Intel "Kaby Lake"

Grall

Invisible Member
Legend
Sweclockers posts today about a new "emergency" 14nm processor to avoid another Broadwell debacle with the upcoming Cannondale chip/11nm process.

35710


One interesting tidbit the eagle-eyed can discern from this product matrix is a H-processor line, possibly mobile/BGA variant, equipped with GT4 graphics and not just one, but actually TWO 128MB Crystalwell eDRAM cache chips...! That has the potential to be a real powerhouse, as far as on-chip graphics is concerned. Veeeery interesting, I must say! Especially if "TBD" in package column means it will be a socketed processor, preferably with 90W+ TDP.
 
One interesting tidbit the eagle-eyed can discern from this product matrix is a H-processor line, possibly mobile/BGA variant, equipped with GT4 graphics and not just one, but actually TWO 128MB Crystalwell eDRAM cache chips...! That has the potential to be a real powerhouse, as far as on-chip graphics is concerned. Veeeery interesting, I must say! Especially if "TBD" in package column means it will be a socketed processor, preferably with 90W+ TDP.

Could you enlighten me on how doubling capacity and possibly bandwidth would help these iGPUs at the performance level they are at especially if we believe what Intel said about first Iris Pro? Broadwell is 20% faster, while Skylake is said to be 50% better than Broadwell. That puts it behind 750Ti, additional 20-30% is required just to be on par.

"32MB is enough"
"effective 100-130GB/s bandwidth due to being a cache"

Also doubling capacity would require tags on the CPU to double as well. Right now it takes 2MB away from the L3. That means rather than 4MB L3 on Core i5 and 6MB L3 on Core i7 left over, we'd only have 2MB on i5 and 4MB on i7. This looks like an extremely hastily put part IMO.

Positives: maybe 10-20% over 1x128MB
Negatives: Double eDRAM TDP(3W to 6W), double OPI TDP(1W to 2W), third 80mm2 die on the package, 2MB more L3 cache taken away and additional die space on the chip for tags, increased package cost.

Why not 1x256MB with double OPI bandwidth?
 
I guess I missed something, somewhere: what was the "Broadwell debacle?"

I'd get it, if we were talking about priming the pump on a new lithography node. Since that obviously isn't the answer, what is it exactly that is being referenced here?
 
I guess I missed something, somewhere: what was the "Broadwell debacle?"

I'd get it, if we were talking about priming the pump on a new lithography node. Since that obviously isn't the answer, what is it exactly that is being referenced here?

The delays, presumably.
 
Was that a "debacle"? I suppose perception is the reality for someone who wants to see it that way.
 
Interesting. Were we in dire need of new processors? Was Intel running out of money? Did the competition for desktop processors finally catch up? Was the channel completely devoid of Haswell to sell?

I think "debacle" is, again, a perception thing in this case. I didn't see a debacle, I saw a pipeline stall that otherwise really didn't mean a whole lot. But that's just, like, my perception man.
 
If you intended to launch anything three quarters of a year ago, and instead you continue to sell your old product in the meantime, you then launch it now instead, and the successor to whatever you're launching now will be launched in barely another two months - well, let's just say that if we were talking about AMD it would be not just a debacle, but a disaster too on top. ;)

Nobody's debating that good ole haswell wasn't a pretty damn decent CPU, but it WAS intended to be retired about nine-ten months ago or so. So let's call a spade a spade, alright?
 
It is interesting to see Intel makes more usage of Crystal Well. I notice that they are to introduce 64MB variant, that should be pretty tiny ~40 mm2.
Intel dual core line is already pretty tiny ( Core M are 84 mm2), INtel may no longer want to bother dual channel memory controller. I wonder if Intel could have Core M replacing the (traditional) core i3 set-up. DDR4 on a single channel along with CW should offer enough bandwidth for solid GPU performances with great power characteristics. The Celeron/Pentium lines would lose CW, hyper threating, etc.
Looking at those chips withtwo CW I wonder if Intel consider moving away from dual channel set-up (super high end and server line aside).
 
Last edited:
Am I the only one reading this CPU's name and thinking "Baby Cake"?
 
My perception of Broadwell's launch was Core M chips being the priority and that 14nm yields were sucking. Why push to get those chips out when there really was not much pressure to do so. Devil's Canyon (Haswell 2.0) was interesting too. I think they just wanted to maximize Haswell ROI while working to make 14nm cheaper to manufacture. I sense the new strategy is "tick tick tock". ;)
 
Last edited:
I'm digging this in light of new leaks from Notebookcheck:

http://www.notebookcheck.net/Intel-Kaby-Lake-Quad-Core-ULV-Processors-in-2017.172188.0.html
http://www.notebookcheck.net/Intel-Detailed-Kaby-Lake-Road-Map-leaked.172282.0.html


yvgDEPR.jpg


wz8UuAm.jpg




Biggest news is that the 15W U line is getting a quad-core model. I imagine many gaming notebook brands wanting to release 12" laptops or 2-in-1s using these along with external GPUs through Thunderbolt 3.
The next Surface Book is probably bound to bring one of these too and then have a discrete GPU in the keyboard.

We also need to be aware that the first batch of Kaby Lake U processors is not coming with HDCP 2.2, meaning it won't be able to display most legal 4K content (bluray or streaming alike).



There are also some Core i7 7500U results hitting the website, where we can see a substantial GPU performance improvement of 10-50% between GT2 Skylake and GT2 Kaby Lake. It might have to do with power and temperature constraints, and/or they may be using desktop-class DDR4. There are some single-threaded CPU improvements too.
 
i'd love a quad core surface pro 5 with a thunderbolt gpu enclosure. I'd get a rx 460 and bring it with me for civ gaming while traveling
 
Back
Top