Intel Skylake / Kaby Lake

[scannall]

Ahem. Care to comment now?

Launch prices:
7900X: $999 ($1 less than I predicted)
7820X: $599 ($19 more than I predicted)
7800X: $389 ($11 less than I predicted)
7740K: $339 ($11 less than I predicted)
7640K: $242 ($8 less than I predicted)

I don't mind being wrong. It happens from time to time. ;-) They certainly fiddled with the SKU's more than I thought they would. And the prices are lower than I expected. I know Ryzen is selling pretty well, perhaps better than I thought even? I have to wonder if it will hurt their Xeon sales for 1P servers though? Or if it's an over reaction? Regardless, there is still plenty of room for Ryzen in there. I think most of us like seeing a good scrap.

 

[mikk]

Plus 7900X is clocked to 3.3Ghz whereas 6950X is 3.0Ghz. At same clock speed single thread performance of 7900X is gonna be even less than the 15% claimed by intel.

3.0-3.5 Ghz for both according to the footnote. +15% IPC compared to Broadwell is quite good considering that Skylake-S to Haswell has improved with ~10%. It's a bigger improvement against a newer µarch. The new cache system seems to help there. With the big clock advantage of 7900x the difference between these two SKUs running stock clock should be huge for Singlethread, because the Turbo should run with full speed.

 

[Sweepr]

Found a new Coffee Lake-S engineer sample entry @ SiSoftware, confirms a 6C/12T model:

Genuine Intel(R) CPU 0000 @ 2.60GHz (6C 12T 2.59GHz, 6x 256kB L2, 9MB L3)
Intel(R) CoffeeLake Mobile Graphics Controller (192SP 24C 1.2GHz, 6.4GB) (OpenCL)

http://ranker.sisoftware.net/show_s...e0ddeccaa29faa8cf4c9f8debbdee3d3f586bb83&l=en

 

[dullard]

I don't mind being wrong. It happens from time to time. ;-) They certainly fiddled with the SKU's more than I thought they would. And the prices are lower than I expected. I know Ryzen is selling pretty well, perhaps better than I thought even? I have to wonder if it will hurt their Xeon sales for 1P servers though? Or if it's an over reaction? Regardless, there is still plenty of room for Ryzen in there. I think most of us like seeing a good scrap.

My post from the other thread:

Skylake: 7820X $599, 7900X $999 (post-Ryzen, Intel went back to the old pricing)
Broadwell: 6850K: $617, 6900K $1089 (pre-Ryzen was a bit higher than their normal pricing)
Haswell: 5930K: $583, 5960X: $999
Ivy Bridge: 4930K: $583, 4960X: $999
Sandy Bridge: 3930K: $583, 3960X: $999
Westmere: 980: $583, 990X: $999
Nehalem (2009 version): 950: $562, 975: $999
Nehalem (2008 version): 940: $562, 965: $999

Why on earth would you expect higher prices? Intel has been using the same price brackets more or less for the last decade.

 

[IntelUser2000]

The footnotes regarding SPECInt_2006 performance differential comparing the 6950X and 7900X doesn't inspire much confidence.

It definitely does not. Skylake E3 is 10% faster per clock than Haswell E3 in SpecCPU. If we assume all-core turbo is the same, maybe we are seeing 2-3% gain for the new cache arrangement. It's the best case though. Because there's such a big gap between base and Turbo for 7900X while the spread is much smaller on 6950X. It's 15% faster in single thread because SKL-X has clock speed advantage. That's with SKL-X using DDR4-2666 and BDW-E using DDR4-2133.

You thought diminishing returns in IPC was bad the past 10 years. Time to see the next 10. With Icelake we would be hard pressed to see 5%+ gains. Oh, how loud the cries would be on enthusiast forums when the gains for waiting 2-3 years become 2-3%.

 

[TheF34RChannel]

Gives you a month to take out a mortgage to afford one :\

Nah I'm good for clicking a buy button.

I presume there's also no word about whether or not current coolers will fit LGA 2066? It looks promising (and a good way to save some money), however. (Clicked this for a look: https://www.arctic.ac/eu_en/products/cooling/cpu.html?cpu_socket_intel=568)

 

[IEC]

My post from the other thread:

Skylake: 7820X $599, 7900X $999 (post-Ryzen, Intel went back to the old pricing)
Broadwell: 6850K: $617, 6900K $1089 (pre-Ryzen was a bit higher than their normal pricing)
Haswell: 6930K: $583, 5960X: $999
Ivy Bridge: 4930K: $583, 4960X: $999
Sandy Bridge: 3930K: $583, 3960X: $999
Westmere: 980: $583, 990X: $999
Nehalem (2009 version): 950: $562, 975: $999
Nehalem (2008 version): 940: $562, 965: $999

Why on earth would you expect higher prices? Intel has been using the same price brackets more or less for the last decade.

While your point relative to Broadwell pricing is correct, the argument is that Intel has been milking the consumer for more or less the last decade due to a distinct lack of competition from AMD. The crying foul is simply because the perception is that competition is back, yet there is no commensurate change in pricing brackets.

However, the increase in cores/threads at each price point is the real competitive change. Wasn't too long ago that quad core was the king and 8-core was the pinnacle of HEDT. Now we are looking at mainstream 6 and 8 cores from both and the pinnacle at HEDT being 16 and 18 core.

Competition is good for the consumer.

 

[EXCellR8]

i suppose $2000 for 18 Intel cores isn't terrible if you're a content creator... with A LOT of content to create.

 

[Bouowmx]

http://www.tomshardware.com/news/in...by-lake-x-x299-basin-falls-core-i9,34545.html

The LGA2066 socket is compatible with LGA2011 cooling solutions

 

[LTC8K6]

Found a new Coffee Lake-S engineer sample entry @ SiSoftware, confirms a 6C/12T model:

Genuine Intel(R) CPU 0000 @ 2.60GHz (6C 12T 2.59GHz, 6x 256kB L2, 9MB L3)
Intel(R) CoffeeLake Mobile Graphics Controller (192SP 24C 1.2GHz, 6.4GB) (OpenCL)

http://ranker.sisoftware.net/show_s...e0ddeccaa29faa8cf4c9f8debbdee3d3f586bb83&l=en

About time.
Mobile? 9mb L3? That was the same as the 6C/6T entry?

 

[VirtualLarry]

This makes me believe that judging by the laziness of game developers, they will likely again target the largest market which at max is running 4C-8T due to Intel dominating the industry in last 5 years. Even if I calculate starting today, 6C-12T will take 2 years easily to become mainstream and widespread enough assuming everyone starts buying 6C-12T with unlimited funds from today.

Dude, Ryzen 5 1600(X) is now entry-level for gaming rigs. 6C/12T minimum for a gaming rig.

 

[TheF34RChannel]

So the 7800X and 7820X are now i7 again? That makes little sense. Meh what's a moniker anyway

 

[dullard]

While your point relative to Broadwell pricing is correct, the argument is that Intel has been milking the consumer for more or less the last decade due to a distinct lack of competition from AMD. The crying foul is simply because the perception is that competition is back, yet there is no commensurate change in pricing brackets.

However, the increase in cores/threads at each price point is the real competitive change. Wasn't too long ago that quad core was the king and 8-core was the pinnacle of HEDT. Now we are looking at mainstream 6 and 8 cores from both and the pinnacle at HEDT being 16 and 18 core.

Competition is good for the consumer.

Competition is good. Yes. But a duopoly vs a monopoly has very little impact on pricing (especially at the low end). Having just two companies in this space is NOT competition. A duopoly has an impact on the high end, but the impact is far less than most people expect before they crunch the numbers. This is especially true if the companies can differentiate their products. The more that Intel can make the 7980XE sound different than the AMD Threadripper, the less the impact that Threadripper has on the 7980XE prices.

Optimal pricing strategy is a known practice. Having a competitor be better than it was in the past doesn't change the optimal pricing. All it changes is what a company should provide at those prices (if it can).

I'm glad that AMD is back. I'm glad that Intel is providing more than it did in the last HEDT chips. But that just doesn't impact the price levels nearly as much as people want.

 

[swilli89]

i suppose $2000 for 18 Intel cores isn't terrible if you're a content creator... with A LOT of content to create.

Well for content creation where Ryzen is close to Skylake's IPC a 16 core ThreadRipper (tm) for probably half or less the price is much better decision.

 

[Bouowmx]

About time.
Mobile? 9mb L3? That was the same as the 6C/6T entry?

Entry looks like a Core i7-"8700H" (like 7700HQ): mobile processor with same core and thread count as standard i7, but only 1.5 MB L3 cache per core.

 

[Sweepr]

Intel® Core™ X-series Processors Listed @ Intel ARK

Launch date listed as Q2-2017, so we can expect the chips before the end of June just like BenchLife predicted (June 26).

http://ark.intel.com/products/family/123588/Intel-Core-X-series-Processors#@desktop

The good element of this design is that a larger L2 will increase the hit-rate and decrease the miss-rate. Depending on the level of associativity (which has not been disclosed yet, at least not in the basic slide decks), a general rule I have heard is that a double of cache size decreases the miss rate by the sqrt(2), and is liable for a 3-5% IPC uplift in a regular workflow. Thus here’s a conundrum for you: if the L2 has a factor 2 better hit rate, leading to an 8-13% IPC increase, it’s not the same performance as Skylake-S. It may be the same microarchitecture outside the caches, but we get a situation where performance will differ.

Fundamental Realisation: Skylake-S IPC and Skylake-X IPC will be different.

http://www.anandtech.com/show/11464...ng-18core-hcc-silicon-to-consumers-for-1999/3

 

[swilli89]

Update: Intel neglects solder on hew high end skylake-x chips. https://www.overclock3d.net/news/cp...x_and_kaby_lake-x_cpus_will_not_be_soldered/1

 

[tamz_msc]

Ahem. Care to comment now?

Launch prices:
7900X: $999 ($1 less than I predicted)
7820X: $599 ($19 more than I predicted)
7800X: $389 ($11 less than I predicted)
7740K: $339 ($11 less than I predicted)
7640K: $242 ($8 less than I predicted)

Congratulations! Now where was HCC chips at the time of your prediction?

Oh wait...

 

[MrTeal]

Actually LCC is 10C, MCC is 18C and HCC is up to 28C. The LCC die should be relatively small.

According to AT, the 18C die is HCC now. There's been some renaming of the server dies.

The enterprise line from Intel has three designs for their silicon – a low core count, a high core count, and an extreme core count: LCC, HCC, and XCC respectively.[\quote]

 

[tamz_msc]

Found a new Coffee Lake-S engineer sample entry @ SiSoftware, confirms a 6C/12T model:

Genuine Intel(R) CPU 0000 @ 2.60GHz (6C 12T 2.59GHz, 6x 256kB L2, 9MB L3)
Intel(R) CoffeeLake Mobile Graphics Controller (192SP 24C 1.2GHz, 6.4GB) (OpenCL)

http://ranker.sisoftware.net/show_s...e0ddeccaa29faa8cf4c9f8debbdee3d3f586bb83&l=en

There goes the 12MB L3.

 

[dullard]

Congratulations! Now where was HCC chips at the time of your prediction?

Oh wait...

So, you are calling me out for not predicting prices of secret processors with no information about their existence, their specifications, nor is there any history of prices to base my estimations on?

This is a new category for Intel, they have created new price levels that they did not have before. The 7920X is cheaper than I would have expected. But, then again, I didn't know about the existence of the 7940X, 7960X, and 7980XE at the time which compete with the 7920X in the high end.

It looks like the 6950X at $1575 is being replaced by both the 7940X at $1399 (cheaper and lower 3rd digit in the model number) and the 7960X at $1699 (more expensive and higher 3rd digit in the model number).

 

[Sweepr]

There goes the 12MB L3.

Wrong according to mikk's source and the latest leak. And if both of them are correct you're not only wrong about this one, but also...

2. CFL-S Core i7 is powered by 6 core 12 threads, LLC 12MB. Compared to the previous year, the number of cores / threads and LLC capacities increased by 50%. The Rebalanced Smart Cache Hierarchy * applied here has not been applied compared to Skylake-X (which probably overlaps in a six-core segment, which will appear a month or two ahead of time).


(*: Skylake-X reduced the L2 cache capacity from 256KB to 1MB, instead of reducing the L3 cache capacity from 2.5MB to 1.375MB and changed the cache policy from inclusive to non-inclusive. I think it will help to improve the IPC. It is a feature that does not apply to KABI Lake-X.


3. CFL-S Core i5 powered by 4 cores 8 threads, LLC 8MB. Compared to the previous version, hyperthreading increased 100% of threads and increased LLC capacity by 33%.


4. CFL-S Core i3 is powered by 4 core 4 threads, LLC 6 / 8MB. Compared to the previous version, Hyper-Threading has been removed, but with 100% increase in core count, the number of threads is equal and LLC capacity has increased by 100%.

Core i7 having a 6C/12T flagship doesn't imply that Core i3/i5 would get more cores/threads as well.

If there is also an unlocked 4C/8T part based on the 4+2 die that is branded as an i7, that would take the place of the i7 7700K.

 

[krumme]

3.0-3.5 Ghz for both according to the footnote. +15% IPC compared to Broadwell is quite good considering that Skylake-S to Haswell has improved with ~10%. It's a bigger improvement against a newer µarch. The new cache system seems to help there. With the big clock advantage of 7900x the difference between these two SKUs running stock clock should be huge for Singlethread, because the Turbo should run with full speed.

Intel themselves states 15% and 10% performance uplift in specint 2006 for 7900x vs. 6950.
And thats for 2133 ram on the bwe and 2667 on the skl x. As seen in the footnotes too.

The 6950 is sold as 3.0 base
https://ark.intel.com/products/9445...ssor-Extreme-Edition-25M-Cache-up-to-3_50-GHz
The 7900x is sold as 3.3 base
http://ark.intel.com/products/123613/Intel-Core-i9-7900X-Processor-13_75M-Cache-up-to-4_30-GHz

15% IPC "plus" ? Its a stretch and then some. Did they regress for frequency then ?
We know where the performance is, and on a new process node i think they actually managed to get a slight bit of higher freq even for base. As is also seen by the spec sheets.

 

[tamz_msc]

Wrong according to mikk's source and the latest leak. And if both of them are correct you're not only wrong about this one, but also...

Or that your BS Benchlife die size chart referred to mobile stuff. All my posts are based on the assumption that 149 mm2 is the desktop part. If it is mobile then it's a different matter altogether.

 

[IEC]

Competition is good. Yes. But a duopoly vs a monopoly has very little impact on pricing (especially at the low end). Having just two companies in this space is NOT competition. A duopoly has an impact on the high end, but the impact is far less than most people expect before they crunch the numbers. This is especially true if the companies can differentiate their products. The more that Intel can make the 7980XE sound different than the AMD Threadripper, the less the impact that Threadripper has on the 7980XE prices.

Optimal pricing strategy is a known practice. Having a competitor be better than it was in the past doesn't change the optimal pricing. All it changes is what a company should provide at those prices (if it can).

I'm glad that AMD is back. I'm glad that Intel is providing more than it did in the last HEDT chips. But that just doesn't impact the price levels nearly as much as people want.

I agree with you. Definitional arguments aside, the "competition" (as perceived by the average consumer) will not lead to as much competition as some hope. AMD and Intel have different strengths and segmentation strategies. Ultimately, this means that there really isn't 1:1 overlap in a lot of cases, making it possible for both to preserve their desired margins.