Discussion Intel current and future Lakes & Rapids thread

[liahos1]

Ian has his article about Tigerlake's die.

Damn, so the Xe iGPU is even more area efficient than I thought. Seems to be under 50mm2. Actually, I'm getting 46-48mm2. So that's something like a 15% increase. In comparison, Gen 11 is 41-43mm2(41mm2 according to wikichip).

Ian is getting 146mm2 for the whole die, which is actually what I got. Out of the ~24mm2 increase, 8mm is due to Willow Cove cores. The increase is pretty much as I said back in post #3831

Looks like Intel arranged Tigerlake in a way that's easier to add cores.

Addendum: At 38mm2 vs 48mm2, the CPU portion is almost as large as the GPU portion. So much for the iGPU taking up majority of die space. This is because the iGPU scales better.

Thats pretty good no? 47mm^2 = 2tflops.

If intel put out a dice similar in size to a 5700xt (251mm ^2) that would be a 10.7tflop gpu before the benefit of dedicated TDP.

 

[IntelUser2000]

Thats pretty good no? 47mm^2 = 2tflops.

If intel put out a dice similar in size to a 5700xt (251mm ^2) that would be a 10.7tflop gpu before the benefit of dedicated TDP.

Tigerlake has 96EUs with what seems like a 1.1GHz clock. That's 1.7TFlops. Surely in the dGPU format the clocks can be higher.

The big advance for Intel is that they are doing this without moving to a new process. Yes, its very good.

Remember, you need to add the display controller, the GDDR6/HBM2 interface, and the PCI Express controller to get the proper die size. Those parts also happen to be the ones that don't scale well with finer processes.

 

[liahos1]

anyone want to take a wild guess at tflops and die size for the supposed 512 eu part haha

 

[IntelUser2000]

I count 37 Icelake based laptops so far. Not all are available in North America, but vast majority are. About 10 models are super cheap ones.

anyone want to take a wild guess at tflops and die size for the supposed 512 eu part haha

The DG2 right?

I have speculated DG may mean Development Graphics, not Discrete Graphics.

According to earlier leaks, Arctic Sound exists(ATS). Ashraf said it'll be the first dGPU. He also said it'll be connected to Intel CPUs via EMIB.

So this year's dGPU may not be discrete more than Kaby-G is.

 

[liahos1]

didn't ashraf also say Raja wanted to "enter the market with a bang" - When i talked to Raja at the analyst day he said they would focus on entry to enthusiast in discrete gpus first and then work their way up the stack in successive years.

He said they would focus on top end performance for DCG gpus. This product is coming out in 2020.

Which goes back to my earlier question? What does a 512 eu gpu look like?

 

[IntelUser2000]

didn't ashraf also say Raja wanted to "enter the market with a bang" - When i talked to him at the analyst day he said they would focus on entry to enthusiast in discrete gpus first and then work their way up the stack in successive years.

I don't know. 512EU doesn't seem to be anything more than mid-end, especially if Nvidia's 7nm client GPUs are coming this year. And its more than just Flops.

They have a year, and its not a long time for silicon development. Looking at DG1 it still seems to have quite a ways to go. Angle of attack starting from the iGPU makes sense since their stronghold is iGPU. At least to me.

Which goes back to my earlier question? What does a 512 eu gpu look like?

8.2TFlops @ 1GHz.

 

[Richie Rich]

I'm sure you knew those numbers too before you looked them up . I already changed lucrative to margins. You either don't understand the difference or conveniently ignored that. If anyone should be ashamed, it should be you because of your ignorance and misinformation.

I own stocks of Apple and AMD so yes, before I invest my money I like to go through the numbers first. Anyway you wrote that server market is more lucrative, which was lie because smartphone market is 7x bigger. Then you changed that to margins but you never wrote numbers. So here we go:
- Intel Profit Margin 27%
- Apple Profit Margin 21%

So Intel has higher margins by 6 % points, however at 7x bigger smartphone market Apple gains 5.5x more money than Intel from servers. So the difference in margins doesn't compensate the market size difference (assuming 100% server market). If Apple would aim for 20% server market share that number would rise from 5.5x somewhere around 20-25x. So yes, Apple and its current management is not motivated to enter server market for just 5% more money (crumbs). Another thing to consider is growth rate. Smartphones and mobile gaming grows much faster so they invest here (and effectively multiplying their future revenue). In other words an investment in slower growing server market would generate relative loss for Apple. However not for Nuvia engineers, they went from employees with normal salary to company owners, fighting for billions.

You claim 82% as if it is fact. No one believes that but you. Also, Jim Keller is a rock star, but he is not a magician. He needs a team of talent and vision. No one man creates a CPU. Does Intel have that? I think so. Their roadmap looks promising if they can get their manufacturing figured out.

It's not about believes but about measured numbers. That +82% IPC advantage is based on SPECint results here at Anandtech. CPU frequencies are known too so everybody can calculate that (A13 runs at 2.65, 9900K at 5 GHz, 3950X at 4.6-4.65 GHz). Please feel free come up with your interpretation. I hope you know how to calculate percentage, do you?

Intel has a huge deficit especially in uarch development. Ice Lake brings only +18% IPC over Coffie Lake and about +25% over 2015 Sky Lake (after 5 years). That's poor 5% IPC increase per year. Another 5-8% with Tiger Lake (Willow Cove) looks also very bad. Cortex A77 jumped +25% from its A76 predecessor in one single year (A77 matched Coffie Lake IPC so we are talking about improvement of very powerfull 4xALU OoO ARM core, not about some slow 2xALU in-order junk). This year will be unveiled A78 with another +25% increase what means generic Cortex core will have higher IPC than Intel for the first time. At time of Golden Cove will enter market in 2022 they will do two another +25% jumps (+56% IPC total). And we know from Apple A13 analysis that there is 82% IPC potential hanging low enough to be picked so it's realistic and doable. And that's not good for Intel future even Golden Cove would bring +20% IPC jump. That's not enough still.


You've already been warned once today about dragging threads off topic with your Apple speculation, and here you're doing it again just a few hours later. Stop.

AT Moderator ElFenix

 

[Thunder 57]

I hope you know how to calculate percentage, do you?

Intel has a huge deficit especially in uarch development. Ice Lake brings only +18% IPC over Coffie Lake and about +25% over 2015 Sky Lake (after 5 years). That's poor 5% IPC increase per year. Another 5-8% with Tiger Lake (Willow Cove) looks also very bad. Cortex A77 jumped +25% from its A76 predecessor in one single year (A77 matched Coffie Lake IPC so we are talking about improvement of very powerfull 4xALU OoO ARM core, not about some slow 2xALU in-order junk). This year will be unveiled A78 with another +25% increase what means generic Cortex core will have higher IPC than Intel for the first time. At time of Golden Cove will enter market in 2022 they will do two another +25% jumps (+56% IPC total). And we know from Apple A13 analysis that there is 82% IPC potential hanging low enough to be picked so it's realistic and doable. And that's not good for Intel future even Golden Cove would bring +20% IPC jump. That's not enough still.


You've already been warned once today about dragging threads off topic with your Apple speculation, and here you're doing it again just a few hours later. Stop.

AT Moderator ElFenix

. If you want to read what I said about AMD/Apple to it in that thread. This thread is about Intel.

Your number games are stupid, too. Intel didn't increase IPC due to manufacturing, not because they can't design CPU's. You just happen to know how unreleased CPU's are going to perform IPC wise? That's bloody impressive! You said Zen 3 could be 40-50%, but Tiger Lake will be 5-8%. When you make up numbers to suit your narrative, it just makes you look foolish. Tiger Lake s looking more impressive. We might finally see some serious competition between it and Zen 3.

 

[amdfan111]

Tiger Lake, Willow Cove, whatever you want to call it. It could be a great product if Intel cared enough to make desktop CPUs, but all they care about is laptop. Fact is, only AMD makes a desktop-first CPU (RyZen, lots of cores, no iGPU crap). If Tiger Lake is so great, why won't Intel make something more substantial than some 2-core, 4-core rubbish? What AMD 4800H has shown is that even a tiny laptop chip can fit 8 cores, but Intel's latest and greatest is the same 4 core nonsense? Tiger Lake has some merit on paper: 18% IPC vs Skylake and 4.3 GHz shouldn't be more than 10% behind 4.8-5.0 GHz Zen3; but why won't Intel make some real CPUs with it? I'm talking 16-32 cores, 100-200 W.

 

[IntelUser2000]

@amdfan111 Intel's share is roughly 70/30 for laptop/desktop. You can assume AMD is almost the opposite. Hence, the difference in focus.

Of course 30% is not a small deal to lose so they'll want to address this, but considering the hole they dug themselves, it'll take some time to fix it.

Silicon design is difficult, so if you are not ready for it, it'll take significant amount of time. Optimizations are needed because it can easily swing 10-20% either way.

 

[CHADBOGA]

Tiger Lake, Willow Cove, whatever you want to call it. It could be a great product if Intel cared enough to make desktop CPUs, but all they care about is laptop. Fact is, only AMD makes a desktop-first CPU (RyZen, lots of cores, no iGPU crap). If Tiger Lake is so great, why won't Intel make something more substantial than some 2-core, 4-core rubbish? What AMD 4800H has shown is that even a tiny laptop chip can fit 8 cores, but Intel's latest and greatest is the same 4 core nonsense? Tiger Lake has some merit on paper: 18% IPC vs Skylake and 4.3 GHz shouldn't be more than 10% behind 4.8-5.0 GHz Zen3; but why won't Intel make some real CPUs with it? I'm talking 16-32 cores, 100-200 W.

Because their manufacturing process is currently not up to it.

 

[ondma]

Tiger Lake, Willow Cove, whatever you want to call it. It could be a great product if Intel cared enough to make desktop CPUs, but all they care about is laptop. Fact is, only AMD makes a desktop-first CPU (RyZen, lots of cores, no iGPU crap). If Tiger Lake is so great, why won't Intel make something more substantial than some 2-core, 4-core rubbish? What AMD 4800H has shown is that even a tiny laptop chip can fit 8 cores, but Intel's latest and greatest is the same 4 core nonsense? Tiger Lake has some merit on paper: 18% IPC vs Skylake and 4.3 GHz shouldn't be more than 10% behind 4.8-5.0 GHz Zen3; but why won't Intel make some real CPUs with it? I'm talking 16-32 cores, 100-200 W.

Where can you buy those 5 ghz zen 3? Oh, wait, it doesnt exist.

 

[Markfw]

Where can you buy those 5 ghz zen 3? Oh, wait, it doesnt exist.

Zen 3 is not out yet. Do you just like to troll ?

 

[uzzi38]

Where can you buy those 5 ghz zen 3? Oh, wait, it doesnt exist.

You will never be able to.

You'll be able to buy a 4.7GHz Zen 3 with better ST performance than the 9900K by a long shot or whatever housefire comes next instead, and still not get ICL/TGL on desktop. Glorious foundry muck-ups at it's finest.

 

[DrMrLordX]

didn't ashraf also say Raja wanted to "enter the market with a bang"

He does like to make a scene. That's what's so great about him. He'll get the bang he wants, eventually . . . but it may not be until 2022. This topic isn't the right place for that discussion, though.

 

[mikk]

Intel normally splits its CPUs into categories, such as H-series for 35W/45W TDP parts, U-series for 15W and 28W hardware, and Y-series for less than that. Those names are still used externally, however Intel has stated to us in conversation that they want to move away from that segmentation as a marketing message.

Intel: 28 W Ice Lake Core i7-1068G7 Coming Q1

www.anandtech.com

It could explain why they have been ADL-UH renamed into ADL-P! No more U and H segments.

 

[uzzi38]

Intel: 28 W Ice Lake Core i7-1068G7 Coming Q1

www.anandtech.com

It could explain why they have been ADL-UH renamed into ADL-P! No more U and H segments.

Which means with ADL all mobile chips will be ULV ones instead of how they are currently.

Well would you look at that, they realised that they'd need to ensure that -H aren't the just binned desktop chips they are currently.

 

[jpiniero]

Intel: 28 W Ice Lake Core i7-1068G7 Coming Q1

www.anandtech.com

It could explain why they have been ADL-UH renamed into ADL-P! No more U and H segments.

That already is happening with Rocket. They are going to ship 35+ W parts of that. Even now there appears to be some OEMs setting PL1 to 35 W on the 6 core Comet Lake U.

Which means with ADL all mobile chips will be ULV ones instead of how they are currently.

Well would you look at that, they realised that they'd need to ensure that -H aren't the just binned desktop chips they are currently.

Probably a cost cutting move, helping out validation for OEMs. It would also mean that "H" would get the PCH onpackage instead of off. Would also explain the LGA 1700 socket, S probably is doing the same thing.

 

[liahos1]

Zen 3 is not out yet. Do you just like to troll ?

do you think zen 3 is going to hit 5ghz? single core? all core?

 

[scannall]

do you think zen 3 is going to hit 5ghz? single core? all core?

I think Intel's 14nm is the last time you will see 5Ghz, on silicon anyway. AMD and Intel are both going wider, which holds back clocks. But that's fine since you will have more throughput. 5Ghz isn't some magic number. At the end of the day, the part that gets the most work done for your use is the best part for you.

 

[Gideon]

NVM delete, wrong thread

 

[Markfw]

do you think zen 3 is going to hit 5ghz? single core? all core?

Regardless of what I think, thisd is not the place to discuss that.

 

[lobz]

Tiger Lake, Willow Cove, whatever you want to call it. It could be a great product if Intel cared enough to make desktop CPUs, but all they care about is laptop. Fact is, only AMD makes a desktop-first CPU (RyZen, lots of cores, no iGPU crap). If Tiger Lake is so great, why won't Intel make something more substantial than some 2-core, 4-core rubbish? What AMD 4800H has shown is that even a tiny laptop chip can fit 8 cores, but Intel's latest and greatest is the same 4 core nonsense? Tiger Lake has some merit on paper: 18% IPC vs Skylake and 4.3 GHz shouldn't be more than 10% behind 4.8-5.0 GHz Zen3; but why won't Intel make some real CPUs with it? I'm talking 16-32 cores, 100-200 W.

Please tell me that was just a typo. Till then I can't react normally.

 

[FriedMoose]

https://twitter.com/i/web/status/1217523000624336897

I'm sure IT World Canada is incorrect and those gains are from IPC and clocks, not through architecture alone. >10% IPC gain would be very unexpected.

 

[coercitiv]

I'm sure IT World Canada is incorrect

Where exactly did they write the gains are IPC exclusive? Fine-tuning an architecture can be done to improve fmax as well, not just IPC.