Intel Skylake / Kaby Lake

[krumme]

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

For the lazy:
" Up to 10 percent faster multithread performance1 over previous generation  Up to 15 percent faster single-thread performance2 over previous generation"
...footnotes:
"1 Based on SPEC*int_rate_base2006 (n copy) comparing Intel® Core™ i9-7900X X-series processor (10C20T) vs. Intel® Core™ i7- 6950X Processor (10C/20T). 2 Based on SPEC*int_rate_base2006 (1 copy) comparing Intel® Core™ i9-7900X X-series processor (10C20T) vs. Intel® Core™ i7- 6950X Processor (10C/20T)"

Its a small uplift, but remember it comes at huge cost savings 1700->1200. Lets hope for reasonably mb prices.

 

[tamz_msc]

" Up to 10 percent faster multithread performance1 over previous generation  Up to 15 percent faster single-thread performance2 over previous generation"
...footnotes:
"1 Based on SPEC*int_rate_base2006 (n copy) comparing Intel® Core™ i9-7900X X-series processor (10C20T) vs. Intel® Core™ i7- 6950X Processor (10C/20T). 2 Based on SPEC*int_rate_base2006 (1 copy) comparing Intel® Core™ i9-7900X X-series processor (10C20T) vs. Intel® Core™ i7- 6950X Processor (10C/20T)"

There is a 7% uncertainty in those results for it not being final silicon. So right now, as of this moment IPC is basically the same.

 

[WingZero30]

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.

 

[gx_saurav]

Great reasoning. Now apply it to your upgrade from i5 to i7 only a few months before i7 7700K will essentially become the i5 8600K (if not worse). What do you think happens to 7700K prices in Aug. when Intel ships cheaper i5 with same computing potential?

Actually what do you think happens to 7700K pricing once 6c/12t SKL-X hits the market at $389 MSRP? When I said 6c-8c is the "new meta" for gaming I did not mean one should automatically buy one. The value game gets a reset though, and people riding the R5 and i5 trains will get a lot more value than you think.

I agree with your point that a 7700K will soon become a i5 8600K or something on that line. New buyers will get more value as now a 4C-8T CPU will be even more affordable.

This means 4C-8T CPUs will sell a lot due to low costs which will make 4C-8T CPU the preferred target for game developers. If game developers utilize more cores then that is always better. But 4C-8T will still hold its own for few years to come. Considering I know how lazy developers are when it comes to optimizing their old code instead of writing new one, this will take few years to saturate. Hence majority of users will be just fine with low cost 4C-8T gaming machines.

Here in India prices don't reduce generally. Compared to January when I purchased a Core i5 7500, a Core i7 7700k reduced only $ 15 in cost even with Ryzen, last week.

For every 10 PCs sold, 6 use Core i3 with onboard graphics and are meant for everyday computing PC or for learning of kids. 3 are Core i5 or Ryzen with a NVidia/AMD GPU and One is Core i7 with GPU. Hardly anyone is willing pay the price of Ryzen 1700/1800 here. People think that if they are going to pay that much money, they might as well buy a Core i7. Ryzen 1500 & 1600 is selling like hot cakes here.

 

[krumme]

There is a 7% uncertainty in those results for it not being final silicon. So right now, as of this moment IPC is basically the same.

hmm It says "+/- 5%"
But yeaa you are right it does look like too small an uplift and basicly the same. What is happening?

 

[tamz_msc]

I agree with your point that a 7700K will soon become a i5 8600K or something on that line. New buyers will get more value as now a 4C-8T CPU will be even more affordable.

This means that a 4C-8T CPUs will sell a lot due to low costs which will make 4C-8T CPU the preferred target for game developers. If game developers utilize more cores then that is always better. But 4C-8T will still hold its own for few years to come.

Here in India prices don't reduce generally. Compared to January when I purchased a Core i5 7500, a Core i7 7700k reduced only $ 15 in cost even with Ryzen, last week.

For every 10 PCs sold, 6 use Core i3 with onboard graphics and are meant for everyday computing PC or for learning of kids. 3 are Core i5 or Ryzen with a NVidia/AMD GPU and One is Core i7 with GPU. Hardly anyone is willing pay the price of Ryzen 1700/1800 here. People think that if they are going to pay that much money, they might as well buy a Core i7. Ryzen 1500 & 1600 is selling like hot cakes here.

That's true, the Ryzen 5 1600 is basically this generation's 6600K/7600K in India. Unless Coffee Lake 6C has a variant at the same price point as the 1600, AMD has this segment pretty much covered.

 

[tamz_msc]

hmm It says "+/- 5%"
But yeaa you are right it does look like too small an uplift and basicly the same. What is happening?

 

[krumme]

It says +/- 5% in the factsheet - difficult to know what is fact then
https://newsroom.intel.com/wp-content/uploads/sites/11/2017/05/Intel-Core-X-series-Fact-Sheet.pdf

 

[gx_saurav]

That's true, the Ryzen 5 1600 is basically this generation's 6600K/7600K in India.

I upgraded to Core i7 7700 because it was the only 4C-8T CPU in 65W which I could use on my existing motherboard. 7700K is 95W & more costly with only 3% better performance in games compared to 7700. Ryzen would have needed a new motherboard all together which would again increase the cost.

Unless Coffee Lake 6C has a variant at the same price point as the 1600, AMD has this segment pretty much covered.

If Intel tries to give 6C-12T in 65W then clock speed will likely need to be reduced to match the TDP target of 65W.

In India, people prefer to buy bigger numbers so a layman buys Ryzen 5 with 6C-12T instead of Core i5 with 4C-4T. In future a person will again prefer to buy a Ryzen 5 with 6C-12T instead of Core i5 with 4C-8T (as this has less core ) or Core i7 CFL-S with 6C-12T (as this is more costly due to being an Intel CPU).

 

[imported_jjj]

Intel made the claim that the 8th series is over 30% faster than Kaby Lake. Managed to find a slide where the disclaimer is visible.
The comparison is in SYSmark 2014 v1.5 and they compare i7-7500U (dual core with base at 2.7GHz and turbo up to 3.5GHz at 15W TDP) with 2x4GB DDR4 2133 vs a 4C/8T I7 at 15W TDP with turbo up to 4GHz,paired with 2x4GB DDR4 2400. There is some more info in that disclaimer but can't quite make out what it says.

 

[Sweepr]

Intel launches Core-X series with up to 18-cores for 1999 USD

Core i9-7980XE
18C/36T
$1999

Core i9-7960X
16C/32T
$1699

Core i9-7940X
14C/28T
$1399

Core i9-7920X
12C/24T
$1199

Core i9-7900X
10C/20T
$999

Core i9-7820X
8C/16T
$599

Core i9-7800X
6C/12T
$389

Core i7-7740X
4C/8T
$369

Core i5-7640X
4C/4T
$242

https://videocardz.com/70005/intel-launches-core-x-series-with-up-to-18-cores-for-1999-usd

The cat is out of the bag!

- Ridiculously fast 18C AVX-512 enabled Core i9-7980XE for nearly the same price as 10C Core i7-6950X (Teraflop CPU)
- 10C down to $999 with Core i9-7900X, up to 4.5 GHz
- 8C down to $599 with Core i9-7820X, up to 4.5 GHz
- 6C Core i9-7800X as the new base model for Skylake-X
- Rebalanced cache hierarchy, which Intel claims increased performance (reviews on June 11-12!)
- New overclocking features including AVX-512 ratio offset
- Core i9 brand and Socket R4 (LGA 2066) confirmed

 

[tamz_msc]

Intel made the claim that the 8th series is over 30% faster than Kaby Lake. Managed to find a slide where the disclaimer is visible.
The comparison is in SYSmark 2014 v1.5 and they compare i7-7500U (dual core with base at 2.7GHz and turbo up to 3.5GHz at 15W TDP) with 2x4GB DDR4 2133 vs a 4C/8T I7 at 15W TDP with turbo up to 4GHz,paired with 2x4GB DDR4 2400. There is some more info in that disclaimer but can't quite make out what it says.

I squinted my eyes reading the extreme right corner, seems to say something along the line of +/-7% margin of error. But as krumme posted it is +/-5% in their newsdesk pdf.

 

[RichUK]

$1199 for the 12 core - Dayum, that's a good price point, given current pricing.

 

[imported_jjj]

I squinted my eyes reading the extreme right corner, seems to say something along the line of +/-7% margin of error. But as krumme posted it is +/-5% in their newsdesk pdf.

Yeah but before that there is something about power policy, might be more relevant. I guess the GPU would matter in this benchmark too and ofc pricing.

 

[csbin]

7900X cpuz benchmark

http://www.bilibili.com/video/av10946196/

 

[krumme]

For comparison, 6.2Ghz 4770k is matched by 5.7Ghz 7700k. There is actually lower IPC differential between HSW-E and SKL-X than between HSW-S and SKL-S.

So, please, quit BSing.

Anandtech found the difference to be 2.7% IPC uplift from broadwell to skl 6700 series on a broad productivity workload they have used for many years. Done on ddr4 2400 with loose timings:
http://www.anandtech.com/show/9483/intel-skylake-review-6700k-6600k-ddr4-ddr3-ipc-6th-generation/9
Here we are talking skl x on 2667 and bwe on 2400 so the miniscule IPC uplift is easily lost in error measuring.

 

[tamz_msc]

7900X cpuz benchmark

http://www.bilibili.com/video/av10946196/

4.4GHz at 1.764V?
Barely any improvement in MT performance with that OC.

 

[imported_jjj]

Intel X299 Launch: 8 World Records and 23 Global First Places!

 

[JoeRambo]

7900X cpuz benchmark

http://www.bilibili.com/video/av10946196/

That is a nice score for 4.4Ghz clock. For comparison 5Ghz Skylake-S has ~the same SP score.

But then again CPU-Z is notoriuos for loving L2 cache, Ryzen with 512KB of L2 was punching above its clock in CPU-Z.

Barely any improvement in MT performance with that OC.

I would strongly suggest anyone with certain agendas to look at "MP scaling factor". It is only 9.87 for ten core. Feel free to drum up BS about "Intel's MP scaling broken, cause Broadwell-EP had >11.7 for ten cores.

 

[lolfail9001]

Anandtech found the difference to be 2.7% IPC uplift from broadwell to skl 6700 series on a broad productivity workload they have used for many years. Done on ddr4 2400 with loose timings:
http://www.anandtech.com/show/9483/intel-skylake-review-6700k-6600k-ddr4-ddr3-ipc-6th-generation/9
Here we are talking skl x on 2667 and bwe on 2400 so the miniscule IPC uplift is easily lost in error measuring.

Anandtech's review was flawed by comparing 5775C to 4770k and 6700k, you know. And well, if you skip it, then measurements line-up.

4.4GHz at 1.764V?
Barely any improvement in MT performance with that OC.

Fairly positive voltage readings are being retarded, it happens on engineering samples, you know.

 

[asendra]

Intel has really felt the AMD competition, who would have thought!

Looking at the line up, the i7-7820X looks to be where intel has really pushed to be as competitive as possible with AMD. 100$ more than the 1800x, but better clocks, better IPC, more PCI lanes etc. Not that bad of a deal.

Only problem is the X299 platform should be quite a bit more expensive..

 

[tamz_msc]

That is a nice score for 4.4Ghz clock. For comparison 5Ghz Skylake-S has ~the same SP score.

But then again CPU-Z is notoriuos for loving L2 cache, Ryzen with 512KB of L2 was punching above its clock in CPU-Z.

It was less due to L2 and more due to the 25% wider uop cache. They fixed the benchmark in a recent update.

Fairly positive voltage readings are being retarded, it happens on engineering samples, you know.

Still that MT score is nothing to write home about. 6950X can easily do 4.3GHz.

 

[Sweepr]

That is a nice score for 4.4Ghz clock. For comparison 5Ghz Skylake-S has ~the same SP score.

If only reviews were out today. I can't wait to see how the rebalanced cache affects CPU limited gaming. We can confirm the specs I leaked here are real as well, up to 4.5 GHz Turbo.

 

[tamz_msc]

I would strongly suggest anyone with certain agendas to look at "MP scaling factor". It is only 9.87 for ten core. Feel free to drum up BS about "Intel's MP scaling broken, cause Broadwell-EP had >11.7 for ten cores.

I would strongly suggest at people with outdated knowledge refrain from posting about possible conspiracies when it is known that MT ratio is affected by background applications.

 

[itsmydamnation]

Anyone going to guess the die size based of DDR4 memory interface relative to broadwell-E?