Intel Skylake / Kaby Lake

Page 415 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

SAAA

Senior member
May 14, 2014
541
126
116
ShopBLT (US) has two Skylake-SP models up for ore-order.



This website only list base clocks for Xeon, and we know from previous leaks that the new lineup packs a new cache structure (total of 1.375 MB L3 per core). 16.5 MB cache '5122' above indicates they have a 12C/24T model with 3.6 GHz base clock. :)

...

200W TDP? I don't care. Price is where one expects, clock is absurdly high, much better than what my crystal ball could predict. Turbo? Can't wait to see some numbers, 14nm+ is definitely awesome.
Bodes very well for 6-8 cores Skylake's, Coffe is interesting but unless 14++ is phenomenal and its IPC grows slightly there shouldn't be competition against a soldered 140W TDP cpu.
I only hope prices are sane and we get 6 cores at less than $400 for both...
 
  • Like
Reactions: Sweepr
Mar 10, 2006
11,715
2,012
126
...

200W TDP? I don't care. Price is where one expects, clock is absurdly high, much better than what my crystal ball could predict. Turbo? Can't wait to see some numbers, 14nm+ is definitely awesome.
Bodes very well for 6-8 cores Skylake's, Coffe is interesting but unless 14++ is phenomenal and its IPC grows slightly there shouldn't be competition against a soldered 140W TDP cpu.
I only hope prices are sane and we get 6 cores at less than $400 for both...

14nm++ is actually pretty awesome, but I think IPC on Coffee will be worse than on SKX. It's going to be an annoying tradeoff.
 

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
14nm++ is actually pretty awesome, but I think IPC on Coffee will be worse than on SKX. It's going to be an annoying tradeoff.

It surely will be interesting to see how the new cache structure on Skylake X performs and what IPC improvements it brings. I still think Coffeelake will be the top 6C/12T CPU as it will easily beat Skylake-X 6C/12T by more than 10% in stock and max clocks which I doubt the Skylake X can overcome with higher IPC. 12% higher transistor performance is going to allow CFL 6C/12T to easily break 5 Ghz and probably go past 5.2 Ghz. imo CFL 6C/12T unlocked CPU will be the CPU to buy in Q3 2017.
 
  • Like
Reactions: ozzy702

ozzy702

Golden Member
Nov 1, 2011
1,151
530
136
An actual IPC regression, or a per core throughput regression due to less average clockspeed headroom?

Coffeelake will essentially have Kabylake/Skylake IPC. Skylake X is a different core than Skylake/Kabylake and has much more cache. Nobody knows what IPC for Skylake X looks like yet but I don't see Intel regressing on IPC with Skylake X so the assumption is made that Skylake X will have higher IPC but lower clocks than Coffeelake. How that balances out is anyone's guess.
 

crashtech

Lifer
Jan 4, 2013
10,521
2,111
146
Coffeelake will essentially have Kabylake/Skylake IPC. Skylake X is a different core than Skylake/Kabylake and has much more cache. Nobody knows what IPC for Skylake X looks like yet but I don't see Intel regressing on IPC with Skylake X so the assumption is made that Skylake X will have higher IPC but lower clocks than Coffeelake. How that balances out is anyone's guess.
Oh, yeah, I'd neglected to consider the new cache setup for SKX. Arachnotronic's post makes much more sense in that context.
 
  • Like
Reactions: Arachnotronic

tamz_msc

Diamond Member
Jan 5, 2017
3,726
3,560
136
It surely will be interesting to see how the new cache structure on Skylake X performs and what IPC improvements it brings. I still think Coffeelake will be the top 6C/12T CPU as it will easily beat Skylake-X 6C/12T by more than 10% in stock and max clocks which I doubt the Skylake X can overcome with higher IPC. 12% higher transistor performance is going to allow CFL 6C/12T to easily break 5 Ghz and probably go past 5.2 Ghz. imo CFL 6C/12T unlocked CPU will be the CPU to buy in Q3 2017.
There we go again, one graph showing better leakage characteristics means that CFL will break OC records on a 6 core chip. It's high time that people stop predicting the future using their wishful thinking coated as absolute facts.
 
  • Like
Reactions: Drazick and dfk7677

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
There we go again, one graph showing better leakage characteristics means that CFL will break OC records on a 6 core chip. It's high time that people stop predicting the future using their wishful thinking coated as absolute facts.
Being realistic is too boring: who wants to discuss on such terms as expecting SKX to clock 5-10% higher than BDWE and CFL 6c to clock 5-10% lower than KBL 4c, which would put similar parts from both sockets within 5% of each other in clocks for stock configs.
 
  • Like
Reactions: Drazick

vissarix

Senior member
Jun 12, 2015
297
96
101
There we go again, one graph showing better leakage characteristics means that CFL will break OC records on a 6 core chip. It's high time that people stop predicting the future using their wishful thinking coated as absolute facts.
After reading for months about Ryzen overclocking to 4.7ghz and 25% higher IPC compared to Skylake we can also try to dream big here...not as much as on the Ryzen thread tho..:cool:
 

tamz_msc

Diamond Member
Jan 5, 2017
3,726
3,560
136
After reading for months about Ryzen overclocking to 4.7ghz and 25% higher IPC compared to Skylake we can also try to dream big here...not as much as on the Ryzen thread tho..:cool:
That is partly true though, if you completely max out the uop cache in each then for certain integer operations Ryzen is 25% faster than Skylake in IPC.;)

Unlike some of the posts on this thread being completely clueless about fitting 12MB L3 on 149mm^2. The SiSoft leaks shows that i7s might be 6C12T with 9MB L3. That's credible speculation.

A 6 core Coffee Lake part breaking 5.2GHz just because of improved transistors is not.
 
  • Like
Reactions: Drazick

ehume

Golden Member
Nov 6, 2009
1,511
73
91
Regardless of what it OC's to, Coffee Lake looks like it will be specified to be a 135W CPU with 6 cores. Q: will it have an inbuilt GPU?

The reason I ask is that I am in the market for a CPU to stress heatsinks, to comparatively test them. A chip that is 135W before OC looks promising indeed. I could only push my 4790K to the low 150's in the days before it became useless for comparative heatsink-testing. There was one good thing about the chip though: the inbuilt GPU did not add heat to the test system the way a discrete GPU would do.
 
  • Like
Reactions: Drazick

LTC8K6

Lifer
Mar 10, 2004
28,520
1,575
126
Regardless of what it OC's to, Coffee Lake looks like it will be specified to be a 135W CPU with 6 cores. Q: will it have an inbuilt GPU?

The reason I ask is that I am in the market for a CPU to stress heatsinks, to comparatively test them. A chip that is 135W before OC looks promising indeed. I could only push my 4790K to the low 150's in the days before it became useless for comparative heatsink-testing. There was one good thing about the chip though: the inbuilt GPU did not add heat to the test system the way a discrete GPU would do.
It's DOA if it's 135W tdp, imo.

I don't expect it to be anywhere near that, and I think 135W tdp would probably eliminate current boards. 7700K is only 91W tdp. You'd be talking about a 50% tdp increase.
 

LTC8K6

Lifer
Mar 10, 2004
28,520
1,575
126
That is partly true though, if you completely max out the uop cache in each then for certain integer operations Ryzen is 25% faster than Skylake in IPC.;)

Unlike some of the posts on this thread being completely clueless about fitting 12MB L3 on 149mm^2. The SiSoft leaks shows that i7s might be 6C12T with 9MB L3. That's credible speculation.

A 6 core Coffee Lake part breaking 5.2GHz just because of improved transistors is not.
The Sisoft leaked chip was 6C6T with 9mb cache.
Presumably if there is a 6C12T version, it would have more cache? 12mb?
 

tamz_msc

Diamond Member
Jan 5, 2017
3,726
3,560
136
The Sisoft leaked chip was 6C6T with 9mb cache.
Presumably if there is a 6C12T version, it would have more cache? 12mb?
Or it could be that HT was disabled. Where is the evience that hints at 12 MB L3? For that either Intel must have found a way to shrink the size of the cache or some other die larger than 149 mm^2 must exist. Where is the evidence for either of these possibilities?
 
  • Like
Reactions: Drazick