Intel Skylake / Kaby Lake

[mikk]

3 months? Do you have a source?

My i7-7700k was made in week 43 according to the batch number, I have bought and received this CPU in week 50 2016. Yes, before Intel officially launched Kabylake.

 

[mikk]

Any clue on this Geekbench result? It's an Ice Lake model according to Geekbench with some abnormal entries. It's running with 1C only which is explainable assuming it is a very early ES. The strange thing is there is a M7-6Y75 entry which is Skylake, but 2.00 Ghz doesn't match to base or turbo clock. L1 Data Cache is different to current mainstream SKUs, upped from 32 KB to 48 KB. L3 Cache much bigger. Processor ID doesn't match either, here is a direct comparison: https://browser.primatelabs.com/v4/cpu/compare/2400363?baseline=901997

Actually the scores are very low, only the memory scores are way better. They are way faster than my i7-7700k running with DDR4-3200 CL14.

 

[lolfail9001]

Any clue on this Geekbench result?

Fabricated perhaps? I mean, model 126? Come on.

 

[Sweepr]

There is a new Sisoftware listing from Cannonlake.

Genuine Intel(R) U 0000 @ 3.30GHz (2C 4T 3.31GHz, 2x 256kB L2, 4MB L3)
http://ranker.sisoftware.net/show_r...d4ecdce5dde9dafc8eb383a5c0a598a88efdc0f8&l=en

There is a difference to the first: Z0 and U0.

Intel Cannonlake Client platform CNL - Z0 Cannonlake Client System
Intel Cannonlake Client platform CNL - U0 Cannonlake Client System

The strange thing is that it reports Gen9 because as we know CNL is using Gen10. Also it reports that it is using a graphics unit with 48 EUs. The recent CNL ID listing says it is GT2 only.

Intel(R) HD Graphics Gen9 (384SP 48C 1GHz, 6.3GB) (OpenCL)
http://ranker.sisoftware.net/show_r...efdbe8d8ebd9e1c7b588b89efb9ea393b5c6fbc3&l=en

So Cannonlake might double the iGPU resources, freaking 48 EUs for the bread and butter mobile chips (Y/U SKUs), on top of a new/updated graphics architecture. Has Intel ever done this before? That would more than make up for Kaby Lake's stagnant 3D performance. About time as well, because dual-cores on a 10nm proccess should be tiny. Also the 3.3 GHz clockspeed looks good for a brand new process, not nearly as mature as their 14nm+/14nm++.

 

[mikk]

Fabricated perhaps? I mean, model 126? Come on.

What do you mean with model 126? Don't you think that a new generation, be it Icelake or Cannonlake or Gemini Lake whatever gets a new model number?

 

[Bouowmx]

Intel Core i5-7200U: Family 6 Model 142 Stepping 9
https://browser.primatelabs.com/v4/cpu/search?q=Intel+Core+i5-7200U

 

[mikk]

Intel Core i5-7200U: Family 6 Model 142 Stepping 9
https://browser.primatelabs.com/v4/cpu/search?q=Intel+Core+i5-7200U

Could you say something? So you are saying Intel must go with a higher number? Is there a rule that prevents Intel going a different route?

There was a huge jump from Skylake to Kabylake beside that Kabylake is only a new stepping, the CPU core itself was unchanged.

Intel Core m7-6Y75 GenuineIntel Family 6 Model 78
Intel Core i7-7Y75 GenuineIntel Family 6 Model 142

No idea why Intel introduced such a big jump. The model number is no indication unless you have sources with a different number. Not to mention that these numbers may change later in production. We can only say that this model number is unknown, there is no CPU on the market with this number. As expected for any CPU in early development.

 

[lolfail9001]

What do you mean with model 126? Don't you think that a new generation, be it Icelake or Cannonlake or Gemini Lake whatever gets a new model number?

What i mean with model 126 is that it would kind of weird to modify CPU-ID output but not concern yourself one bit with ASIC string output. Besides the memory latency result stinks of fake. Or geekbench getting tripped out by cache structure, possible too.

No idea why Intel introduced such a big jump.

Actually it is a very minor jump. They literally swapped 2 bits, that's it. Kaby Lake was really meant to be a new Skylake stepping, so much that they did not even concern themselves with changing model number on ES phase, and then went "oh damn" in final phase and did the laziest change possible to mask it being a different stepping only.

 

[jpiniero]

The latency score could be explained due to the LPDDR4X being connected via EMIB maybe?

 

[mikk]

What i mean with model 126 is that it would kind of weird to modify CPU-ID output but not concern yourself one bit with ASIC string output. Besides the memory latency result stinks of fake. Or geekbench getting tripped out by cache structure, possible too.

All these subscores scores are too boring for a fake imho because they are very low and why only 1 CPU core? Beside that almost all system infos have to be faked not only the ID: Model, L1 data cache, L3 cache, motherboard, northbridge, southbridge, bios

Memory scores could have a big impact due to new solutions, be it on-die memory or on-chip connected memory (EMIB). And it supports LPDDR4x according to the motherboard info.

http://www.anandtech.com/show/11021/sk-hynix-announces-8-gb-lpddr4x4266-dram-packages

 

[Ajay]

The latency score could be explained due to the LPDDR4X being connected via EMIB maybe?

Seems to me if latency goes up that much with EMIB, the EMIB suXors...
I think it's some other issue (if even real).

 

[lolfail9001]

Beside that almost all system infos have to be faked not only the ID

Guess how all that information is reported

Seems to me if latency goes up that much with EMIB, the EMIB suXors...

Geekbench measures latency in operations/second, this thing has latency between L1 and L2 cache level if it's any true, actually.

So the issue is that it is unrealistically good.

 

[imported_ats]

3 months? Do you have a source?

My i7-7700k was made in week 43 according to the batch number, I have bought and received this CPU in week 50 2016. Yes, before Intel officially launched Kabylake.

Batch number doesn't refer to start of manufacturing but end of manufacturing.

 

[mikk]

Batch number doesn't refer to start of manufacturing but end of manufacturing.

Then this production schedule is based on the end of production. Because there is usually a ~2 months difference from production to RTS in these type of Roadmaps. Intel could do a paper launch at Computex or a presentation, product availability not before early August I think.

 

[IntelUser2000]

So Cannonlake might double the iGPU resources, freaking 48 EUs for the bread and butter mobile chips (Y/U SKUs), on top of a new/updated graphics architecture. Has Intel ever done this before?

With Ivy Bridge, yes. Theoretical was 2x Sandy Bridge. Arguably its harder nowadays but still. Performance increase will depend on thermals because on the U chips the increase was only 30%: http://www.anandtech.com/show/5872/intel-dual-core-ivy-bridge-launch-and-ultrabook-review/5

That would more than make up for Kaby Lake's stagnant 3D performance. About time as well, because dual-cores on a 10nm proccess should be tiny. Also the 3.3 GHz clockspeed looks good for a brand new process, not nearly as mature as their 14nm+/14nm++.

3.3GHz if its base is an improvement over 7600U.

Geekbench measures latency in operations/second, this thing has latency between L1 and L2 cache level if it's any true, actually.

So the issue is that it is unrealistically good.

So much so its probably misreporting it. Icelake is a new architecture right? Perhaps that's got something to do with it?

The latency scores on the ST says ~3ns and MT its about ~6ns. Yea, that's not right.

The total bandwidth is believable though. Maybe not the copy.

 

[JoeRambo]

There was a huge jump from Skylake to Kabylake beside that Kabylake is only a new stepping, the CPU core itself was unchanged.

Intel Core m7-6Y75 GenuineIntel Family 6 Model 78
Intel Core i7-7Y75 GenuineIntel Family 6 Model 142

Actually the jump is not that huge, once one considers that they are both family E.
78 is 4E
142 is 8E

Btw 4C Kaby 7700K is guess what - 9E, so still perfectly fine.

126 is 7E, indicating unknown CPU model ?

 

[lolfail9001]

Actually the jump is not that huge, once one considers that they are both family E.

The jump is even smaller once you consider that 78 is 01001110b and 142 is 10001110b. They literally panic swapped 2 bits when management told them they would sell a new stepping as a new line-up and that was it.

 

[Sweepr]

Any clue on this Geekbench result? It's an Ice Lake model according to Geekbench with some abnormal entries. It's running with 1C only which is explainable assuming it is a very early ES. The strange thing is there is a M7-6Y75 entry which is Skylake, but 2.00 Ghz doesn't match to base or turbo clock. L1 Data Cache is different to current mainstream SKUs, upped from 32 KB to 48 KB. L3 Cache much bigger. Processor ID doesn't match either, here is a direct comparison: https://browser.primatelabs.com/v4/cpu/compare/2400363?baseline=901997

Actually the scores are very low, only the memory scores are way better. They are way faster than my i7-7700k running with DDR4-3200 CL14.

Nice find. Is it expected to have ES up and running at this point? Any chance client Ice Lake is actually a 2018 product, at least for mobile? Exciting times ahead.

 

[Ajay]

Nice find. Is it expected to have ES up and running at this point? Any chance client Ice Lake is actually a 2018 product, at least for mobile? Exciting times ahead.

IDK. Thought Ice Lake was going be on 10nm+. Oh wait, isn't 10nm+ due in 2018? I think my brain is starting to hurt with all the shifts, delays, re-branding, etc.

 

[LTC8K6]

Nice find. Is it expected to have ES up and running at this point? Any chance client Ice Lake is actually a 2018 product, at least for mobile? Exciting times ahead.

Well, they had a 10nm CL chip running in a 2 in 1 laptop at CES.

http://www.digitaltrends.com/computing/intel-cannon-lake-10nm-processor-teased-2in1-ces-2017/

 

[Sweepr]

With Ivy Bridge, yes. Theoretical was 2x Sandy Bridge. Arguably its harder nowadays but still. Performance increase will depend on thermals because on the U chips the increase was only 30%: http://www.anandtech.com/show/5872/intel-dual-core-ivy-bridge-launch-and-ultrabook-review/5

3.3GHz if its base is an improvement over 7600U.

Although Ivy Bridge was impressive thanks to the (then) new graphics architecture and higher clockspeeds, it only had 33% more EUs than Sandy Bridge's iGPU. If the SiSoftware entry for Cannon Lake is correct, we're getting 48 EUs, twice what Broadwell/Skylake/Kaby Lake GT2 offered + Gen 10 architecture (previous GFXBench finding). It's easier to understand why CFL-U 4+3e is being offered now. They want to avoid consumers/OEMs having to choose between quad-core and faster graphics in this power bracket (KBL-U Refresh gets 4C/8T but only 24 EUs).

 

[IntelUser2000]

Although Ivy Bridge was impressive thanks to the (then) new graphics architecture and higher clockspeeds, it only had 33% more EUs than Sandy Bridge's iGPU. If the SiSoftware entry for Cannon Lake is correct, we're getting 48 EUs, twice what Broadwell/Skylake/Kaby Lake GT2 offered + Gen 10 architecture (previous GFXBench finding). It's easier to understand why CFL-U 4+3e is being offered now. They want to avoid consumers/OEMs having to choose between quad-core and faster graphics in this power bracket (KBL-U Refresh gets 4C/8T but only 24 EUs).

Cores are mature so there likely won't be big per shader improvements like Ivy Bridge had. So theoretical increase is the same.

Thermals are the big deal in the 15W space. I doubt it'll turn out to be better than 30 percent there. If anything I'd bet on the gains being less.

 

[Bouowmx]

Thermals aren't an issue (not like NVIDIA GPU Boost); just the fact that big integrated graphics is running on paltry power budget (see AMD Bristol Ridge 15 W).

 

[IntelUser2000]

Thermals aren't an issue (not like NVIDIA GPU Boost); just the fact that big integrated graphics is running on paltry power budget (see AMD Bristol Ridge 15 W).

Right.

Wasn't Cannonlake the one with 48EUs? Is that changed too or it was always Icelake?

Early on in development GPUs had room for noticeable gains at same process and power. Now it's nearly all dependent on power and thermal constraints. The last generation that brought big performance-wise changes were Gen 7. Gen 8 and 9 advances were in corner cases scenarios and overall minimal. I do not believe it'll change in Gen 10.

So assume Cannonlake brings 30% gain. Icelake for a other 30%?

 

[Ajay]

Cores are mature so there likely won't be big per shader improvements like Ivy Bridge had. So theoretical increase is the same.

Thermals are the big deal in the 15W space. I doubt it'll turn out to be better than 30 percent there. If anything I'd bet on the gains being less.

Why? I would think Intel is capable of re-architecturing the iGPU to increase perf and perf/watt. Or do you think they will be lazy and just go with a 'dumb' shrink?