Intel Skylake / Kaby Lake

[twothreefive]

Ryzen has Haswell IPC at best.

This is what I have heard/read. If you take a 10% IPC improvement from Haswell to Skylake and add in the "potentially" better overclocks on the Skylake X vs Ryzen, the price difference for the Intel might be worth it. Granted, Ryzen will be more power efficient. A friend built a 1700x ssytem and hit a wall after 3.9 GHZ. 4.0 GHZ was not stable unless a huge increase in the voltage was applied. This seems to be the standard result on Ryzen. Don't get me wrong, I am glad AMD is competitive again and lit a fire under Intel! The Ryzen will have advantages in price and power efficiency. Given the leaked results though, I think the Skylake X will pull ahead in multi core and single core performance. Will it be worth the price? Hard to say. Prices differences for CPUs at the higher end are never linear....

 

[ManyThreads]

Ryzen has Haswell IPC at best.

Consensus seems to be that Ryzen is 7% behind Kaby Lake IPC - not sure how that compares to Haswell IPC.

 

[Atari2600]

Consensus seems to be that Ryzen is 7% behind Kaby Lake IPC - not sure how that compares to Haswell IPC.

With SMT?

 

[wildhorse2k]

How exactly do you reckon that considering there's binning involved?

12C will share design with 14C, 16C and 18C. The best parts will go to 16C and 18C to get the highest clocks for them. 12C will get the worst. Which is why it is expected not to OC well. It is also questionable whether their IMC will be as good as 8/10C and support very high RAM clock speeds.

 

[coercitiv]

12C will share design with 14C, 16C and 18C. The best parts will go to 16C and 18C to get the highest clocks for them. 12C will get the worst. Which is why it is expected not to OC well.

According to Silicon Lottery.com:

As of 2/23/17, the top 27% of tested 7600Ks were able to hit 5.1GHz or greater. (1.408V CPU VCORE or less)
As of 2/22/17, the top 28% of tested 7700Ks were able to hit 5.1GHz or greater. (1.408V CPU VCORE or less)

As of 2/23/17, the top 13% of tested 7600Ks were able to hit 5.2GHz or greater. (1.44V CPU VCORE or less)
As of 2/22/17, the top 7% of tested 7700Ks were able to hit 5.2GHz or greater. (1.44V CPU VCORE or less)

Now tell me if you agree with the following statement:

7600K shares the same core with 7700K. The best parts go to 7700K to get the highest clocks for them. 7600K get the worst. Which is why it is expected not to OC well.

Looking forward to 10C "premium" $1000 SKU overtaking the 12C "trash" $1200 SKU through sheer overclocking.

 

[ManyThreads]

With SMT?

I believe it is with SMT, but same number of cores/threads. AMD themselves uses that 7% number a lot too when doing comparisons. I've also seen articles that only measured single core with similar results but I can't find them at the moment. I'd be happy to be pointed to some better analysis, but that is what I recall from when I was doing Ryzen research.

 

[wildhorse2k]

According to Silicon Lottery.com:

Now tell me if you agree with the following statement:

You know you can get weird results if your sample size isn't sufficient?

 

[Timmah!]

12C will share design with 14C, 16C and 18C. The best parts will go to 16C and 18C to get the highest clocks for them. 12C will get the worst. Which is why it is expected not to OC well. It is also questionable whether their IMC will be as good as 8/10C and support very high RAM clock speeds.

12C will share design with the 10C, not with 18C. It will be actually top, uncut part of its design "branch", thus the one i would want to buy

 

[Topweasel]

12C will share design with the 10C, not with 18C. It will be actually top, uncut part of its design "branch", thus the one i would want to buy

Yeah originally I think Skylake-X was going to top out at 10c, which would have been harvested parts. But 12 is the actual core count of that die.

 

[Shivansps]

10c is LCC and 12c MCC as far as i know.

 

[Ajay]

10c is LCC and 12c MCC as far as i know.

Correct. Thought I think the naming is LCC, HCC and XCC (Xtreme Core Count, I guess).

 

[Timmah!]

12C will share design with the 10C, not with 18C. It will be actually top, uncut part of its design "branch", thus the one i would want to buy

10c is LCC and 12c MCC as far as i know.

From the article on this very site:

The first half of the Skylake-X processor llineup follows this trend. Intel will launch four Skylake-X processors based on the LCC die, which for this platform will have a maximum of 12 cores. All processors will have hyperthreading.

@Ajay: Incorrect

 

[itsmydamnation]

I believe it is with SMT, but same number of cores/threads. AMD themselves uses that 7% number a lot too when doing comparisons. I've also seen articles that only measured single core with similar results but I can't find them at the moment. I'd be happy to be pointed to some better analysis, but that is what I recall from when I was doing Ryzen research.

If you want actual IPC taking the stilits data and removing the workloads where 256bit SIMD actually helps then Zen is 2% faster then haswell and 7% behind skylake.

 

[Ajay]

From the article on this very site:

The first half of the Skylake-X processor llineup follows this trend. Intel will launch four Skylake-X processors based on the LCC die, which for this platform will have a maximum of 12 cores. All processors will have hyperthreading.

@Ajay: Incorrect

From that same Article

All the processors in the enterprise line are typically made from these three silicon maps: a 10-core LCC silicon die

. So Ian must have been having a bad day. AFAIK, the 12 core CPU will come from the HCC die.

 

[Timmah!]

From that same Article . So Ian must have been having a bad day. AFAIK, the 12 core CPU will come from the HCC die.

Further down the article:

Still covering the LCC core designs, the final processor in this stack is the Core i9-7920X. This processor will be coming out later in the year, likely during the summer, but it will be a 12-core processor on the same LGA2066 socket for $1199 (retail ~$1279), being part of the $100/core mantra. We are told that Intel is still validating the frequencies of this CPU to find a good balance of performance and power, although we understand that it might be 165W rather than 140W, as Intel’s pre-briefing explained that the whole X299 motherboard set should be ready to support 165W processors.

I could see him doing a mistake, but if he writes twice in the same article that 12C is LCC, i have hard time believing its a typo or coincidence. But you never know. For now, i choose to believe 12C is LCC. We shall see who was right in August.

 

[Topweasel]

From that same Article . So Ian must have been having a bad day. AFAIK, the 12 core CPU will come from the HCC die.

I think the confusion is that the LCC die goes up to 12 cores per https://en.wikichip.org/wiki/intel/microarchitectures/skylake#Die . But it makes sense that Intel wouldn't want to waste all of it's fully functional dies on consumer products. It also makes sense that Intel would be fine using its worst MCC dies on a consumer CPU.

 

[TahoeDust]

Reviewers have had their review kits for a few days now...I can't believe we are not seeing any leaks. I'm sure some of you in the know must be hearing things. Spill it.

 

[Timmah!]

Well, OBR already started teasing how 7900x is massively superior to 6950x thanks to highly improved clocks and how its gonna come close to 12C Threadripper.

 

[JoeRambo]

12C is almost certainly LCC. It was planned to bring 12C long ago ( and they are delayed tiny bit to gather stockpile of good dies?). Now HCC is different beast, will require NUMA (on chip) support to extract most of performance and will probably have some performance gotchas ( like increased memory latency due to dual mem controllers etc <- Broadwell-EP HCC had those, maybe due to new uncore architecture Intel did away with that)

Intel already has incredibly lucrative market for HCC cores -> Xeons that absorb all cores good and bad, it's the Threadripper that forced them to bring them to HEDT market.

 

[Topweasel]

12C is almost certainly LCC. It was planned to bring 12C long ago ( and they are delayed tiny bit to gather stockpile of good dies?). Now HCC is different beast, will require NUMA (on chip) support to extract most of performance and will probably have some performance gotchas ( like increased memory latency due to dual mem controllers etc )

Intel already has incredibly lucrative market for HCC cores -> Xeons that absorb all cores good and bad, it's the Threadripper that forced them to bring them to HEDT market.

Slides as early as January had 10c max.

Edit: not for LCC but for Skylake-X. 12c Was a more recent addition. The 14+ options where last minute options with the mobo guys saying they haven't even seen any of these chips when asked at Computex.

 

[JoeRambo]

Slides as early as January had 10c max.

That was marketing driven, then Ryzen happened. Unless you want us to believe Intel did not add 2 more cores to Broadwell-EP LCC?

 

[Topweasel]

That was marketing driven, then Ryzen happened. Unless you want us to believe Intel did not add 2 more cores to Broadwell-EP LCC?

That's why I added my edit. LCC was always 12 cores. But I think Intel wanted to hold back on the 12c for servers. Then Ryzen made them decide to offer fully unlocked LCC. Then ThreadRipper became more than a rumor and Intel went into a panic.

 

[JoeRambo]

That's why I added my edit. LCC was always 12 cores. But I think Intel wanted to hold back on the 12c for servers. Then Ryzen made them decide to offer fully unlocked LCC. Then ThreadRipper became more than a rumor and Intel went into a panic.

Yeah it's pretty much that's what happened. Those KabyLake "hedt" CPUs make me also believe Intel was planning to do away with unlocked desktop line CPUs, but I guess we'll never know. Now that AMD has so good and fully unlocked core it is game over for market schemers at Intel.

 

[Sweepr]

Already told you months ago. Skylake-SP LCC = 10-core, MCC = 18C, HCC = 28C.​

 

[Topweasel]

Already told you months ago. Skylake-SP LCC = 10-core, MCC = 18C, HCC = 28C.​

https://en.wikichip.org/wiki/intel/microarchitectures/skylake#Die says otherwise.

12/18/28