Tuesday, May 28th 2019

Intel "Ice Lake-Y" MCM Pictured Up Close

Here are some of the first pictures of the 10th generation Core "Ice Lake-Y" multi-chip module, designed for ultra low-power platforms, such as notebooks and 2-in-1 convertibles. These chips have a TDP target as low as 8W to 15W, are built in a BGA (ball-grid array) MCM (multi-chip module) to minimize Z-height and conserve mainboard PCB real-estate, respectively. The larger dies of the two of course is the 10 nm "Ice Lake" SoC with 4 "Sunny Cove" CPU cores, and a Gen11 GT2 iGPU. The smaller die is the PCH (platform controller hub), or chipset. The RVP (reference validation platform) is a motherboard that lets you test every possible connectivity option of the platform, by manually switching around PCIe lanes, SATA links, GPIO, LVDS, and TMDS lanes via jumpers. Intel usually hands these out to OEMs, system integrators, and system software developers. We also spotted the "Ice Lake" 4-core wafer.
Add your own comment

43 Comments on Intel "Ice Lake-Y" MCM Pictured Up Close

#26
Darmok N Jalad
Vya Domus
Compared to an APU from 6 years ago ? That's not exactly ground breaking.
No, but keep in mind it wasn’t that long ago that mobile gaming was a painful experience unless you had a honking gaming laptop. And also consider that the A12X is a passively cooled chip in a device that is 1lb, as thick as a pencil, and gets around 10 hours of battery life. To me, that's saying something.

Besides, XboxOne provides and acceptable gaming experience for many people. Does it need to get even better? Sure, but we are headed that direction. Imagine an A12X with a higher TDP and active cooling. Or imagine a Ryzen APU with more memory channels/bandwidth. I think we aren’t that far off.
Posted on Reply
#27
Prince Valiant
Darmok N Jalad
No, but keep in mind it wasn’t that long ago that mobile gaming was a painful experience unless you had a honking gaming laptop. And also consider that the A12X is a passively cooled chip in a device that is 1lb, as thick as a pencil, and gets around 10 hours of battery life. To me, that's saying something.

Besides, XboxOne provides and acceptable gaming experience for many people. Does it need to get even better? Sure, but we are headed that direction. Imagine an A12X with a higher TDP and active cooling. Or imagine a Ryzen APU with more memory channels/bandwidth. I think we aren’t that far off.
Running at full power for 10H without throttling would be impressive. I presume that isn't the case here?
Posted on Reply
#28
Steevo
18% more IPC when compared to prior models with all security fixes in place and HT disabled? Sure, I will buy that, right when I buy they 4.1Ghz is it's actual sped and not what one core will reach for a second during boot before it's heat limited and then never seen again.

Intel is in trouble, they know it and now sound like the vaporware of AMD bulldozer past.
Posted on Reply
#29
I No
All this talk yet nobody said jack about the acutal usage of the chip which would be ultrabooks. You know those craptastic laptops everyone and their mother gets for every major business for their employees? Outside of this however I don't see anything else though, unless that 18% IPC uplift turns out to be ture but I wouldn't hold my breath.
Posted on Reply
#30
Steevo
I No
All this talk yet nobody said jack about the acutal usage of the chip which would be ultrabooks. You know those craptastic laptops everyone and their mother gets for every major business for their employees? Outside of this however I don't see anything else though, unless that 18% IPC uplift turns out to be ture but I wouldn't hold my breath.
Ultrabooks are fine for auditors and people who want to read an ebook, but so is mobile. It's essentially for devices that a pen and paper, and cell phone can replace, and seeing their adoption where I work it's left up to actual laptops or desktops to do the heavy lifting.

There will be a division in computing where cell phones are as fast and take over more, laptops and desktops will still win performance, and Ultrabooks will slow as people need more connectivity and power than they offer.
Posted on Reply
#31
Darmok N Jalad
Prince Valiant
Running at full power for 10H without throttling would be impressive. I presume that isn't the case here?
No, it isn't, but it's still an example of future potential. I'm not sure what a fully loaded iPad would do to battery life, but that's always been true. It's the better idle states and low-load power use that have given us the major gains in battery life.
Posted on Reply
#32
Aquinus
Resident Wat-man
Darmok N Jalad
No, it isn't, but it's still an example of future potential. I'm not sure what a fully loaded iPad would do to battery life, but that's always been true. It's the better idle states and low-load power use that have given us the major gains in battery life.
Most of the time nowadays the battery is getting chewed up by the backlighting on the display, not by the SoC. Power and thermal management for CPUs and GPUs these days has gotten pretty good.
Posted on Reply
#33
Midland Dog
Tomorrow
18% for which we only have Intel's word. In which workload against what model was this 18% achieved?
Also minus god knows how much percent IPC loss due to security mitigations.

2400 to 3200 RAM does not help much. Especially on CPU limited scenarios. Helps more on the iGPU side.
which is where the focus should be
Posted on Reply
#34
londiste
Tomorrow
18% for which we only have Intel's word. In which workload against what model was this 18% achieved?
Also minus god knows how much percent IPC loss due to security mitigations.
Averaged over a number of benchmarks:

Slides are covered in a bunch of places but Anandtech seems as good a source as any: https://www.anandtech.com/show/14436/intel-10th-gen-10nm-ice-lake-cpus
This is compared to Skylake, so security mitigations have minimal impact in this comparison. If any, security mitigations have heavier impact on Skylake and will widen the gap.
Iso-frequency. I wouldn't put it beyond Intel to test different core counts but from results it would be pretty impossible for that to be the case.
Posted on Reply
#35
efikkan
Intel, AMD, and Nvidia have one thing in common; they usually pick a "best case scenario" when displaying their new tech, Nvidia cherry-picked games when showing off Turing, AMD did it with Radeon VII with very few games and now with Zen 2 with Cinebench and Blender. The benchmarks might not be fake, but they are usually a little on the optimistic side.

So when Intel claims an 18% "IPC" gain for Ice Lake (-U/-Y), I will take it with a grain of salt. But still, Sunny Cove is a major architectural overhaul, and even if the real gain is more like ~12-15%, it's still very good and would be comparable to Sandy Bridge -> Skylake.

But remember, this is just the low-power variant. I'm more interested to see the improvements for Ice Lake-SP, which have a different cache configuration, memory controller and various core features.
Posted on Reply
#36
Darmok N Jalad
efikkan
Intel, AMD, and Nvidia have one thing in common; they usually pick a "best case scenario" when displaying their new tech, Nvidia cherry-picked games when showing off Turing, AMD did it with Radeon VII with very few games and now with Zen 2 with Cinebench and Blender. The benchmarks might not be fake, but they are usually a little on the optimistic side.

So when Intel claims an 18% "IPC" gain for Ice Lake (-U/-Y), I will take it with a grain of salt. But still, Sunny Cove is a major architectural overhaul, and even if the real gain is more like ~12-15%, it's still very good and would be comparable to Sandy Bridge -> Skylake.

But remember, this is just the low-power variant. I'm more interested to see the improvements for Ice Lake-SP, which have a different cache configuration, memory controller and various core features.
Apparently when Intel introduced Ice Lake graphics, they showed it could keep up with or beat the Ryzen IGPs. In the fine print, the Intel machine had like 40%+ more memory bandwidth!
Posted on Reply
#37
londiste
In this case, based on data we have been presented - Ice Lake's +18% on average seems to be based on broader set of benchmarks compared to Zen2's +15% in Cinebench R20.
Posted on Reply
#38
TheGuruStud
londiste
In this case, based on data we have been presented - Ice Lake's +18% on average seems to be based on broader set of benchmarks compared to Zen2's +15% in Cinebench R20.
Yes, they were lying. We all know that. Cherry pick enough useless things while crippling the other system = bigger numbers.
Posted on Reply
#39
R-T-B
btarunr
Yes, every uncut wafer that ever made it out of a cleanroom or an argon-filled container is trashed.
#savethewafers
Posted on Reply
#40
Rahnak
All companies cherry pick numbers. According to PCWorld's Mark Hachman you take that 18% IPC over Skylake, factor in the lower boost clocks over Whiskey Lake and you end up with a modest perf increase (I think he said 2-3% or 3-4%?) over Whiskey Lake.

Anyway, I'm very much looking forward to all the reviews on these new CPUs. So many questions.
Posted on Reply
#41
Xajel
So why this glue is better?
Posted on Reply
#42
rvalencia
R0H1T
They haven't actually disclosed which models boost to 4.1 Ghz, because they've listed 3 TDP numbers 9/15/28W which could be significant.
Also 8665u already boosts to 4.8Ghz so this is a slight downgrade for single core boost.
8665U has weaker 24 IEU IGP.
Posted on Reply
#43
Xzibit
londiste
In this case, based on data we have been presented - Ice Lake's +18% on average seems to be based on broader set of benchmarks compared to Zen2's +15% in Cinebench R20.
In the performance benchmark Intel has them running on different windows build

For ICL
Intel
OS: Windows* 10 RS5-17763.316
For WHL
Intel
OS: Windows* 10 RS4-17134.112.
On the gaming its vs a actual product (HP spectre x360 13t 13-ap0038nr). Should have use the same restrictions for both.

Its also interesting that on the game test Intel decided to use the same windows build for comparison but not for the performance test.

[quote=Intel]
1920×1080 – Fullscreen]Inconsistent to say the least. ICL pre-production test bench vs a final product in one test then a pre-production in another and they throw in the mixing of OS builds. All while not even mentioning ICL clocks.
Posted on Reply
Add your own comment