Coffeelake thread, benchmarks, reviews, input, everything.

[TheF34RChannel]

---

 

[Jan Olšan]

One thing is known for certain; Intel will release their 8 core chip around the same time or sooner than AMD releases their updated Ryzen chips. If you want to know what Intel is going to do, just watch AMD. Yep, those days are back again. Who knows how long they will stay though. Hopefully for a long time.

No, I think they'll do it as soon as they can get Z390 ready. They need to get those chips out ASAP, and that platform out doubly so.

What sources do we actually have for Intel planning an octocore Coffee Lake ? Is there anything but the dubious forum post by the Eurocom employee? That one could have been just a mistake or based on misunderstanding of the leaked roadmaps and outdated Cannon Lake information.

I mean, we can't rule it out, but octocore version was in no leaked documents about the Coffee platform, so far. Unless intel added it to its plans like six months later than it started to work on the hexacores and at the same time somehow managed to keep that secret, there's no way, IMHO.
Well, except if they manage to make such processor from Skylake-D dies (Xeon D based on Skylake, which lacks iGPU - not sure this is possible to do. The chip was aimed for different platform/BGA and for lower clocks, unless Intel also made sure it can be used as LGA-1151 plan B from the start).

In any case, if such chip was real, we should start to see leaks, if not now, then in the near future. ES chips have to exist 9 months before release at least. So basically, I am quite sceptical.

As for Z370 supporting the 8 core who knows.

If the octocore happens, it will most likely support Z370 too, it's not like there is any difference from the pov of the whole platform - if Z270/Z370 is able to support the hexacores, it can support octocore too. There could be power delivery shenanigans again - which would show that Intel is deeply in improvising mode - instead, which is unrelated to chipset itself. But I doubt it.

 

[mikk]

True but that was an emergency paper launch aimed at destroying Ryzen DIY sales. The real release is in January which would be the typical year.

CFL 8C is an emergency model as well just like CFL 6C. Both weren't planned.

Icelake on desktop is dead.

Nonsense.

 

[jpiniero]

CFL 8C is an emergency model as well just like CFL 6C. Both weren't planned.

Not planned yes but the need to rush it out ASAP isn't there.

 

[PeterScott]

CFL 8C is an emergency model as well just like CFL 6C. Both weren't planned.

When lead time in years, and tens of millions in masking costs alone, everything is planned.

CL may have been part of contingency plan (for 10nm issues), but it was planned.

 

[Arachnotronic]

When lead time in years, and tens of millions in masking costs alone, everything is planned.

CL may have been part of contingency plan (for 10nm issues), but it was planned.

Coffee Lake was added to the roadmap in 2H 2015. The 6-core version taped out in late 2016 (a fact posted on these forums actually), and it launched in early October.

An 8 core version should take less time to do than the 6 core version because there is no process change.

 

[mikk]

When lead time in years, and tens of millions in masking costs alone, everything is planned.

CL may have been part of contingency plan (for 10nm issues), but it was planned.

It wasn't planned until Intel realized they had to cancel Cannonlake for desktop entirely because of their ongoing 10nm issues. Coffeelake is an emergency generation for the canceled Cannonlake, both 6C and 8C.

 

[jpiniero]

6 yes, 8 no. 8 is coming because Icelake is now falling apart.

 

[Dayman1225]

6 yes, 8 no. 8 is coming because Icelake is now falling apart.

That's a big assumption you have no proof of. Jumping to conclusions is stupid without facts or proof to back it up.

 

[Arachnotronic]

6 yes, 8 no. 8 is coming because Icelake is now falling apart.

Here we go again.

 

[Dayman1225]

There is absolutely no indication that Icelake is falling apart or is delayed. So where you got that from, I do not know.

Now talking about Coffee Lake. The 8 core SKU coming out much later than the rest of the lineup tells me they did not originally plan this die and slotted it in due to Ryzen. BUT we have still yet to confirm that the 8c sku exists.

 

[TheF34RChannel]

It wasn't planned until Intel realized they had to cancel Cannonlake for desktop entirely because of their ongoing 10nm issues. Coffeelake is an emergency generation for the canceled Cannonlake, both 6C and 8C.

I sincerely doubt that Intel realised in 2015 they had to can CNL, and equally, it was planned (back up or not) because they cannot shake a tree and have CPUs falling out at will - if only

 

[DrMrLordX]

Icelake on desktop is dead.

Eh?

Edit: WTH am I talking about. I like budget chips, not these $1000 monstrosities.

Don't worry man, you don't need to get a $1k chip anymore to get awesome-ness-ness-ness. But $300 is still over the usual Larry range.

Ryzen Launched Feb-Mar 2017, I think the earliest anyone expects the refresh Pinnacle Ridge (Ryzen 12nm), is Feb-Mar 2018 (some are saying as late as Q3).

So I don't think Ryzen 2 is expected before Feb-Mar 2019. Leaked Slides show Ryzen 2 in 2019.

I would be pleased as punch to get AMD releasing on a schedule like that. It would be very very bad for all of us if . . . well see below.

The other part of it is that Icelake's clocks are going to be lower, likely enough to wipe out gains from any IPC increase and then some. So Icelake on desktop with an 8 core CFL available would be redundant at best.

Didn't stop them from launching Haswell.

What sources do we actually have for Intel planning an octocore Coffee Lake ? Is there anything but the dubious forum post by the Eurocom employee? That one could have been just a mistake or based on misunderstanding of the leaked roadmaps and outdated Cannon Lake information.

That's really all we've got. But what do you think about a future in which Intel releases nothing in 2018? Icelake looks like a 2019 product now. Without an 8c Coffeelake, that means Intel has 0 new desktop products from Oct 2017 to maybe July 2019. That's assuming they even launch Icelake for the desktop, though under this scenario Intel really HAS to do it. They have to launch something. They can't just wait until Sapphire Rapids!

And no I don't see Intel launching a mesh-based Skylake-X/Cascade Lake chip in 2018 as Coffeelake 8c.

There is absolutely no indication that Icelake is falling apart or is delayed. So where you got that from, I do not know.
.

8c or no 8c, Intel just can't afford to fail on Icelake anyway. They already punted on desktop Cannonlake. They have that patched-up okay-ish - people are buying Kabylake and Coffeelake. Problem solved. But I don't think they can get around another botched launch.

 

[Dayman1225]

8c or no 8c, Intel just can't afford to fail on Icelake anyway. They already punted on desktop Cannonlake. They have that patched-up okay-ish - people are buying Kabylake and Coffeelake. Problem solved. But I don't think they can get around another botched launch.

Not even that - With Zen 2 out being out around the same time as Icelake - Intel could not afford to put out another version of Skylake on 14nm - They would need a new arch (Icelake) and a smaller more efficient node to compete. Icelake is coming regardless if it clocks lower.

 

[jpiniero]

There is absolutely no indication that Icelake is falling apart or is delayed. So where you got that from, I do not know.

Besides all the rumors that 10 nm is in shambles? It's not Icelake that's the problem, it's 10 nm.

I agree that if it was just clocks Intel would continue on as planned, but they wouldn't do the 8 core CFL. It's one or the other, not both.

 

[Dayman1225]

Besides all the rumors that 10 nm is in shambles? It's not Icelake that's the problem, it's 10 nm.

The same reports we've been hearing over and over for months? If we don't see Cannonlake throughout H1 2018 - I'll agree with yah that Icelake may have problems coming out when it's meant to.

I agree that if it was just clocks Intel would continue on as planned, but they wouldn't do the 8 core CFL. It's one or the other, not both.

Why not both? If the Coffee lake 8c is real and it indeed clocks higher than Icelake would that means Coffee lake would be faster in MT at least. But Icelake would be much more efficient. Trade offs are made. But I do still believe current Icelake will arrive in H1 2019 as expected.

 

[elhefegaming]

let me get something straight.

Next year H310 B360 etcs will come out, which are (by definition) inferior to the z370, but those WILL support ICL just like Z390? is that correct?

 

[Mulrian]

We don't know anything yet. Z370 might support ICL or whatever 8 core SKU's Intel release next.

 

[Jan Olšan]

let me get something straight.

Next year H310 B360 etcs will come out, which are (by definition) inferior to the z370, but those WILL support ICL just like Z390? is that correct?

No way, Ice Lake is a new socket, new platform. Do you actually expect compatibility across architeture cycles, because Intel broke compatibility in the last third of its current architecture cycle? That's naïve IMHO.

I expect there will be no differences in compatibility and upgrade paths between Z370 and H310-Z390. Unless Intel got crazy and completely overhauled the platform policies.

 

[jpiniero]

Why not both? If the Coffee lake 8c is real and it indeed clocks higher than Icelake would that means Coffee lake would be faster in MT at least. But Icelake would be much more efficient. .

Esp with the way desktop sales are going, Intel wouldn't have spent the money to design the 8 core CFL unless they really needed to.... and assuming 10nm yields are that bad, they have to do something to fill the void.

 

[Arachnotronic]

Esp with the way desktop sales are going, Intel wouldn't have spent the money to design the 8 core CFL unless they really needed to.... and assuming 10nm yields are that bad, they have to do something to fill the void.

Design costs for an 8 core CFL would not be high. The hard work (taping out the key IPs to 14nm++) is already done. Adding a few ring stops and a few cores requires a new die to be made, but in terms of difficulty/complexity an 8 core CFL shouldn't be hard to do at all.

Also, to your point that 8C CFL is replacing ICL, this doesn't make sense. Even if you assume no improvement in per core perf, there are going to be improvements elsewhere in the SoC that will matter to the OEMs. Also, if Intel can't increase ST perf for enthusiasts, they could do a 10 core mainstream SKU that'd fetch a lot of $$.

 

[PeterScott]

Design costs for an 8 core CFL would not be high. The hard work (taping out the key IPs to 14nm++) is already done. Adding a few ring stops and a few cores requires a new die to be made, but in terms of difficulty/complexity an 8 core CFL shouldn't be hard to do at all.

Also, to your point that 8C CFL is replacing ICL, this doesn't make sense. Even if you assume no improvement in per core perf, there are going to be improvements elsewhere in the SoC that will matter to the OEMs. Also, if Intel can't increase ST perf for enthusiasts, they could do a 10 core mainstream SKU that'd fetch a lot of $$.

From what I read it is photomasking costs that have escalated through the roof, and for a new chip of different dimensions, I would expect you would need a whole new set of masks.

 

[Arachnotronic]

From what I read it is photomasking costs that have escalated through the roof, and for a new chip of different dimensions, I would expect you would need a whole new set of masks.

Of course you'd need a mask set, but an 8 core CFL will probably generate literally billions in revenue during its life time, $10-20 million for a new mask set is nothing.

 

[jpiniero]

Also, to your point that 8C CFL is replacing ICL, this doesn't make sense. Even if you assume no improvement in per core perf, there are going to be improvements elsewhere in the SoC that will matter to the OEMs. Also, if Intel can't increase ST perf for enthusiasts, they could do a 10 core mainstream SKU that'd fetch a lot of $$.

The point is that they are doing the 8 core Coffee Lake because 10 nm yield is so bad that an 8 core Icelake isn't realistic in the timeframe needed.

 

[Arachnotronic]

The point is that they are doing the 8 core Coffee Lake because 10 nm yield is so bad that an 8 core Icelake isn't realistic in the timeframe needed.

That's right, but why does that mean that Ice Lake is cancelled for the desktop? Do you think Intel will run the entirety of 2019 with Coffee Lake?