Anandtech: The Ice Lake Benchmark Preview: Inside Intel's 10nm

Snowdog

[H]F Junkie
Joined
Apr 22, 2006
Messages
11,262
AT got to spend some time with intels new core. While there are some nice gains in specific tests, overall progress is limited because while IPC improves, clock speed regresses almost an equal amount:
https://www.anandtech.com/show/14664/testing-intel-ice-lake-10nm/4

111797.png


111763.png


111764.png


111788.png


A quick calculation of 1.47/1.42 means that even Intel is only predicting an absolute gain of ~3.5% for Ice Lake over current generation systems.

The reason why the difference is so small is because of IPC and frequency. Intel is touting a median IPC advantage on the new Sunny Cove cores of +18% against Skylake. That isn’t something we were able to test in the short time we had with the system, but +18% should provide a healthy bump – we actually see a number of key microarchitectural improvements bubble up through in our SPEC testing.

But at the same time, the frequency has decreased – our Whiskey Lake Huawei Matebook system was +500 MHz on the base frequency (+38%), and +700 MHz on the turbo frequency (+18%). If it were not for the vast increase in memory speed, moving from LPDDR3-2133 to LPDDR4X-3733, one might have predicted that the Core i7-1065G7 Ice Lake processor and the Core i7-8565U Whisky Lake processor would have performed equally.
 
All mobile/embedded, so lower clocks in general are expected. That's 500/700 less than another mobile part? Wooh, That's rough. For Intel's sake, hopefully power scales gently for their desktop parts.
 
All mobile/embedded, so lower clocks in general are expected. That's 500/700 less than another mobile part? Wooh, That's rough. For Intel's sake, hopefully power scales gently for their desktop parts.

Agree, hopefully as their 10nm mature a bit more, we see better clock for the desktop variants.
 
10 nm from Intel is never going to produce a chip for desktop that anyone wants. Intel knows this and why they are trying to get to 7 nm as fast as they can now as they need EUVL for monolithic design. If they switch to a chiplet design then maybe things might be different.
 
All mobile/embedded, so lower clocks in general are expected. That's 500/700 less than another mobile part? Wooh, That's rough. For Intel's sake, hopefully power scales gently for their desktop parts.

We are entering an age of lower max clocks for everything. As process nodes shrink beyond 14nm, leakage current makes it more and more difficult to sustain high clocks.

AMD has been seeing this at both 12nm and 7nm, and Intel is now seeing it as well, and they are likely seeing more of it earlier due to their 10nm woes.

I feel fairly certain that thhis is not just a mobile issue, but that instead, the age of 5ghz clocks is permanently over, and clocks are going to keep going down with every die shrink.

In CPU design It's going to be important to pump up IPC gains going forward to compensate.
 
https://www.ultrabookreview.com/22169-whiskey-lake-laptops/
i7 8565u clocks: 1.8 base; 4.6 max [GHZ].

and it is -500/700

Yeah 10nm is dead on the water.

edit to explain:
pretty great for laptops, i just don't see this being able to scale properly for desktop parts, at the moment.

EDIT:
the gaming test is worthless unless whoever made the OP explains that the iGPU has 64 EU vs the 24 EU of the past so what you are looking at is the much bigger iGPU.
 
Last edited:
EDIT:
the gaming test is worthless unless whoever made the OP explains that the iGPU has 64 EU vs the 24 EU of the past so what you are looking at is the much bigger iGPU.

It's useful for mobile users that would be limited to the IGP, but it's not that useful for general gaming where a desktop dGPU would be used. Still, the higher scores show that the lower clockspeeds aren't really hampering gaming performance, which means that a lower-clocked Ice Lake part with a mobile dGPU isn't going to lose performance versus a Skylake-based part.

pretty great for laptops, i just don't see this being able to scale properly for desktop parts, at the moment.

I see this repeated regularly, while there's no evidence available one way or the other, and there are good arguments for and against Intel's 'mobile first' approach to 10nm.
 
https://www.ultrabookreview.com/22169-whiskey-lake-laptops/
i7 8565u clocks: 1.8 base; 4.6 max [GHZ].

and it is -500/700

Yeah 10nm is dead on the water.

edit to explain:
pretty great for laptops, i just don't see this being able to scale properly for desktop parts, at the moment.

EDIT:
the gaming test is worthless unless whoever made the OP explains that the iGPU has 64 EU vs the 24 EU of the past so what you are looking at is the much bigger iGPU.

Yup, doesn't help Intel 14nm is so mature that it is hard for their 10nm to clock high. I wouldn't say gaming test is worthless as you are getting more iGPU performance while drawing the same amount of power.
 
IdiotInCharge, it shows average and at most 95th percentile, yes those are bad at the moment, we kinda moved to 1% after years of stuttering and you know this mate, it is part of the devil in the details with how this "preview" was allowed to be made, it shows the best lights hiding all those terrible crinkles underneath :)

I do think that for laptop it will be good, but i wouldn't be praising it at all.
 
we kinda moved to 1% after years of stuttering and you know this mate, it is part of the devil in the details with how this "preview" was allowed to be made, it shows the best lights hiding all those terrible crinkles underneath :)

Well, I'd prefer actual frametime analysis, but we get what Anandtech got and looks to be the best we're going to get at the moment.

I do think that for laptop it will be good, but i wouldn't be praising it at all.

I game on an i7 8550u with the IGP. I also game on a 9900k with a 1080Ti under water.

What Intel is providing for the ultrabook space is a pretty big jump, and miles ahead of their closest competition. They support VRR and it moves from a 'want' to a 'buy' for me, in the right chassis.


[as a tangent, my 8550u, in a thin ultrabook, can pull well over 1080p60 in MOBAs and the like- Ice Lake would easily pull 1080p120, and with VRR, would be a dream to use across the board]
 
At the end of the day all my users want is something light that has a good battery and strong wifi, and a 1080p screen in the 13-15" range, if it can do email, excel, and a few websites along the way with out stuttering then I won't hear from them until one of them inevitably spills a coffee on it and blows it out.
 
until one of them inevitably spills a coffee on it and blows it out.

Then they need one of those Latitudes that's rated to stop 7.62x39, and can deal with the extra weight...


[mostly joking on the bullet-stopping point, but my work Latitude sure feels like it could]
 
My main disappointment is that I think they could launch this on both 14nm and 10nm. I realize they would have had to do alot of the work twice to make it happen but I think it would be worth it for their desktop and server market if 7nm is even a year or two away. If we could get 6-8 core desktop parts with 15%+ IPC that would be a compelling reason for many workloads to not get ryzen 3.
 
These are fine for low power parts... and that is all Intel 10nm will ever be. Yes they have said so and fully admitted their 10nm designs don't really work properly.

There are very likely never going to be desktop parts based on these. Intel doesn't need to release them either. Their current desktop parts are good enough for the hardcore Intel fans... and will continue to be justified by some due to slightly higher clocks. Intel will stick with those until they have 7nm parts to ship. They have all but admitted 10nm is a lost cause. We are more then half way through 2019 and they have announced nothing for desktop parts based on 10nm... and are already talking about 7nm for next year and Intel chiplet tech to anyone that will listen.

Anyone hoping for a desktop ice lake part for the masses to fight Ryzen2, I'm sorry it's not happening. Intel is going to have some very interesting parts next year if their head engineer is to be believed. Until then they have admitted they where to ambitious with 10nm and it has cost them.
 
These are fine for low power parts... and that is all Intel 10nm will ever be. Yes they have said so and fully admitted their 10nm designs don't really work properly.

There are very likely never going to be desktop parts based on these. Intel doesn't need to release them either. Their current desktop parts are good enough for the hardcore Intel fans... and will continue to be justified by some due to slightly higher clocks. Intel will stick with those until they have 7nm parts to ship. They have all but admitted 10nm is a lost cause. We are more then half way through 2019 and they have announced nothing for desktop parts based on 10nm... and are already talking about 7nm for next year and Intel chiplet tech to anyone that will listen.

Anyone hoping for a desktop ice lake part for the masses to fight Ryzen2, I'm sorry it's not happening. Intel is going to have some very interesting parts next year if their head engineer is to be believed. Until then they have admitted they where to ambitious with 10nm and it has cost them.

Intel is talking 2021 for 7nm, not 2020. Unless they plan to iterate on 14nm again, or release nothing at all, I'd have a hard time believing that we won't see some kind of 10nm desktop chips from them next year.
 
Intel is talking 2021 for 7nm, not 2020. Unless they plan to iterate on 14nm again, or release nothing at all, I'd have a hard time believing that we won't see some kind of 10nm desktop chips from them next year.

Perhaps but I really dont think so. They have 7nm set for 2021 on the nose... they will be talking about it leaking benches summer/fall 2020.

It looks like ice lake is going to increase IPC and greatly reduce clock speed. Right now the only clear win they have over AMD in the desktop market is clock speed. I just don't see them giving up that moral win for a net wash in increased IPC vs a loss of the moral clock win. From what I'm reading at anand it sounds like performance gains would probably be single digit overall. We know from Intels history their marketing wonks in general get their way... I can't see them releasing a part that only clocks to 4.6-4.8 best case even if it is slighly faster. It will get trashed as a side step. I can see Intel more or less sticking to what they got till early 2021 (and perhaps late 2020 if they whip their engineers hard enough)

In the minds of the masses Zen2 is a homerun but it isn't a grand slam either. Intel still has a few arguments on the higher end with higher clocks and a more mature platform. Unless Zen2+ is much more then a simple update I don't see Intel rushing more side step parts for the desktop. 2021 is close enough that by mid next year they will be firing up the hype engine for 7nm Intel chiplets AND XE all at once. All they have to do to stall Zen2+ upgrades for old Intel customers (people already on Zen2 then are gone for awhile anyway) is start talking about Intel chiplets featuring XE graphics ect.
 
Then they need one of those Latitudes that's rated to stop 7.62x39, and can deal with the extra weight...


[mostly joking on the bullet-stopping point, but my work Latitude sure feels like it could]
My carpenters are all equipped with Lifebooks... so yeah those just run till they don't usually the raid controller on those that give out first most sad....
 
Perhaps but I really dont think so. They have 7nm set for 2021 on the nose... they will be talking about it leaking benches summer/fall 2020.

It looks like ice lake is going to increase IPC and greatly reduce clock speed. Right now the only clear win they have over AMD in the desktop market is clock speed. I just don't see them giving up that moral win for a net wash in increased IPC vs a loss of the moral clock win. From what I'm reading at anand it sounds like performance gains would probably be single digit overall. We know from Intels history their marketing wonks in general get their way... I can't see them releasing a part that only clocks to 4.6-4.8 best case even if it is slighly faster. It will get trashed as a side step. I can see Intel more or less sticking to what they got till early 2021 (and perhaps late 2020 if they whip their engineers hard enough)

In the minds of the masses Zen2 is a homerun but it isn't a grand slam either. Intel still has a few arguments on the higher end with higher clocks and a more mature platform. Unless Zen2+ is much more then a simple update I don't see Intel rushing more side step parts for the desktop. 2021 is close enough that by mid next year they will be firing up the hype engine for 7nm Intel chiplets AND XE all at once. All they have to do to stall Zen2+ upgrades for old Intel customers (people already on Zen2 then are gone for awhile anyway) is start talking about Intel chiplets featuring XE graphics ect.
I am not too concerned about Desktop performance, their existing stuff is solid there and currently standing its ground. Laptops is where the bulk of sales are going and Intel is playing the numbers, iron out the process and fine tune it there so in a year or so they can release some desktop parts probably in the 45-65 watt range which would cover the bulk of the AIO's and small form factor desktops, which all combined is what 85% of the market?
 
I am not too concerned about Desktop performance, their existing stuff is solid there and currently standing its ground. Laptops is where the bulk of sales are going and Intel is playing the numbers, iron out the process and fine tune it there so in a year or so they can release some desktop parts probably in the 45-65 watt range which would cover the bulk of the AIO's and small form factor desktops, which all combined is what 85% of the market?

I agree if we see 10nm desktop parts they will probably be mainstream parts. I don't think they are going to release the 10nm 3900/3950 killer with pricing to match. They can sit on what they got for the high end and ride the mindshare that 5ghz gives them until they can release a real next gen part.

AMD will be hard pressed to compete with their mobile parts as you say.... and OEMs are more then willing to go Intel for a year or two even if AMDs mobile parts do win some benchmarks. Especially as in the back rooms with OEMs Intel is already talking about their Chiplet stacking tech. AMD isn't convincing OEMS to evict Intel from their high end designs with one product cycle. :) Intel can sit out for awhile and ride what they have on the market already.

When AMD ryzen2 rumors where circling sales slowed... and when Intel 3D chiplet rumors fire up next year the same effect will keep Intel from bleeding much.
 
I agree if we see 10nm desktop parts they will probably be mainstream parts. I don't think they are going to release the 10nm 3900/3950 killer with pricing to match. They can sit on what they got for the high end and ride the mindshare that 5ghz gives them until they can release a real next gen part.

AMD will be hard pressed to compete with their mobile parts as you say.... and OEMs are more then willing to go Intel for a year or two even if AMDs mobile parts do win some benchmarks. Especially as in the back rooms with OEMs Intel is already talking about their Chiplet stacking tech. AMD isn't convincing OEMS to evict Intel from their high end designs with one product cycle. :) Intel can sit out for awhile and ride what they have on the market already.

When AMD ryzen2 rumors where circling sales slowed... and when Intel 3D chiplet rumors fire up next year the same effect will keep Intel from bleeding much.
Intel has a track record of solid support and documentation, which the Ryzen 3000 stuff (Which I am still happly awaiting my 3900X) is showing AMD's deficiencies in this department with pretty much all their board partners. Until AMD really gets that sorted out Dell, HP, Lenovo, the big guys are going to offer their chips but they aren't going to push them hard, because their after costs are higher.
 
Intel is talking 2021 for 7nm, not 2020. Unless they plan to iterate on 14nm again, or release nothing at all, I'd have a hard time believing that we won't see some kind of 10nm desktop chips from them next year.

FY19: Mobile - 14++ WKL w/ some 10nm CNL
Desktop - All 14++ CFL(R)
FY20: Mobile - 10nm ICL
Desktop - 14++× CML (Comet lake)
FY21: TGL in either 10+ or 7nm for both

My prediction, anyhow. Most thought ICL was going to be a 10+nm part but it is looking like it is the same crappy node as CNL with a different architecture.

It's really looking like FY21 (Oct 2020) before Intel desktop escapes 14nm with Tiger Lake.
 
If 14++nm Skylake smokes 10nm Skylake (CNL), and 10nm Sandy Cove matches or beats 14++ Skylake.

Imagine what 14++ Sandy Cove could have been...
 
More testing done, this time from Tom's:
https://www.tomshardware.com/reviews/intel-10nm-ice-lake-test-benchmarks,6257.html

Icelake tested in both 15w and 25w against 2 Whiskey Lake Cpus and 1 Kaby Lake(R) all at 15w.

Handbrake: The 15w Ice Lake fell right above the Whiskey Lake and below the Kaby(R). The other WKL was much worse, but I am thinking it was single channel memory.

CB15: The 15w ICL was nearly the same as the KBL and WKL. Again, the other WKL seemed to trail. At 25w, the ICL managed 670 points and 2.8 ghz. An 8/16 version at 4.2 ghz would scale up to 2000 points, which is Ryzen 3000 territory.

Compression/Decompression showed a nice boost, but I don't think it will be able to dethrone Ryzen 3000.

Encryption was off the charts showing a 3.5x increase over the older Intel parts, even at 15w.
 
So 5-15% IPC security speed cheating loss, 18% best case gain gain with fixes gives actual IPC change roughly 10-3%?
 
<goes directly to GPU tests to see what Intel has been hollering abt re: "Gen. 11" GFX>
- no benches vs previous Iris Plus iGPUs
- no benches vs Ryzen u-series APU w/ Vega 11
but hey, it beats the pants off Intel's prior midrange iGPUs! :rolleyes: Next up: GeForce 3080Ti set to outperform GeForce 2060 for mad gainz!

"Gen. 11" simply means "we put Iris Plus in more SKUs this time", change my mind.
 
"Gen. 11" simply means "we put Iris Plus in more SKUs this time", change my mind.

It's an improved IGP, and there are more cores and they're putting 'Iris' graphics on more SKUs. Of course, outside of Intel's slides, Anandtech's testing is about as good as we're going to get at the moment.
 
It's an improved IGP, and there are more cores and they're putting 'Iris' graphics on more SKUs. Of course, outside of Intel's slides, Anandtech's testing is about as good as we're going to get at the moment.

I'm going to remain skeptical until I see direct comparisons to say an i7 8559U. :) Don't get me wrong--adding Iris GFX to more mainstream parts is great, but I'm not going to give Intel a pat until I see just how far down the stack the new Iris parts will go, and how they compare to previous Iris iGPUs. They've hyped Gen 11 iGPUs hard and now it's time to put up
 
I'm going to remain skeptical until I see direct comparisons to say an i7 8559U. :) Don't get me wrong--adding Iris GFX to more mainstream parts is great, but I'm not going to give Intel a pat until I see just how far down the stack the new Iris parts will go, and how they compare to previous Iris iGPUs. They've hyped Gen 11 iGPUs hard and now it's time to put up

I think the bigger point is that they're pushing Iris into lower TDP parts. That's a big deal for the smallest and thinest laptops, the ones that cannot accommodate a dGPU. And that is most that come with an i7 8559U. Once you get into dGPU territory, I expect Nvidia and (godsdamn hopefully) AMD to have some discrete parts on deck.
 
I'm going to remain skeptical until I see direct comparisons to say an i7 8559U. :) Don't get me wrong--adding Iris GFX to more mainstream parts is great, but I'm not going to give Intel a pat until I see just how far down the stack the new Iris parts will go, and how they compare to previous Iris iGPUs. They've hyped Gen 11 iGPUs hard and now it's time to put up

Looks like both i7s and and two (of three) i5s for the U-series Ice Lake will use Iris Pro. The i7s and one i5 will have 64 execution units with the other i5 Iris at 48. The last i5 and the single i3 skip Iris and have 32 EUs. On the Y-series side the one i7, both i5s, and one of the two i3s use Iris Plus. The i7 and one i5 have 64 EUs, the other i5 and the i3 have 48, and the non-Iris i3 is at 32. So it looks like they're bringing Iris to most of the Ice Lake lineup.

https://cdn.arstechnica.net/wp-content/uploads/2019/08/Intel-Slide-10nm-1.jpg
 
I think the bigger point is that they're pushing Iris into lower TDP parts. That's a big deal for the smallest and thinest laptops, the ones that cannot accommodate a dGPU. And that is most that come with an i7 8559U. Once you get into dGPU territory, I expect Nvidia and (godsdamn hopefully) AMD to have some discrete parts on deck.

I just realized that's a 28W part I cited, so yeah fair point. More gfx for less power is good. I also hope AMD has something for laptops higher in the market than the barely-better-than-iGPU RX 560 types (KBL-GL and Apple-only mobile Vega notwithstanding)
 
I just realized that's a 28W part I cited, so yeah fair point. More gfx for less power is good. I also hope AMD has something for laptops higher in the market than the barely-better-than-iGPU RX 560 types (KBL-GL and Apple-only mobile Vega notwithstanding)

Same. Also, VRR. Intel hasn't made any noise about this lately, so AMD may be the vendor of choice- you know, if they ship a reasonably efficient part. I'd own one now if they did, and I'm skeptical that they'll come anywhere near Ice Lake.
 
Same. Also, VRR. Intel hasn't made any noise about this lately, so AMD may be the vendor of choice- you know, if they ship a reasonably efficient part. I'd own one now if they did, and I'm skeptical that they'll come anywhere near Ice Lake.

AFAIK VRR is officially supported starting with Ice Lake, so there's that.
 
AFAIK VRR is officially supported starting with Ice Lake, so there's that.

By the gods I hope so.

Intel just hasn't said much about it, but if it's available, I'd love to see it in a Razer or XPS laptop. And a 120Hz panel. Because reasons.
 
By the gods I hope so.

Intel just hasn't said much about it, but if it's available, I'd love to see it in a Razer or XPS laptop. And a 120Hz panel. Because reasons.

I was about to make a snark abt 480P/120 gaming, but 120hz with VRR means LFC which can make a world of difference for low-framerate gaming so, yeah. That would be neat.
 
I was about to make a snark abt 480P/120 gaming, but 120hz with VRR means LFC which can make a world of difference for low-framerate gaming so, yeah. That would be neat.

Oh I just run less intensive games. League of Legends is (somehow) a group favorite, and the 8550U in my ultrabook has no problem exceeding 1080p60 on lowest settings. So give me VRR to take care of the tearing and give me 120Hz for smooth everything and you can have my money!
 
Back
Top