Intel Coffee Lake Leaked News

[DXMember]

still on 14nm? Copy Lake is boooooring, where's Cannon Lake and Ice Lake?

[MageTank]

Just now, DXMember said:

still on 14nm? Copy Lake is boooooring, where's Cannon Lake and Ice Lake?

Mobile Cannonlake should be here by the end of this year, but we won't see desktop Cannonlake until 2018. Ice Lake will be 2019. Ice Lake is also rumored to bring FIVR back, so that should be interesting. 

[Taf the Ghost]

6 hours ago, Suika said:

I wouldn't say I was expecting an IPC boost from Coffee Lake, rather the existing increase to IPC that was seen moving to Broadwell and then to Skylake. That combined with higher frequency chips would tempt me to upgrade, so if what you're saying is true, I may jump to Z370 then.

The leak confirms it's the Kaby Lake design + 2 cores on the 6 core design. Which should mean that an OC'd K SKU could very much end up being the best gaming CPU for a rather long time. But that's only if you're getting it up to the 5 Ghz range under OC. But that's under OC, at stock this will actually be a regression from the 7700k in a number of scenarios.

 

The other thing is that the Z370 platform will support the Icelake CPUs coming in 2018, when Intel goes to their 10nm+ process. It's a uArch change (update? Still mostly Skylake? We're not sure, from what I can tell) not just the normal die-shrink situation.  (Cannonlake is the die shrink, but 10nm isn't an improvement over 14nm+ except for some mobile applications.)

[Taf the Ghost]

1 hour ago, MageTank said:

Mobile Cannonlake should be here by the end of this year, but we won't see desktop Cannonlake until 2018. Ice Lake will be 2019. Ice Lake is also rumored to bring FIVR back, so that should be interesting. 

Last information we have is no desktop Cannonlake. We'll see Icelake sometime late in 2018 on 10nm+ process. 

[That_PC_Kid]

1 hour ago, Taf the Ghost said:

(Cannonlake is the die shrink, but 10nm isn't an improvement over 14nm+ except for some mobile applications.)

The 10nm Cannonlake could be a lot better than 14nm it's just at the moment Intel can't make it any better because of complications. The 10nm test CPU's that they make do pretty much the same performance as the 14nm. The reason for shrinking a die is to make it more efficient. If 10nm is the same as 14nm efficiency wise then they haven't accomplished their goal.

[Jito463]

Reading all this back and forth made me laugh for a moment.  Just 20 years ago, we would have thought it stupid to be considering single-digit/low teens nanometer processes.  Anyone that suggested it would probably have been laughed out of the room.  Now we're casually throwing the numbers around, arguing about which is more efficient, heh.

 

Not related to the subject at all, just a random thought that popped into my head.  You may now resume your regularly scheduled discussion.

[Taf the Ghost]

46 minutes ago, Jito463 said:

Reading all this back and forth made me laugh for a moment.  Just 20 years ago, we would have thought it stupid to be considering single-digit/low teens nanometer processes.  Anyone that suggested it would probably have been laughed out of the room.  Now we're casually throwing the numbers around, arguing about which is more efficient, heh.

 

Not related to the subject at all, just a random thought that popped into my head.  You may now resume your regularly scheduled discussion.

Sub-10nm has been a topic, if rare, for a rather long time. The interesting question was always the steps to getting there. After the PC Boom of the 90s, node shrinks started to get really rapid. But it's still interesting there ended up being some surprisingly large steps after they figured out a few of the shrinks below 100nm. 

[leadeater]

2 hours ago, Jito463 said:

Reading all this back and forth made me laugh for a moment.  Just 20 years ago, we would have thought it stupid to be considering single-digit/low teens nanometer processes.  Anyone that suggested it would probably have been laughed out of the room.  Now we're casually throwing the numbers around, arguing about which is more efficient, heh.

 

Not related to the subject at all, just a random thought that popped into my head.  You may now resume your regularly scheduled discussion.

130nm is the way it is and will never get below 100nm 

 

It's kind of amazing to think that Pentium 3 started off as 250nm and Pentium 4 180nm

[Suika]

7 hours ago, MageTank said:

Most of the IPC gain we saw from Broadwell to Skylake, came from the memory subsystem swapping to DDR4. Most of the tests that showed a tangible gain in IPC, were tests that coincidentally scaled with memory bandwidth.

Interesting, so then ignoring the potential of mesh cache for the moment, that would mean Haswell-E, already running DDR4, would make to be even less of an upgrade? I'm sure there's more to it, but it's beginning to sound like I'm better off waiting even longer, or even trying to hit 4.6GHz.

 

Although, the idea of an NVMe compatible M.2 slot is still tempting, but I could probably swap motherboards for $50 if that's what I want.

[Taf the Ghost]

3 hours ago, Suika said:

Interesting, so then ignoring the potential of mesh cache for the moment, that would mean Haswell-E, already running DDR4, would make to be even less of an upgrade? I'm sure there's more to it, but it's beginning to sound like I'm better off waiting even longer, or even trying to hit 4.6GHz.

 

Although, the idea of an NVMe compatible M.2 slot is still tempting, but I could probably swap motherboards for $50 if that's what I want.

Well, the important detail is that Cores can only be as "good" as the memory sub-systems that provide them information. If a core is wasting 1/2 of its cycles doing nothing, you've built a CPU that's too big & too "wide" and thus costs too much to produce for the backing sub-systems. There's a lot of inter-related developments that are needed in CPU design, and Intel's actual advantage has normally been in the memory systems compared to AMD. 

[That_PC_Kid]

10 hours ago, DXMember said:

still on 14nm? Copy Lake is boooooring, where's Cannon Lake and Ice Lake?

Sadly Coffee Lake is still on 14nm. I'll probably upgrade to Ice Lake when that comes out.

[mr moose]

9 hours ago, Jito463 said:

Reading all this back and forth made me laugh for a moment.  Just 20 years ago, we would have thought it stupid to be considering single-digit/low teens nanometer processes.  Anyone that suggested it would probably have been laughed out of the room.  Now we're casually throwing the numbers around, arguing about which is more efficient, heh.

 

Not related to the subject at all, just a random thought that popped into my head.  You may now resume your regularly scheduled discussion.

Not a truer words has been said, this is exactly what I remind people of  when they tell me that wireless technology will never be as fast as wired.  Which force of omnipotent knowledge can be that sure of future technology?   

 

Hell I even remember people saying copper is copper, you won't get faster than coaxial BNC networking.  hahahahahahahahaha...

[tom_w141]

15 hours ago, That_PC_Kid said:

Intel CPUs have single threaded performance due to higher clock speeds, whilst the 5% higher IPC over Ryzen has little to do with it.

Fixed the above ^

15 hours ago, That_PC_Kid said:

The extra cores will be good for YouTubers, anyone who does editing, encryption, decryption, streaming, multi-tasking, encoding, programming, and many more.

Exactly which is why Ryzen is a good buy now.

 

14 hours ago, Bleedingyamato said:

Basibally they hit a brick wall for improvement and haven't figured out how to get around it yet?

They have hit the limit of the 14nm process.

[tom_w141]

14 hours ago, DXMember said:

Copy Lake

LMAO

[MageTank]

12 hours ago, Suika said:

Interesting, so then ignoring the potential of mesh cache for the moment, that would mean Haswell-E, already running DDR4, would make to be even less of an upgrade? I'm sure there's more to it, but it's beginning to sound like I'm better off waiting even longer, or even trying to hit 4.6GHz.

 

Although, the idea of an NVMe compatible M.2 slot is still tempting, but I could probably swap motherboards for $50 if that's what I want.

As @Taf the Ghost said, your memory and cache subsystem matter a lot when it comes to CPU speed in general. Even when discussing IPC. Haswell-E did have DDR4, but the memory controller was "weak" in comparison to Broadwell-E. With it being new at the time, it wasn't nearly as refined. That's why you see people struggling to hit 3200 and beyond on Haswell-E, yet have people hitting 3600 with relative ease on Broadwell-E. 

 

We must remember, IPC is not a static value that exists across the entire spectrum of your CPU. It will vary drastically depending on the instruction set/workload. Some instruction sets scale extremely well with memory bandwidth. AVX is the best example of this, as you can see the difference to which it scales, by comparing DDR4 bandwidth, to your cache bandwidth. Your cache bandwidth is magnitudes faster, and that is why Small FFT Prime95 (the length's that fit entirely into cache) gets so much hotter than larger FFT lengths. It allows you to use a lot more bandwidth to process those AVX workloads. Your GFLOPS (using an AVX2 program like Linpack/LinX) improve in proportion to memory/cache clock, on a much larger scale than raw CPU clocks. It's why IPC in regards to just CPU's themselves, are deceiving. Even comparing say, AMD to Intel, is different from a "pure CPU" standpoint, given how vastly different their IMC's are. Intel has a clear and strong advantage in the IMC department, even after Ryzen's latest AGESA updates.

 

I can say that, the difference in performance going from Haswell-E to Broadwell-E was far less pronounced than it was going from Haswell to Skylake. Sadly, we cannot make apples to apples comparisons for going from Haswell-E to Skylake-E, as the cache restructure has certainly changed benching performance. The amount of GFLOPS one can pull in Linpack with the larger L2 cache is absolutely insane. Not to mention it's memory controller is extremely potent, being able to run quad channel DDR4 4000mhz+ depending on the board. It would be nothing to see people hit 110GB/s bandwidth, with 45ns latency (if tweaked correctly). In comparison, we only hit about 60GB/s bandwidth, 35ns latency with fully tweaked DDR4 on consumer Skylake/Kaby Lake.

[juri-han]

alot of people keep pushing the idea of zen2 by q4 of 2018 if thats true and you bought a coffee lake ....

[Jito463]

32 minutes ago, juri-han said:

alot of people keep pushing the idea of zen2 by q4 of 2018 if thats true and you bought a coffee lake ....

I don't know about a coffee lake, but I wouldn't mind a lake of coffee.

 

[That_PC_Kid]

6 hours ago, tom_w141 said:

Exactly which is why Ryzen is a good buy now.

That is true but once Coffee Lake comes out Ryzen will be the budget option.

[That_PC_Kid]

1 hour ago, juri-han said:

alot of people keep pushing the idea of zen2 by q4 of 2018 if thats true and you bought a coffee lake ....

But if Zen 2 were to come out after Coffee Lake would it be all that bad? You still have better IPC with Intel.

[MageTank]

7 minutes ago, That_PC_Kid said:

But if Zen 2 were to come out after Coffee Lake would it be all that bad? You still have better IPC with Intel.

Both will have IPC that are within spitting distance of each other from a strictly gaming standpoint. It's the potentially higher clock speeds, and superior IMC that wins me over on Intel's side. I can't stand the thought of not having access to specific timings to make any kit of ram compatible with my CPU. The fact that you are forced to use specific memory kits on Ryzen in order to get the most out of your CPU, annoys me. Especially given the fact that manual tweaking will only get you so far, before you are limited by the lack of 2DPC and multi-rank timings.

 

If Zen2 brings us a stronger IMC, I'd certainly be down to give it a shot. If it fails to do so, and gives us another crippled IMC like we have on the current Zen architecture, then I just don't see myself using it. 

[juri-han]

49 minutes ago, MageTank said:

Both will have IPC that are within spitting distance of each other from a strictly gaming standpoint. It's the potentially higher clock speeds, and superior IMC that wins me over on Intel's side. I can't stand the thought of not having access to specific timings to make any kit of ram compatible with my CPU. The fact that you are forced to use specific memory kits on Ryzen in order to get the most out of your CPU, annoys me. Especially given the fact that manual tweaking will only get you so far, before you are limited by the lack of 2DPC and multi-rank timings.

 

If Zen2 brings us a stronger IMC, I'd certainly be down to give it a shot. If it fails to do so, and gives us another crippled IMC like we have on the current Zen architecture, then I just don't see myself using it. 

think that would be one of the first things they would try and work on, i assumed zen+ would be that plus a minor clock speed bump but if they skip zen+ and go right for the 7nm zen2 you would get ram issues sorted out on top of a major boost the clock speeds

[Taf the Ghost]

58 minutes ago, MageTank said:

Both will have IPC that are within spitting distance of each other from a strictly gaming standpoint. It's the potentially higher clock speeds, and superior IMC that wins me over on Intel's side. I can't stand the thought of not having access to specific timings to make any kit of ram compatible with my CPU. The fact that you are forced to use specific memory kits on Ryzen in order to get the most out of your CPU, annoys me. Especially given the fact that manual tweaking will only get you so far, before you are limited by the lack of 2DPC and multi-rank timings.

 

If Zen2 brings us a stronger IMC, I'd certainly be down to give it a shot. If it fails to do so, and gives us another crippled IMC like we have on the current Zen architecture, then I just don't see myself using it. 

Hopefully we get a better IMC in even in the Zen refresh next year, though sub-timings are going to be a little wonkier on Zen going forward because of the interaction with the Infinity Fabric.

 

Q2`2019 is probably the latest we'd see Zen2, but I'm sure AMD hopes for Q4`2018, which would be probably 12 months after Tapeout. (The kind of hilarious part is that consumer Ryzen CPUs are pretty much the Engineering Samples, since the rollout was that good.)

[MageTank]

1 minute ago, Taf the Ghost said:

Hopefully we get a better IMC in even in the Zen refresh next year, though sub-timings are going to be a little wonkier on Zen going forward because of the interaction with the Infinity Fabric.

 

Q2`2019 is probably the latest we'd see Zen2, but I'm sure AMD hopes for Q4`2018, which would be probably 12 months after Tapeout. (The kind of hilarious part is that consumer Ryzen CPUs are pretty much the Engineering Samples, since the rollout was that good.)

We have access to most tertiary timings, but the lack of tREFI, 2DPC timings, and multi rank timings really hurt latency and compatibility. My 3200 C14 kit can do 3800 C14 on my 7700k, but can barely do 3200 C16 on Ryzen. That's with a ton of tweaking to even get it to post (not even counting stress tests). The issue being, they are 16GB DIMM's, and are dual rank as a result. Not being able to tweak multi-rank tertiary timings to ease stress off the IMC, you are essentially going in blind, praying that it works.

 

I did see a way to completely disable rank interleaving, but at that point, I am throwing away all performance benefits of going with a multi-rank kit in the first place. The ironic part is, AMD completely acknowledges the impact of multi-rank kits from a performance standpoint: https://community.amd.com/community/gaming/blog/2017/07/14/memory-oc-showdown-frequency-vs-memory-timings

[Demonking]

So basically Intel is doing what AMD doing here.

[tom_w141]

2 hours ago, That_PC_Kid said:

That is true but once Coffee Lake comes out Ryzen will be the budget option.

Costing less than a 4 core/4 thread i5 it is already the budget option and the best option for a mid range builder or gamer/streamer. Also the 7700k only helps you if you have a 144Hz refresh rate monitor anyway. You can't even see a difference between 120 and 140fps (and that is with the difference exaggerated. irl its closer). FPS has diminishing returns above 100. 30-60 is a stark difference, 60-100 is also a large difference, 100-144 not so much.