• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

Discussion Intel current and future Lakes & Rapids thread

Page 158 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

lobz

Golden Member
Feb 10, 2017
1,625
2,088
136
Let me explain why both of you are right.

Stilt is doing testing that's entirely single threaded. Due to the architectural choices between AMD and Intel, SMT gains are better on AMD.

So in strictly single threaded mode, they can be equal, yet at the same AMD is being faster per thread.

Also regarding AT's SpecCPU 2017 testing suite, annoyingly they are running the _Rate suite set at 1 rather than just running the one meant for single thread, which is the Integer/FP speed suite. But with SpecCPU 2006 suite they are running the Speed suite.

Like, what are they doing?

According to AT's Spec tests, Icelake is 30%+ faster than Whiskeylake, not 18%. Intel's choice of being so exact with gains are strange. Why not just say 15-20%? 18%? Are you sure its not 17.9995%? :rolleyes:

Let's also call it performance per clock, not "IPC". I'm sick of this term. People get extremely confused and start arguing about stupid things. Like they say its per cycles and it performs different in reality.

No, its called performance per clock.
I'd say you're almost perfectly right in this case. Almost, because I'm 99.99% sure that if that test suite puts SKL on par with Zen 2, then ICL will never have a 18% advantage against SKL in a million years in the same suite. So while you're right about that confusion and those differences you've mentioned, I was simply saying he can't state general advantage numbers based on one test and using the 18% number from a completely different source. I'm sure we'll see a somewhat better comparison between all of these when ICL comes to the desktop.

Oh wait...
 

Gideon

Golden Member
Nov 27, 2007
1,260
2,373
136
I'd say you're almost perfectly right in this case. Almost, because I'm 99.99% sure that if that test suite puts SKL on par with Zen 2, then ICL will never have a 18% advantage against SKL in a million years in the same suite. So while you're right about that confusion and those differences you've mentioned, I was simply saying he can't state general advantage numbers based on one test and using the 18% number from a completely different source. I'm sure we'll see a somewhat better comparison between all of these when ICL comes to the desktop.

Oh wait...
We can hopefully compare Renoir to Ice Lake, early next year
 
  • Like
Reactions: lobz

Ajay

Diamond Member
Jan 8, 2001
8,442
3,280
136
Rocketlake might be too large to manufacture cheaply with 8 Willow Cove cores and 32EU Gen 12 iGPU. If they enlarge the L3 to 3MB and do a split L2 to extend it to 1.25MB, we might end up having monolithic 8 core RKL being close to 300mm2, which will not help their supply issues with 14nm chips.
Well, sans iGPU and two core, I think it’s doable as a 'halo' product. I’m sure Intel has run simulations to determine the cpu performance under several variations of the willow cove architecture and found the one that yields the highest gaming performance on the current 14 nm process. If they can salvage a win on top binned dice (probably some KF/KS part w/o the GPU), then they can remain the standard bearer as a gaming CPU. That puts the whole downstream line in a favorable light. I would expect Intel will put this to the test soon in FPGA mock-ups.

Intel is going to lose the ST crown to Zen3, so they have nothing to lose when trying to compete with Zen4.
 

lobz

Golden Member
Feb 10, 2017
1,625
2,088
136
If Willow Cove features an 11% IPC increase over Sunny Cove then that will result in ~30% IPC increase over Skylake. (My 11% guess for Willow Cove is complete conjecture by the way, I have no evidence of this. Just making assumptions based on previous architectures and trying to piece together leaks)

However to a degree you are correct, clock speeds may take a hit, so let's take a rough estimate of clock speeds regressing ~10% as an example. Then the overall performance increase will still be in the region of 20% - Still a vast improvement.
That's the exact same level of wishful thinking that made AMD to make a hype video of vega with the unmistakable hint of 'POOR VOLTA' in it.
 

IntelUser2000

Elite Member
Oct 14, 2003
7,398
2,096
136
Well, sans iGPU and two core, I think it’s doable as a 'halo' product. I’m sure Intel has run simulations to determine the cpu performance under several variations of the willow cove architecture and found the one that yields the highest gaming performance on the current 14 nm process.
The leak says the 8 core has the GT1 Gen 12/Xe iGPU, so its quite likely its MCM. It's not just about saving costs on otherwise very profitable enthusiast K chips, but rest of the lineup. Even the quad core will be fairly large and I'm estimating it could be large as 200mm2, and that'll be problematic as a high volume product for -S chips. Current high volume -S chips are in the 120mm2 range, and I think historically its a high for them and they really want to bring that to 100mm2.

If you see the chips that went into Black Friday sales systems, Intel must be selling some chips at $50-70. Packaging is $5-7, so some might easily cost $20 to produce. For the K chips, even $40 is good if they can sell the thing for $150-200.

This extreme price sensitivity is why eDRAM and HBM makes zero sense as a high volume product. Intel ARK prices are misleading because while the non eDRAM U chips are listed the same as the eDRAM equipped ones, realistically Intel is selling for much higher and you can only see such chips in the highest laptop configurations.

Also there's no guarantee that the next generation product will do better. When screwups happen the opposite is true. We'll see.

Intel is going to lose the ST crown to Zen3, so they have nothing to lose when trying to compete with Zen4.
Not sure how it'll do against the competition.

But there's a potential clock advantage for low clocked high perf/clock CPUs. If they are clocking the RKL/TGL S and K dies conservatively,* that means it might actually have overclocking headroom.

Right now the 9900K and KS have almost no overclocking headroom. If RKL can clock 4.2GHz stock but overclock to 4.6GHz, it'll be faster than any CFL CPU.

*This is similar to initial batch of P4800X Optane being rated at 30DWPD but later ones being raised to 60DWPD, or 660P having 200 TBW(1TB size) endurance and 665P having 300TBW endurance. The difference is better characterization as engineers understand more of the new product.
 
Last edited:

lobz

Golden Member
Feb 10, 2017
1,625
2,088
136
The leak says the 8 core has the GT1 Gen 12/Xe iGPU, so its quite likely its MCM. It's not just about saving costs on otherwise very profitable enthusiast K chips, but rest of the lineup. Even the quad core will be fairly large and I'm estimating it could be large as 200mm2, and that'll be problematic as a high volume product for -S chips. Current high volume -S chips are in the 120mm2 range, and I think historically its a high for them and they really want to bring that to 100mm2.

If you see the chips that went into Black Friday sales systems, Intel must be selling some chips at $50-70. Packaging is $5-7, so some might easily cost $20 to produce. For the K chips, even $40 is good if they can sell the thing for $150-200.

This extreme price sensitivity is why eDRAM and HBM makes zero sense as a high volume product. Intel ARK prices are misleading because while the non eDRAM U chips are listed the same as the eDRAM equipped ones, realistically Intel is selling for much higher and you can only see such chips in the highest laptop configurations.

Also there's no guarantee that the next generation product will do better. When screwups happen the opposite is true. We'll see.



Not sure how it'll do against the competition.

But there's a potential clock advantage for low clocked high perf/clock CPUs. If they are clocking the RKL/TGL S and K dies conservatively,* that means it might actually have overclocking headroom.

Right now the 9900K and KS have almost no overclocking headroom. If RKL can clock 4.2GHz stock but overclock to 4.6GHz, it'll be faster than any CFL CPU.

*This is similar to initial batch of P4800X Optane being rated at 30DWPD but later ones being raised to 60DWPD, or 660P having 200 TBW(1TB size) endurance and 665P having 300TBW endurance. The difference is better characterization as engineers understand more of the new product.
I'm so happy to have read this thread through. I've learned from a few people here that they're not the blind fans some replies in other threads may have suggested. In fact, you, for example are frankly more knowledgeable than me, and you don't let your sentiments get in the way of that knowledge - unlike many others :) This thread was some much needed fresh air for me on this forum!

Sorry for OT :)
 

mikk

Diamond Member
May 15, 2012
3,120
938
136
Only makes sense since that is what it looks like they are doing with U. It would give them the option to not include the IGP with K parts too.
Only makes sense for a GT2 imho. I wouldn't expect a chiplet for a GT1 with only 32 EUs. If they do plan a GT2 for RKL at some point a separate 10nm die could make sense. If you look in the tweakers roadmap desktop was fully 14nm. The problem is you are speaking like it had been confirmed from Intel which isn't.
 

Ajay

Diamond Member
Jan 8, 2001
8,442
3,280
136
Okay, so I am missing something. Is there a new tweakers roadmap? I can only find one from the spring for SIPP.
 

uzzi38

Golden Member
Oct 16, 2019
1,506
2,878
106
There were like 2-3 all at the same time and at least 2 in the same article
 

uzzi38

Golden Member
Oct 16, 2019
1,506
2,878
106
Okay, so I am missing something. Is there a new tweakers roadmap? I can only find one from the spring for SIPP.
Both are in here, you gotta scroll down

 
  • Like
Reactions: Ajay

uzzi38

Golden Member
Oct 16, 2019
1,506
2,878
106
Im hoping that the IGP is vastely improved, GT1-GT2 mainly.
For Tiger Lake or Rocket Lake?

Rocket should be in between Gen 9 and Gen 11 for the iGPU - 32EUs of Gen 12 will be around there anyway. Tiger on the other hand is a completely different story. The iGPU on that thing will be very nice.
 

Ajay

Diamond Member
Jan 8, 2001
8,442
3,280
136
Danke. It's the same ones I've seen before. Still do not understand why Intel would include a GPU in an i7 or i9 cpu, unless it is some bizarre OEM requirement.
 

jpiniero

Diamond Member
Oct 1, 2010
9,181
1,812
126
Only makes sense for a GT2 imho. I wouldn't expect a chiplet for a GT1 with only 32 EUs. If they do plan a GT2 for RKL at some point a separate 10nm die could make sense. If you look in the tweakers roadmap desktop was fully 14nm. The problem is you are speaking like it had been confirmed from Intel which isn't.
Can't see them doing a chiplet for U and monolithic for S. Wouldn't make sense. The die size of the GPU is still going to be fairly large but the CPU die size is the issue.

Figure the reason that 14 nm was only mention for S was because there is no option for the 10 nm like U gets.
 

IntelUser2000

Elite Member
Oct 14, 2003
7,398
2,096
136
Rocket should be in between Gen 9 and Gen 11 for the iGPU - 32EUs of Gen 12 will be around there anyway. Tiger on the other hand is a completely different story. The iGPU on that thing will be very nice.
The 32EU Gen 11 is ~20% faster than Gen 9, and 64EU Gen 11 is 2x as fast as Gen 9. Things not scaling linearly benefits the lower end more.

Might only end up being 10-20% slower than the 64EU Gen 11. Of course that's if whatever variant going in RKL behaves similarly to the one in Tigerlake and assuming TGL graphics is really 2x as fast as Gen 11.

Thanks @lobz
 

uzzi38

Golden Member
Oct 16, 2019
1,506
2,878
106
It's 2x faster when comparing 25W TGL vs 15W ICL.

You need to account for the Gen 11 in the 2x figure Intel provided being clocked lower.
 

coercitiv

Diamond Member
Jan 24, 2014
4,299
5,494
136
Still do not understand why Intel would include a GPU in an i7 or i9 cpu, unless it is some bizarre OEM requirement.
Ditching the iGPU has several implications:
  • the entire i5 ->i7 desktop lineup ditches the iGPU, leadng to a clear loss in value for the i5
  • AMD and Nvidia GPUs get a free lunch
  • the i7 and i9 lose computing power in video editing / recording / transcoding (against a competition with much more raw power available in these tasks)
It's not all black and white in the case against the iGPU on i7/i9.
 

uzzi38

Golden Member
Oct 16, 2019
1,506
2,878
106
The 32EU Gen 11 is ~20% faster than Gen 9, and 64EU Gen 11 is 2x as fast as Gen 9. Things not scaling linearly benefits the lower end more.

Might only end up being 10-20% slower than the 64EU Gen 11. Of course that's if whatever variant going in RKL behaves similarly to the one in Tigerlake and assuming TGL graphics is really 2x as fast as Gen 11.

Thanks @lobz
Okay, now that I've gotten some sleep, it occured to me you've made two incorrect assumptions in that:

1. Intel's 2x figure came from a slide where they compared 15W ICL-U vs 25W TGL-U (well, they didn't specify what the chips were, but clearly stated they were comparing 15W vs 25W)

2. TGL-U has 96EUs vs ICL-Us 64EUs. That 2x figure isn't at the same EU count either.
 

Spider-Man

Junior Member
Oct 29, 2018
4
3
41
So, according to Anandtech, Zen 2 has ~7% higher IPC than CFL-R (P0).

Now, they used DDR4-3200 on AMD and DDR4-2666 on Intel. That's a ~20% difference in memory speed. They also divided performance by GHz, which disadvantages the higher clocked CPU because most workloads don't scale perfectly with clock speed.

The Stilt tests a large suite and locks both CPUs to the same core and memory frequencies. SKX-R also ends up faster than both even when you look at the ER results.

Phoronix shows that the hardware fixes going from Skylake-SP to Cascade Lake-SP provide a positive uplift in performance (a smaller penalty). They even show a positive uplift in performance going from CFL-R (P0) to CFL-R (R0).
 
  • Like
Reactions: Zucker2k and mikk

ASK THE COMMUNITY