Intel Skylake / Kaby Lake

[TheF34RChannel]

It's a new video. He got 5GHz with a "pre-tested" CPU running Prime95 with delid and temperatures hovered at 88 degrees. He says that average retail samples should do 4.7GHz after a delid.

Thanks! He's pushing his delid tool I suspect as those temps are conflicting with some other news around the web. We'll know Monday eh.

 

[mikk]

Investigating a bit more...

Coffee Lake-S = IMVP8

From Richtek Vcore Product Selection List for Intel and AMD platforms:

Skylake = IMVP8
Kaby Lake = IMVP8
Ice Lake = IMVP9
Tiger Lake = IMVP9


CPU VR = IMVP8/9

Next year's 300-series chipsets LGA 1151 motherboards might support next-generation Intel products. Now why do I say 2018 MBs and not this year's Z270/Z370 MBs? Let's go back to this DigiTimes article:

That would explain why Intels next generation code name is called ....lake, because the CPU pin count doesn't differ from Skylake and beyond.

 

[TheF34RChannel]

That would explain why Intels next generation code name is called ....lake, because the CPU pin count doesn't differ from Skylake and beyond.

Think it won't differ until what comes after Tiger Lake?

 

[jpiniero]

When is this myth ever going to die? It's like we are still stuck in 2010 or something. Current gen consoles forced developers to parallelize their engines using the task based parallelism model. That coupled with the increase in multithreaded aware drivers has actually given higher core count CPUs the advantage in gaming.

It only goes so far though. I suppose there might be some games which benefit from QCM or the extra L3. More likely the L3.

I am not expecting much of a benefit from Coffee Lake. The real benefit is helping out the i5, which does need 6C6T or 4C8T.

 

[Sweepr]

Quoting der8auer for those of you too lazy to watch the video.

Skylake-X i9 7900X OC to 5000 MHz using AIO ! (en)

From my point of view Skylake-X is a pretty impressive CPU. We now have have a 10-core that's a lot cheaper than the Broadwell-E before and it clocks a lot higher. Even with the stock Intel paste I was able to reach 4.8 GHz on the CPU (i9-7900X) using Corsair 280 (mm) AIO. So there is still some headroom, I guess if you use a custom new water cooling you might be able to hit 4.9 GHz on a very very good chip without delidding. So after this test I delidded the CPU and replaced the stock TIM with liquid metal and this helped me push the CPU with an AIO to 5 GHz.

So we had Broadwell-E before who could run like 4.3-4.4 GHz, it cost 1700€. Now we have Skylake-X which is a lot cheaper and we CAN push it to 5 GHz, so what's all this negative press about? I don't really understand it. So from my point of view this is a very impressive CPU, so we have very high single-thread performance on the 10-core and also high MT performance, which we didn't have before in a Broadwell-E (probably comparing to mainstream).

...I think you can maybe hit 5.1 GHz on a custom water cooling loop if you have a very very good chip. Keep in mind that this chip was already pre-tested so it's already a really good CPU. On average CPU you might be able to get 4.7-4.8 GHz if you delidded it and on a pre tested CPU 5 GHz should be possible.

Full video.

 

[Sweepr]

10nm++ is too late for Ice Lake. I think it'll either be 10nm or 10nm+.

You know, I wouldn't rule out a mid-2018 10nm Ice Lake launch, then mid-2019 10nm+ Tiger Lake launch, then mid-2020 for the next Lake on 10nm++. 7nm is 2021 for client anyway.

Alternatively, there could be a 10nm CNL, 10nm+ ICL, 10nm+ TGL, and then 10nm++ part.

I expect Ice Lake to be a 10nm+ product. They will need the enhanced version of the process to (on top of the new architecture) deliver competitive performance compared to aggressively clocked Kaby Lake Refresh & Coffee Lake parts @ 14nm++. Notice Intel refers to it as their 2nd generation 10nm processor. Considering the recent tape in, let's say April/May 17, I expect H2 2018 launch for client Ice Lake (possibly as soon as August/September). Will be interesting to see if mobile or desktop is first this time.

That would explain why Intels next generation code name is called ....lake, because the CPU pin count doesn't differ from Skylake and beyond.

Indeed. And it would give LGA 1151 300-series chipsets motherboard owners a really nice upgrade path. They could use one of their existing Skylake/Kaby Lake CPUs or buy a new Coffee Lake chip being launched in early 2018 (apparently new K models are also in the cards) and later on upgrade to (IMVP9) Ice Lake / Tiger Lake. Hope we get more substantial leaks to back this up in the future.

 

[Arachnotronic]

Quoting der8auer for those of you too lazy to watch the video.

Skylake-X i9 7900X OC to 5000 MHz using AIO ! (en)



Full video.

Solid. So we've got another Haswell-E -- similar frequency capability to the mainstream counterpart but with a lot more cores.

Also impressive that they kept similar frequency capability even with 4x the L2 cache without too much of an L2 cache latency increase (12 cycle -> 14 cycle).

This is a great piece of engineering from Intel.

 

[Arachnotronic]

That would explain why Intels next generation code name is called ....lake, because the CPU pin count doesn't differ from Skylake and beyond.

I think Intel is just naming everything indiscriminately after Lakes

 

[Jan Olšan]

If Ice Lake was 10++, where would 10+ go? Ice is the second 10nm generation, obviously 10nm++ should be on the third one (Tiger IIRC?). For Coffee, it was even the fourth gen, but that is probably just inconsistent naming.

 

[Arachnotronic]

If Ice Lake was 10++, where would 10+ go? Ice is the second 10nm generation, obviously 10nm++ should be on the third one (Tiger IIRC?). For Coffee, it was even the fourth gen, but that is probably just inconsistent naming.

Broadwell = 14nm, Skylake = 14nm, Kaby Lake = 14nm+, Coffee Lake = 14nm++.

 

[IndyColtsFan]

Pretty stoked! Pre-ordered a 7900K + X299 Strix board from ShopBLT

I'm REALLY anxious to hear your results, as that is more or less the same config I'm considering if I go with SKL-X. At this rate, I may end up building a SKL-X box and a Threadripper box!

 

[raghu78]

When is this myth ever going to die? It's like we are still stuck in 2010 or something. Current gen consoles forced developers to parallelize their engines using the task based parallelism model. That coupled with the increase in multithreaded aware drivers has actually given higher core count CPUs the advantage in gaming.

Observe:

Source

The whole point of owning an Intel HEDT CPU, is that you don't have to sacrifice gaming performance for productivity, or vice versa. You can have your cake and eat it too.

The majority of games still run faster on 7700k.

http://www.techspot.com/review/1348-amd-ryzen-gaming-performance/page5.html

6900k beating a 7700k is not the norm. I can bet a CFL 6C/12T will beat any Skylake-X for gaming. I do agree that Ryzen in its current version has a few quirks (core clocks especially) to work out before it appeals to gamers in the same way as Skylake-X. I think Zen on 14nm+ early next year could make this a much closer contest even in games. We are already seeing developers show that if they optimize for Ryzen it could be very competitive for gaming.

https://www.pcper.com/reviews/Graphics-Cards/Rise-Tomb-Raider-Gets-Ryzen-Performance-Update
https://www.pcper.com/reviews/Processors/Ashes-Singularity-Gets-Ryzen-Performance-Update

Due to AMD being absent from high end CPU segment for 5+ years and developers not even bothering to optimize for AMD CPU architectures the recovery in gaming will be a bit slower but it will be steady and grow with each AMD generation going forward.

 

[raghu78]

It's a new video. He got 5GHz with a "pre-tested" CPU running Prime95 with delid and temperatures hovered at 88 degrees. He says that average retail samples should do 4.7GHz after a delid.

so best case for Skylake-X with delidding and water cooling is 5Ghz. Avg silicon will be 4.8 Ghz most likely. Impressive feat. Kudos to Intel. CFL 6C/12T is going to set new records for OC on air/water with and without delidding. The high clocks will be the main reason for Skylake-X to be more attractive to gamers atleast till AMD can get higher clocks with Zen on 14nm+ next year. I still think AMD is going to price Thread ripper very aggressively and I would not be surprised to see 16C/32T at 3.4-3.5 Ghz base at USD 1000. Even after overclocking the 7900X at 4.8-5.0 Ghz won't have a chance against the flagship TR at 3.9-4.0 Ghz in multithreaded workloads. 25% clock headroom and 10% IPC cannot overcome 60% more cores and better SMT scaling. So for a lot of people who do serious work on their PC , TR will still be unbeatable in terms of value. I am looking forward to Pinnacle Ridge and I think AMD should be able to match what they achieved with Bristol Ridge with max clocks of 4.7 Ghz. The continuous improvements to AGESA should provide a more mature AM4 platform by early next year which will make Pinnacle Ridge vs Skylake-X and the upcoming Cascade Lake-X very interesting.




Talking of AMD in the skylake thread. Multiple mod warnings yet you're still posting about AMD in this thread.


esquared
Anandtech Forum Director.

 

[TheF34RChannel]

so best case for Skylake-X with delidding and water cooling is 5Ghz. Avg silicon will be 4.8 Ghz most likely. Impressive feat. Kudos to Intel. CFL 6C/12T is going to set new records for OC on air/water with and without delidding. The high clocks will be the main reason for Skylake-X to be more attractive to gamers atleast till AMD can get higher clocks with Zen on 14nm+ next year. I still think AMD is going to price Thread ripper very aggressively and I would not be surprised to see 16C/32T at 3.4-3.5 Ghz base at USD 1000. Even after overclocking the 7900X at 4.8-5.0 Ghz won't have a chance against the flagship TR at 3.9-4.0 Ghz in multithreaded workloads. 25% clock headroom and 10% IPC cannot overcome 60% more cores and better SMT scaling. So for a lot of people who do serious work on their PC , TR will still be unbeatable in terms of value. I am looking forward to Pinnacle Ridge and I think AMD should be able to match what they achieved with Bristol Ridge with max clocks of 4.7 Ghz. The continuous improvements to AGESA should provide a more mature AM4 platform by early next year which will make Pinnacle Ridge vs Skylake-X and the upcoming Cascade Lake-X very interesting.

A 16C winning at MT against a 10C isn't really surprising, comparing the two, however, is. We'll have to compare equal core counts. Makes more sense to me than lining up parts by price, but that's just me.

 

[raghu78]

A 16C winning at MT against a 10C isn't really surprising, comparing the two, however, is. We'll have to compare equal core counts. Makes more sense to me then lining up parts by price, but that's just me.

Why are we to compare equal core counts when the pricing is completely different. People with unlimited budget can go ahead and buy a 18 core Xeon and try and hit the best clocks with delidding and custom watercooling. But for everyone else who has a budget comparing performance for a given price is very relevant and perfectly logical. AMD will bring higher clocks next year with Pinnacle Ridge and at that time they might bump up the pricing a bit as they get even more competitive with higher max clocks and single thread performance. The golden rule of pricing is its dictated by competitive dynamics. If Ryzen 7 and Ryzen TR did not exist we would never have seen Intel bring 18 cores to Skylake-X. You would probably be paying USD 2000 for a 12C LCC die.






Want to talk about AMD? Start another thread.

Just not in this thread.



esquared
Anandtech Forum Director

 

[jpiniero]

Coffee Lake taped in ~November 2016 (https://forums.anandtech.com/thread...icroarchitecture.2489451/page-2#post-38580224) and we're going to see it available for purchase in August 2017 -- less than a year delta.

But that's only because Intel is rushing Coffee Lake out. Without the rushing out it would have been released in January. Realistically you are looking at Icelake U/Y mobile at the end of 2018 and the other models in early 2019. Then Tigerlake client would be 2019/2020 and Saphire Rapid in 2020/2021. Of course Intel hitting those dates is probally unrealistic.

 

[raghu78]

But that's only because Intel is rushing Coffee Lake out. Without the rushing out it would have been released in January. Realistically you are looking at Icelake U/Y mobile at the end of 2018 and the other models in early 2019. Then Tigerlake client would be 2019/2020 and Saphire Rapid in 2020/2021. Of course Intel hitting those dates is probally unrealistic.

Exactly. I would say Intel has a lot of proving to do with 10+ and 10++ . I think best case for Icelake is early Q4 2018 and worst case is late Q1 or early Q2 2019. I also think Intel's big challenge is going to be when they can get Icelake-SP out in volume given the massive die sizes of server chips. With Intel struggling to yield at Intel 10nm with tiny sub 100 sq mm die chips and having no meaningful production till early 2018 it remains to be seen how quickly they can get 10+ in good shape to ramp Icelake desktop/notebook and then Icelake server which will have much larger die sizes than 300 sq mm. If there are delays to 10+ that will hurt Intel's server business in a huge way as AMD has wisely moved to multiple die already and would be targetting 7nm Zen 2 based server chips with 48C/96T for mid-Q3 2019. I think multiple die with EMIB and server first will start at 10++ in 2020 (probably mid-2020).

 

[TheF34RChannel]

Why are we to compare equal core counts when the pricing is completely different. People with unlimited budget can go ahead and buy a 18 core Xeon and try and hit the best clocks with delidding and custom watercooling. But for everyone else who has a budget comparing performance for a given price is very relevant and perfectly logical. AMD will bring higher clocks next year with Pinnacle Ridge and at that time they might bump up the pricing a bit as they get even more competitive with higher max clocks and single thread performance. The golden rule of pricing is its dictated by competitive dynamics. If Ryzen 7 and Ryzen TR did not exist we would never have seen Intel bring 18 cores to Skylake-X. You would probably be paying USD 2000 for a 12C LCC die.

Because when I'm in the market for a 10C I'm not looking for a 16C? Only AMDs perspective is 'compare by price points' because that's all they usually have going for them. If you're looking for price/performance you're in the wrong thread I'm afraid; this is an Intel thread, this is not an AMD thread. This stuff is wearing me out. It's a rather simple equation: do you want price/performance? Buy AMD. Do you want raw power and pure performance? Buy Intel. And believe me, I don't mean to sound harsh and I too wish Intel were cheaper but this is the reality we live in.

 

[jpiniero]

Exactly. I would say Intel has a lot of proving to do with 10+ and 10++ .

True. I've mentioned this before but we still don't know what's wrong with 10 for sure though. Yields are likely crap but it could be that 14++ is just simply cheaper and Intel can get away with it.

As for server, based upon that earlier leak, it sort of makes sense now. Perhaps Intel will (try to at least) skip Icelake Server altogether and go to 10++ Tigerlake Server in mid 2019 and 7 Saphhire Rapid in mid 2020. This with EMIB.

 

[Arachnotronic]

Exactly. I would say Intel has a lot of proving to do with 10+ and 10++ . I think best case for Icelake is early Q4 2018 and worst case is late Q1 or early Q2 2019. I also think Intel's big challenge is going to be when they can get Icelake-SP out in volume given the massive die sizes of server chips. With Intel struggling to yield at Intel 10nm with tiny sub 100 sq mm die chips and having no meaningful production till early 2018 it remains to be seen how quickly they can get 10+ in good shape to ramp Icelake desktop/notebook and then Icelake server which will have much larger die sizes than 300 sq mm. If there are delays to 10+ that will hurt Intel's server business in a huge way as AMD has wisely moved to multiple die already and would be targetting 7nm Zen 2 based server chips with 48C/96T for mid-Q3 2019. I think multiple die with EMIB and server first will start at 10++ in 2020 (probably mid-2020).

So you think 10nm/10nm+ will be a huge struggle for Intel, but GloFo/TSMC 7nm-based products will be cake for AMD?








Talking of AMD in the skylake thread. Multiple mod warnings yet you're still posting about AMD in this thread.


esquared
Anandtech Forum Director./b]

 

[frozentundra123456]

The majority of games still run faster on 7700k.

http://www.techspot.com/review/1348-amd-ryzen-gaming-performance/page5.html

6900k beating a 7700k is not the norm. I can bet a CFL 6C/12T will beat any Skylake-X for gaming. I do agree that Ryzen in its current version has a few quirks (core clocks especially) to work out before it appeals to gamers in the same way as Skylake-X. I think Zen on 14nm+ early next year could make this a much closer contest even in games. We are already seeing developers show that if they optimize for Ryzen it could be very competitive for gaming.

https://www.pcper.com/reviews/Graphics-Cards/Rise-Tomb-Raider-Gets-Ryzen-Performance-Update
https://www.pcper.com/reviews/Processors/Ashes-Singularity-Gets-Ryzen-Performance-Update

Due to AMD being absent from high end CPU segment for 5+ years and developers not even bothering to optimize for AMD CPU architectures the recovery in gaming will be a bit slower but it will be steady and grow with each AMD generation going forward.

Yes, the majority of games do still run faster on a 7700k. That is why all the anti quad core hysteria is a bit (well actually a lot) overblown, at least for gaming. Multithreaded productivity is another situation of course, but that is kind of an apples to oranges comparison IMO. It is kind of like saying a truck will hall more bricks than a sports car. Ya think??

Bottom line, as always is buy the cpu based on your use case. Hell, for the vast majority of users, even an i5 or i3 is more than adequate.

 

[Ajay]

That's great news (der8auer OC)! SKX looks like it will perform very well for overclockers, despite the rants about no solder (including mine).

 

[richierich1212]

That's great news (der8auer OC)! SKX looks like it will perform very well for overclockers, despite the rants about no solder (including mine).

Intel provided him with a golden sample. Great marketing.

 

[plopke]

I was listening to linustech+jayztwocents and they mention delay of coffee lake to 2018 see youtube video , so what will we be getting then a kaby lake refresh this summer/after summer? 6core kaby lake refresh ? I am confused.

https://www.youtube.com/watch?v=XcNiKUm5j1s&t=3251s#

 

[IEC]

Intel provided him with a golden sample. Great marketing.

We'll see how the average retail chip clocks, but if it is as Roman (der8auer) says and the average under AIO cooling is 4.7-4.8GHz non-AVX, then it would be compelling for people interested in Intel HEDT. 90C+ on an AIO, open bench @ 4.8GHz with a 23°C ambient means you are looking at easily 200W+ at those clocks, though. I've asked him to clarify the stock all-core turbo and stock voltage so I can more accurately estimate TDP with an OC, but this should suffice for back of the napkin calculation:

4.8GHz vs 3.3GHz base @ 140W TDP --> +45% just from clockspeed, not including voltage. Even without a voltage increase, we're already at 200W+. With? Going to need a custom loop to even think about upper 4's, much less 5GHz.