Intel Skylake / Kaby Lake

[Sweepr]

Navin Shenoy (Intel) talks about Ryzen @ J.P. Morgan Conference

I don’t know what their product will actually perform at in a non-controlled way. Obviously every vendor is going to show off the best benchmark they can find. So the question is, how does it actually perform on general use cases that everybody’s going to use and we’ll see.
I’m very confident based on the data I’ve seen that we will continue to be the absolute prime leader on performance on every benchmark that matters for single-threaded performance. On multi-threaded, we’ll have a whole set of products that we launched actually last year, our Broadwell-E platform, our ten core, our eight core, and our six core Broadwell-E platform.
I’m confident that that product line, on multithreaded, will be a leader bar none. We’ll see, it’s going to be fun. Going to be a fun 2017; it’s been a while since we’ve had some interesting dynamics in the desktop CPU space, so we’ll see how it plays out.

They are not talking Skylake-X/Kaby Lake-X yet, no surprise. We should hear more at Computex and Gamescom.


ASRock Z270M-STX MXM Micro-STX Motherboard Pictured

https://smallformfactor.net/news/asrock-z270m-stx-mxm-pictured

Cool form factor.

 

[mikk]

If Coffee Lake is meant as a backup plan for 10nm delays, then surely it will be based on the new platform (With FIVR)

Because?

Kabylake was a backup as well, not much changes there.

 

[VirtualLarry]

ASRock Z270M-STX MXM Micro-STX Motherboard Pictured

So, even before mini-STX gains traction in the marketplace, they're (replacing?) it with a bigger micro-STX?

Sounds cool, in theory, I guess.

This allows for constant Intel Turboboost clocks without throttling since their was no need to throttle. However, this means that overclocking is not possible (as mentioned)

 

[SAAA]

Because?

Kabylake was a backup as well, not much changes there.

Well Coffelake is coming much later and the die must be redesigned anyway for 6 cores, so some kind of change in the architecture isn't implausible.

Then we also have this:

...that points to identical IGP but slightly larger die for CFL 4+2 vs KBL 4+2.
I know 3mm^2 isn't the end of the world but cores are so small today that any IPC change Cannonlake could bring might well fit in there.
Or FIVR maybe? Though it doesn't make much sense on desktop and Icelake should bring that back from what I've read.

 

[mikk]

Much later doesn't mean anything. A slightly bigger size could be caused by an updated display controller. CPU and GPU core are probably unchanged.

 

[Arachnotronic]

Much later doesn't mean anything. A slightly bigger size could be caused by an updated display controller. CPU and GPU core are probably unchanged.

Coffee Lake could also be built in a new variant of the process that sacrifices density slightly for another improvement in speed/power consumption.

Remember, Coffee Lake != Kaby Lake Refresh U, I think that Kaby Lake Refresh U will be on 14nm+ (since it's a Kaby Lake) while the Coffee Lake name implies a next iteration of 14nm.

 

[Justinbaileyman]

Oh wow that is fricken cool!! I am totally into tiny computers so this is right up my alley!! What is that longish black port thing sitting right in front of the the I/O ports? Is it some sort of PCIe slot or something?

 

[AtenRa]

Oh wow that is fricken cool!! I am totally into tiny computers so this is right up my alley!! What is that longish black port thing sitting right in front of the the I/O ports? Is it some sort of PCIe slot or something?

Yes that is a MXM slot for graphics cards

more pics in the link bellow
https://smallformfactor.net/news/asrock-z270m-stx-mxm-pictured

 

[CakeMonster]

So, Coffee is next for desktop? And we assume it will be 14nm, and both 4c8t and 6c12t? Am I reading that right?

In that case, those hunting for better IPC/single thread performance for games will have very little reason to be excited unless Intel is planning a huge surprise with architectural revamp on 14nm. The 6c part will have a hard time getting to the same speeds as KL and CL 4c.

 

[lolfail9001]

In that case, those hunting for better IPC/single thread performance for games will have very little reason to be excited

Uh, yeah? Those can lay on the couch until at the very least Icelake and maybe even Sapphire Rapid (or whatever the name was).

 

[jpiniero]

So, Coffee is next for desktop? And we assume it will be 14nm, and both 4c8t and 6c12t? Am I reading that right?

Yep.

In that case, those hunting for better IPC/single thread performance for games will have very little reason to be excited unless Intel is planning a huge surprise with architectural revamp on 14nm. The 6c part will have a hard time getting to the same speeds as KL and CL 4c.

Those people would be better off with Skylake-X... of course you would have to pay up though.

 

[CakeMonster]

At this point, I'm actually willing to pay up, but for the last 4+ years, 4c/8t has still delivered better clock speed and given me no reason to go for HEDT or more cores for gaming. And from what I understand that wouldn't change with SL-X nor with CL? Or have I missed something about X?

 

[jpiniero]

At this point, I'm actually willing to pay up, but for the last 4+ years, 4c/8t has still delivered better clock speed and given me no reason to go for HEDT or more cores for gaming. And from what I understand that wouldn't change with SL-X nor with CL? Or have I missed something about X?

Skylake-X is different since it's based upon (what should be) a faster core than mainstream Skylake. The max upside is 20% I'd say, but maybe 10-15%.

 

[CakeMonster]

That seems a bit hard to believe given that its still called Skylake, and that kind of suggest fewer changes than Kaby, which had practically none. But I have little reason to upgrade my 6700K for Kaby anyway, so I'll just hold on and see what it brings.

 

[jpiniero]

That seems a bit hard to believe given that its still called Skylake, and that kind of suggest fewer changes than Kaby, which had practically none. But I have little reason to upgrade my 6700K for Kaby anyway, so I'll just hold on and see what it brings.

Well if you have a 6700K you don't need to upgrade for some time.

 

[IntelUser2000]

Skylake-X is different since it's based upon (what should be) a faster core than mainstream Skylake. The max upside is 20% I'd say, but maybe 10-15%.

The performance increase is probably less than 5%. The increased L2 cache might be better suited for server and for consumers it might even hurt if latency increases by a decent amount.

 

[jpiniero]

The performance increase is probably less than 5%. The increased L2 cache might be better suited for server and for consumers it might even hurt if latency increases by a decent amount.

And the AVX-512 units, which is pretty much confirmed by the Sandra scores. I think mainstream Skylake was gimped further than that because of the desire to make it work in tablets.

 

[lolfail9001]

The performance increase is probably less than 5%. The increased L2 cache might be better suited for server and for consumers it might even hurt if latency increases by a decent amount.

My only concern when i hear about cache changes is oc potential. I suspect Skylake-X won't do Kaby Lake justice in overclocking potential despite similar manufacturing node (or rather, exact same one).

Also, AVX 512 means we're cooking eggs on CPUs now during stability tests.

 

[StrangerGuy]

https://smallformfactor.net/news/asrock-z270m-stx-mxm-pictured
Cool form factor.

I wished all desktop mobos are SODIMM by now. Other not being able to put near-pointless RAM heatsinks there is really no good technical or cost reason for DDR4 to have a desktop DIMM factor judging from the past few years.

 

[VirtualLarry]

But... do they make SO-DIMMs that clock to 4000? I was under the impression that desktop DIMMs were performance-optimized, and SO-DIMMs were density-optimized.

If they make DDR4-4000 SO-DIMMs, in 16GB per DIMM sizes, then sure, I'd be for switching desktop boards to SO-DIMMs. But until then, no.

 

[IntelUser2000]

And the AVX-512 units, which is pretty much confirmed by the Sandra scores. I think mainstream Skylake was gimped further than that because of the desire to make it work in tablets.

I disagree on the gimp part. I'd call it consumer optimized. They should do even more and capitalize on what Apple supposedly did with their chips, by lowering frequency but increase perf per clock like Conroe. Going from 4.5GHz to 3.5GHz but 40% perf per clock increase. Or whatever. The must know that 5GHz is a wall and 10 years from now we'll have CPUs performing 10% better per clock with same clock speeds.

This is why we need storage as memory solutions like 3D Xpoint to succeed. I'd be fine with going back to Clarkdale like CPU if fully optimized platform in 2022 or something offers truly instant computing(Instant boot with no sleep states or system S states, instant application launch, instant disk bound maintenance operations like anti virus). In fact, that would be the real "640KB is enough" platform. CPUs are dead aside from replacements for the vast majority of people.

I was under the impression that desktop DIMMs were performance-optimized, and SO-DIMMs were density-optimized

IMO somewhat true. Probably perf/$/density optimized. Looking at Apple you could go smaller than SODIMM and end up even faster at 4266MHz. But that small with the capacity expected on Desktops? Probably very costly. At least it'll end up costing US a lot.

 

[Nothingness]

Also, AVX 512 means we're cooking eggs on CPUs now during stability tests.

It means that when AVX is used voltage and clock are lowered, to prevent heat issues.

 

[Sweepr]

Kaby Lake individual core overclocking

 

[PPB]

That mini stx board is useless considering mxm boards are scarse and they go for 1.5 the cash of a desktop counterpart but without cooling.

If intel aibs really want to push sff, they need to increase the availability and diversity of their mini itx PIO boarda. Those are excelent for sff because they get rid of the need of a riser, and you can use good ol' pcie gpu format without compromises

Sent from my XT1040 using Tapatalk

 

[Bouowmx]

Is that actually per-core overclocking, or overclocking based on number of active cores, as the names, ex. "4-core ratio limit", suggest?

ASUS and Broadwell: