Intel Skylake / Kaby Lake

[jpiniero]

The added IPU looks pretty large. Considering it's about the same size as the 4+2 Kaby, it does make me think that perhaps there is some amount of shrinkage in 14++.

 

[Arachnotronic]

The added IPU looks pretty large. Considering it's about the same size as the 4+2 Kaby, it does make me think that perhaps there is some amount of shrinkage in 14++.

There isn't.

 

[Excessi0n]

Why would it be a marketing decision? Is it so hard to understand, that a 6-core will need much more power than a 4-core and hence power deliver of the socket had to be adjusted? We will see it in reviews. The 8700k will use at least 30% more power than a 7700k at full load.

So? A lot of Z170/Z270 motherboards have overkill power delivery which could easily handle that. Mine's fine with my overclocked 6700K, for example, even though it draws twice as much power as it does at stock!

 

[mikk]

The added IPU looks pretty large. Considering it's about the same size as the 4+2 Kaby, it does make me think that perhaps there is some amount of shrinkage in 14++.

~5 mm² bigger I would say.

 

[IntelUser2000]

~5 mm² bigger I would say.

I don't think they added anything. The U chips have a few differences.

U: 12 PCIe lanes, uses 4GT/s QPI for connection to the chipset(they call it OPI)
non-U: 16 PCIe lanes, uses 8GT/s DMI for connection to the chipset

You can see that U has some blocks that regular chips don't have and vice versa. Also if you compare the two dies, the U chips have almost no empty space, while the regular chip has it on the top rightmost corner of the chip. They moved the memory controller portion all the way to the right to fill up the blank space, cut some PCIe lanes, and put an IPU in there, it looks like.

Regardless, Kabylake-R is a significant rework not a stepping change.

 

[Edrick]

You seem quite upset. I find this oddly strange, especially considering I have never been wrong around here and people know this. 8700K will WRECK any reason to get HEDT for 90% of people out there. You are among the leet 10% cause 2 more cores for reasons.

Not upset at all. Just pointing out the ridiculous claim you made with zero facts to back it up. But seeing how you are the only person on these forums who is never wrong, who am I to argue.

I will make sure to check Intel's HEDT sales after the 8700K gets released. I should see a 90% drop in sales after that according to you.

 

[moonbogg]

That's correct. All the enthusiast (overpriced) gaming sales will go to 8700K and any sales requiring actual CPU power will go to the competition. I see no way around this happening. If I wanted a beast gaming CPU, I couldn't resist 6/12 coffee lake cores at 4.8ghz or whatever for hopefully under $400. Still too expensive, but I wouldn't be able to resist that personally. If I wanted a monster editing or rendering rig, or just beast multi task performance, how could I NOT buy a 16/32 chip for $1000? I absolutely would and I'd be totally jazzed about it too. X299 is in a dead zone.

 

[Lodix]

The added IPU looks pretty large. Considering it's about the same size as the 4+2 Kaby, it does make me think that perhaps there is some amount of shrinkage in 14++.

~5 mm² bigger I would say.

Intel's 14nm++ is less dense than previous 14nm.

EDIT: But aren't this chips made at 14nm+ ?

 

[crashtech]

So? A lot of Z170/Z270 motherboards have overkill power delivery which could easily handle that. Mine's fine with my overclocked 6700K, for example, even though it draws twice as much power as it does at stock!

Yes, but what about an OCed hexacore? Do you know for certain that the current socket has enough power pins to feed that?

 

[pantsaregood]

Yes, but what about an OCed hexacore? Do you know for certain that the current socket has enough power pins to feed that?

Given that getting Z170 and Z270 to push more than 250W to the CPU with a good VRM and cooling isn't unusual? Yes. Yes I do.

There are Z170 and Z270 boards designed to provide up to 488W to the CPU. I don't think the i7-8700K is going to exceed that.

 

[Sweepr]

Core i7-8700K (ES) - New Leaked Benchmark Results

https://zhuanlan.zhihu.com/p/28803237

 

[jpiniero]

Weird, wonder why the single thread is slower than the 7700K.

 

[crashtech]

Given that getting Z170 and Z270 to push more than 250W to the CPU with a good VRM and cooling isn't unusual? Yes. Yes I do.

There are Z170 and Z270 boards designed to provide up to 488W to the CPU. I don't think the i7-8700K is going to exceed that.

A link would be helpful, as well as substantiation that such power delivery numbers don't break specs.

 

[Shivansps]

In that case, only 6C should me incompatible with 200 chipsets.

 

[CakeMonster]

Still confused about Z390. Is it supposed to be for CFL? Or is it meant for the next generation? Was Z370 put together now because Z390 wouldn't be ready in time? Is it worth waying x months to buy a 8700K so that I can get it with Z390?

 

[Arachnotronic]

Still confused about Z390. Is it supposed to be for CFL? Or is it meant for the next generation? Was Z370 put together now because Z390 wouldn't be ready in time? Is it worth waying x months to buy a 8700K so that I can get it with Z390?

Z390 is a new PCH, with some features like native USB 3.1 and some integrated connectivity stuff manufactured in 14nm rather than 22nm, so should be slightly more efficient, too.

But, it's really not a big deal.

 

[Arachnotronic]

Weird, wonder why the single thread is slower than the 7700K.

If the benchmark was run with multiple background threads hitting the cores, even if they were relatively light, you would probably see this thing running at all-core turbo speed (4.3GHz) rather than single-core turbo speed.

 

[Arachnotronic]

That's correct. All the enthusiast (overpriced) gaming sales will go to 8700K and any sales requiring actual CPU power will go to the competition. I see no way around this happening. If I wanted a beast gaming CPU, I couldn't resist 6/12 coffee lake cores at 4.8ghz or whatever for hopefully under $400. Still too expensive, but I wouldn't be able to resist that personally. If I wanted a monster editing or rendering rig, or just beast multi task performance, how could I NOT buy a 16/32 chip for $1000? I absolutely would and I'd be totally jazzed about it too. X299 is in a dead zone.

I like my 7900X because paired with fast RAM + uncore OC to 3.2GHz + CPU OC to 4.6GHz, I really get the best of both worlds: beastly gaming performance (what I mainly use my PC for) and beastly multi-tasking/thread-heavy performance when I need it (rarely, but it does happen).

SKX is a good all-around CPU, especially if you take the time to tweak it properly.

 

[pantsaregood]

A link would be helpful, as well as substantiation that such power delivery numbers don't break specs.

A single 8-pin CPU connector is capable of delivering 288W, so my numbers were actually wrong. Two 8-pin connectors can theoretically provide up to 576W to the CPU.

This ignores the fact that Coffee Lake is painfully unlikely to use >50% more power than Kaby Lake or Skylake - the die isn't even 50% bigger. Z170 and Z270 are quite happy to run even when the CPUs in them are running at over twice their stock power consumption levels. If you're dead set on seeing it, I'll downclock my i7-6700K on my EVGA Z170 FTW to stock and then bring it up to 5.0 GHz just so you can see how significantly it increases power consumption - while the board doesn't really care despite having far from the best VRM on Z170.

 

[Wyrm]

I like my 7900X because paired with fast RAM + uncore OC to 3.2GHz + CPU OC to 4.6GHz, I really get the best of both worlds: beastly gaming performance (what I mainly use my PC for) and beastly multi-tasking/thread-heavy performance when I need it (rarely, but it does happen).

SKX is a good all-around CPU, especially if you take the time to tweak it properly.

What maximum sustained bandwidth were you able to achieve with the overclocked uncore/memory? Is your cooling a custom loop or an AIO?

 

[crashtech]

A single 8-pin CPU connector is capable of delivering 288W, so my numbers were actually wrong. Two 8-pin connectors can theoretically provide up to 576W to the CPU.

This ignores the fact that Coffee Lake is painfully unlikely to use >50% more power than Kaby Lake or Skylake - the die isn't even 50% bigger. Z170 and Z270 are quite happy to run even when the CPUs in them are running at over twice their stock power consumption levels. If you're dead set on seeing it, I'll downclock my i7-6700K on my EVGA Z170 FTW to stock and then bring it up to 5.0 GHz just so you can see how significantly it increases power consumption - while the board doesn't really care despite having far from the best VRM on Z170.

The 8-pin connector(s) are only one minor part of the discussion, the more relevant issue is how many pins on the socket provide power, and whether it's feasible to share their maximum current carrying capacity with two more cores without creating other problems.

 

[Arachnotronic]

What maximum sustained bandwidth were you able to achieve with the overclocked uncore/memory? Is your cooling a custom loop or an AIO?

240mm AIO.

 

[Wyrm]

240mm AIO.

Thank you. These are great numbers but probably not the absolute best you can do with the Skylake-X platform.

If I go with SKL instead of TR, I will likely go further for my bandwidth limited applications (image processing). G.Skill has Trident Z Black series 4GHz 64GB Samsung dies gradually showing up in retail:

 

[Arachnotronic]

Thank you. These are great numbers but probably not the absolute best you can do with the Skylake-X platform.

If I go with SKL instead of TR, I will likely go further for my bandwidth limited applications (image processing). G.Skill has Trident Z Black series 4GHz 64GB Samsung dies gradually showing up in retail:

Yeah, I only have DDR4-3000 (bought SKX day 1 before the QVL lists were out so I had to play it safe), but it's still a lot better than stock DDR4-2666 and mesh speed of 2400.

 

[TahoeDust]