerably.

I believe Intel have said they will use 22nm for eDRAM in Broadwell.

Now if they will only use 64MB for Broadwells eDRAM, it will be close to ~40mm2

 

[SAAA]

To anyone with knowledge, looking at AtenRa's 2+3 guess, is such an oblong die reasonable? Are there any problems in manufacturing a thin and long die?
For heat dissipation it makes sense anyway, spreading apart the components rather than packing them in a square thing.

 

[SAAA]

I believe Intel have said they will use 22nm for eDRAM in Broadwell.

Now if they will only use 64MB for Broadwells eDRAM, it will be close to ~40mm2

It's >80mm2 in 22nm but that's for 128MB so 20mm2 for 64MB at 14nm sounds reasonable. But cost wise I'm not sure, even if it's a tiny die yields are affecting all 14nm parts right now.

 

[AtenRa]

To anyone with knowledge, looking at AtenRa's 2+3 guess, is such an oblong die reasonable? Are there any problems in manufacturing a thin and long die?
For heat dissipation it makes sense anyway, spreading apart the components rather than packing them in a square thing.

Haswell 2+3 is almost the same. I dont believe there is any problem unless 14nm has some peculiarities we dont know.

 

[Enigmoid]

Ok lets have some pics,

<snip>

Now if you measure each component we have,

Core-M 2+2 = 82mm2
GT2 = ~41mm2
Single CPU Core + L3 = ~10,5mm2
System Agent, Display Engine and Memory Controller = ~12mm2
Memory I/O = ~8mm2

For 2+3 (GT3) we have
GT3 = 2x41 = ~82mm2
CPU Cores = 2x10,5 = ~21mm2
System Agent etc = ~12mm2
Memory I/O = 8mm2 + Empty space 4mm2 = ~12mm2

= Total 2+3 die size = ~127mm2

For 4+3(Quad core + GT3)
GT3 = 2x41 = ~82mm2
CPU Cores = 4x10,5 = ~42mm2
System Agent etc = ~12mm2
Memory I/O = 8mm2

= Total 4+3 die size = ~144mm2

For 4+4 (GT4)
GT3 = 3x41 = ~123mm2
CPU Cores = 4x10,5 = ~42mm2
System Agent etc = ~12mm2
Memory I/O = 8mm2 + Empty space 8mm2 = ~16mm2 *

= Total 4+4 die size = ~193mm2

Here are some personal drownings estimations, could be completely wrong but what the hell

Great job!

Don't forget that not all of the slice is replicated.

Haswell non- replicated area.

Guessing these areas won't actually be there. About 1 core + L3 must be subtracted per duplicated slice making the dies even smaller.

It looks like die sizes will be more around
117 mm^2 for 2 + 3
134 mm^2 for 4 + 3
172 mm^2 for 4 + 4