Several Gen11 GPU Variants Referenced in Latest Intel Drivers

[btarunr]

The latest version of Intel Graphics drivers which introduce the company's latest UWP-based Graphics Command Center app, hide another secret in their INF. The file has pointers to dozens of variants and implementations of the company's next-generation Gen11 integrated graphics architecture, which we detailed in a recent article. Intel will implement Gen11 on two key processor microarchitectures, "Ice Lake" and "Lakefield," although later down the line, the graphics technology could trickle down to low-power Pentium Silver and Celeron SoC lines, too, with chips based on the "Elkhart Lake" silicon.

There are 13 variants of Gen11 on "Ice Lake," carved using execution unit (EU) count, and LP (low-power) aggressive power management. The mainstream desktop processors based on "Ice Lake," which are least restrained in power-management, get the most powerful variants of Gen11 under the Iris Plus brand. Iris Plus Graphics 950 is the most powerful implementation, with all 64 EUs enabled, and the highest GPU clock speeds. This variant could feature on Core i7 and Core i9 brands derived from "Ice Lake." Next up, is the Iris Plus Graphics 940, with the same EU count, but likely lower clock speeds, which could feature across the vast lineup of Core i5 SKUs. The Iris Plus 930 comes in two trims based on EU count, of 64 and 48, and could likely be spread across the Core i3 lineup. Lastly, there's the Iris Plus 920 with 32 EUs, which could be found in Pentium Gold SKUs. There are various SKUs branded "UHD Graphics Gen11 LP," with EU counts ranging from 32 to 64.



Intel's "Lakefield" heterogeneous SoC comes with four models of Gen11 graphics, based on two key variants of Gen11, GT1 and GT2, which are physically different in size. The GT1 variant of this iGPU is smaller than GT2, and likely has 32 EUs going by past trends of GT1 configurations. All four models are labelled "UHD Graphics LKF," with no further model numbering. Future releases of Intel drivers could reference "Elkhart Lake," the upcoming low-power SoC that incorporates Gen11 graphics architecture.

View at TechPowerUp Main Site

 

[Deleted member 172152]

So basically, most sku's won't have anywhere near the graphics performance of the 64 eu gen11 igpu if this is legit, kinda proving my point that gen11 (64eu) should actually be compared with iris plus 640 and higher.

AMD has nothing to worry about against a puny 48 eu's.

 

[notb]

So basically, most sku's won't have anywhere near the graphics performance of the 64 eu gen11 igpu if this is legit, kinda proving my point that gen11 (64eu) should actually be compared with iris plus 640 and higher.

Why would they? Most users are perfectly fine with the performance current IGP offer. They'll be more interested in features (hardware encoding being the most important), which Intel is also improving with every generation.

But the top gen11 variant will match 2400G APU or the GT 1030, which most likely was the target.

 

[Deleted member 172152]

Why would they? Most users are perfectly fine with the performance current IGP offer. They'll be more interested in features (hardware encoding being the most important), which Intel is also improving with every generation.

But the top gen11 variant will match 2400G APU or the GT 1030, which most likely was the target.

Well... Looks more like mx130 (940mx) currently, so well below mx150 AND gt1030.

But sure, for general purpose a basic igpu will suffice.

 

[notb]

But sure, for general purpose a basic igpu will suffice.

Since it's an IGP, features and low idle power consumptions will always be more important than performance.

But now Intel has very good reasons to boost performance as well. Many apps started to utilize GPGPU. Sometimes it's for GPU-friendly computations, sometimes just for rendering the UI (like Firefox does).
GPU boost could potentially give AMD APUs an advantage - despite the CPU part being weaker.

 

[jeremyshaw]

Some minor guesswork analysis from the SKUs (Comments in parenthesis).

Y series:
GT2Y64 - Iris Plus 930 (64EU?)
GT2Y48 - UHD Gen 11 LP (48EU)
GT2Y48LM - UHD Gen 11 LP
GT2Y32 - Iris Plus 930 (32EU)
GT2Y32LM - UHD 910 (clearly LM means something, since it's enough to change the designation)

U series:
GT2U64 - Iris Plus 940
GT2U6425W - Iris Plus 950 (25W version, I think. Gets a different billing!)
GT2U48 - Iris Plus 940
GT2U48LM - UHD Gen11 LP (again, LM has to mean something)
GT2U32 - UHD Gen11 LP
GT2U32LM - UHD 920

?? F series?:
GT0 - UHD Gen11 LP (traditionally, GT0 has been "no graphics," afaik, but that no longer seems to be the case)
GT0P5 - UHD Gen11 LP (what is P5 for?)

Lakefield:
GT2 - UHD LKF
GT1P5 - UHD LKF
GT1 - UHD LKF
GT0 - UHD LKF (in the past, I think GT0 would have meant to GPU, but it means something else now [afaik])



As for the LM deal, last time Intel used "LM" for CPUs was in the 1st gen mobile Core i series chips. back then, it meant it was one step below the fullpower "M" series, but one step above the "UM" series. That being said, it probably has no relation to the LM name here, though casually glancing about, the LM designation seems to imply a weaker version (or at least a lower name, for most of the "LM" designated GPUs).

There is more clear segmentation this time around than previous Intel IGPs (not that there was enough power in them to create lower end versions that weren't horrible). GT2 now seems to encompass (potentially) a wide range of available EUs.

It also seems the Y series may get a full GT2 (unlike the current and previous gens), but it still has a lower name, indicating lower clocks.

The most interesting SKU to me is the GT2U6425W, since it implies the higher performance at a TDP-Up of 25W (and gets a different name, too). Of course, this will depend on Intel's willingness to enforce better cooling schemes.

 

[John Naylor]

One interesting fact is that while the higher adoption rate of Turing compared to pascal has been the subject of much discussions, the fact that more people are using discrete GPUs than in recent weeks has drawn less attention. In just the last month .... we've seen nVidia pick up 1.1% market share and AMD and Intel lose 0.6 and 0.5 respectively

Intel's recent market share by month = 9.3 / 10.5 / 10.6 / 10.6 / 10.15
AMD's recent market share by month = 13.9 / 15.2 / 15.1 / 15.3 / 14.69
Nvidia's recent market share by month = 75.6 / 74.1 / 74,2 / 74.0 / 75.1

While by no means indicative of a long term trend, since the recent price drops, two things are evident

AMD gained market share gains against nVidia over 3 previous months have now begun to slide since recent price adjustments. Expect AMD to adjust pricing, will nVidia follow suit.
Intel's fortunes have also reversed, will we continue to see more folks opt for discreet cards in upcoming months ?

While this is just something that **may** turn into somewhat of a trend, it does offer some hope that card prices will continue to drop.

 

[Space Lynx]

Why would they? Most users are perfectly fine with the performance current IGP offer. They'll be more interested in features (hardware encoding being the most important), which Intel is also improving with every generation.

But the top gen11 variant will match 2400G APU or the GT 1030, which most likely was the target.

little late though isn't it? 3400g 7nm will be released round same time as this is released.

 

[jeremyshaw]

little late though isn't it? 3400g 7nm will be released round same time as this is released.

We already know the mobile 3000 APUs are "12nm." Desktop is likely to do the same in the upcoming months, unless if AMD is working contrary to what Dr. Su has already stated (either that or they are developing 3 different I/O dies, two we have already seen, and one unannounced version with a GPU inside).

 

[Aerpoweron]

The Intel GT4 is in the neighborhood of the 2400G iGPU. But it is quite hard to get your hands on that. The GT4 was even a seperate die.

Guess what on the picture is the CPU and what is the GPU. It is an i7 6785R CPU
And as you can see, Intel has released it in 2016.

 

[jeremyshaw]

The Intel GT4 is in the neighborhood of the 2400G iGPU. But it is quite hard to get your hands on that. The GT4 was even a seperate die.

Guess what on the picture is the CPU and what is the GPU. It is an i7 6785R CPU
And as you can see, Intel has released it in 2016.

It's not hard. The bigger one, which contains the CPU, GPU, IMC, L3$, etc. Smaller one is 128MB of eDRAM setup in a sideport fashion (contrary to the earlier L4$ setup of Broadwell).

 

[Aerpoweron]

It's not hard. The bigger one, which contains the CPU, GPU, IMC, L3$, etc. Smaller one is 128MB of eDRAM setup in a sideport fashion (contrary to the earlier L4$ setup of Broadwell).

How was the Broadwell set up? The 5675C has one quite small die and one rectangular bigger on, more than twice the size of the smaller die.

Oh, and could you give me the source of the information you have, to the setup, or what is where in the dies?

I always assumed the smaller Die is the CPU and the bigger die is the GPU with the 128MB of eDRam.

 

[john_]

If we see a significant upgrade in integrated graphics from Intel, AMD will have to take action and offer even faster integrated graphics in it's own APUs. That will push discrete cards even higher than where Nvidia has pushed them today. Nvidia started pushing prices up, because the low end market was killed by AMD's APUs and Intel's "not so terrible", integrated graphics. In 2020-2021 we might see the death of everything under $150.

PS With "killed" and "death" I mean that companies don't sell the huge numbers they used to sell, not that they don't sell at all.

 

[Assimilator]

I always assumed the smaller Die is the CPU and the bigger die is the GPU with the 128MB of eDRam.

I believe you are correct.

 

[Aerpoweron]

I believe you are correct.

I don't believe in believing

That is why i want source material. But it is quite hard to get any useful documentation with google. It was hard enough to get this cpu (i7 6785R).

Here are some Fire Strike Benchmarks with different APU / CPUs with intigrated Graphics. But Some CPUs have issues. The A12-9800 throttles down to 1.4 GHz with heavy GPU-Load is applied. The 6785R throttles to 2.5 / 2.6 GHz when under heavy CPU-Load. (have to find the right flag to set)
So be careful with the results.

https://www.3dmark.com/compare/fs/18529321/fs/18141498/fs/15053139/fs/18026073

I think the i7 6785R can do better with higher clocked DDR4 Memory. The "computer" i have with it only runs DDR3 1600. But i am not even sure if the board can handle faster DDR3 Ram (SoDimm). And i was not able to find any board with the 6785R and DDR4

 

[eidairaman1]

AI is Intel's perogative

 

[notb]

little late though isn't it? 3400g 7nm will be released round same time as this is released.

Are you so sure that 3400G GPU will be much faster than 2400G's? Would it be proper new tech or just rebranded, shrunk stuff we have today?

Either way, not a big deal. Intel doesn't need to make the fastest IGP around. They haven't for a long time and it didn't harm them.
They only need their IGP to stay relevant. Not in context of AMD, but in context of next gen systems.

Intel has to provide a solution that will power single 6/8K screens or multiple 4K screens.
And this has to be a frugal chip that can be put into ultrabooks as well.

Improved gaming performance will be just a welcome side-effect.

 

[Aerpoweron]

I have to agree notb. And don't forget, Intel was even the first company to get a complete DX12 featured GPU. And that was month before AMD or Nvidia.

What i find always funny, when the GPU market shares are compared. You mostly see discrete GPUs. But when you take all GPUs in account (intigrated as well) Intel is far ahead of Nvidia and AMD. The most recent thing i could find was 66% Intel as of Q1 2018.

https://www.ultragamerz.com/new-gpu...own-amd-down-and-intel-gpu-shipments-were-up/

Q3 2018

https://www.statista.com/statistics/754557/worldwide-gpu-shipments-market-share-by-vendor/

 

[Vya Domus]

n 2020-2021 we might see the death of everything under $150.

I believe it will be the complete opposite, dedicated graphics chips are becoming more and more expensive (to manufacture and to sell), I think APUs will slowly become more and more relevant.

Those millions that play Fortnite don't need a thousand dollar 2080ti, a 150$ APU is enough. The reason why this didn't happened yet is thanks to the years of atrociously crappy GPUs Intel stuffed inside their chips. Most people aren't even aware of the fact that video cards are no longer an absolute necessity even when it comes to light gaming.

Slowly but surely these systems will get integrated over time.

 

[notb]

I believe it will be the complete opposite, dedicated graphics chips are becoming more and more expensive (to manufacture and to sell), I think APUs will slowly become more and more relevant.

Haven't you both said the exact same thing? That a boost in integrated GPU performance will lower demand for cheap cards?

Those millions that play Fortnite don't need a thousand dollar 2080ti, a 150$ APU is enough. The reason why this didn't happened yet is thanks to the years of atrociously crappy GPUs Intel stuffed inside their chips. Most people aren't even aware of the fact that video cards are no longer an absolute necessity even when it comes to light gaming.

Intel IGP were never crappy. They're just optimized for different things, not raw performance.
I believe the only party "unaware" here is you... or maybe it's just your hatred towards Intel. ;-)

Millions that play Fortnite and countless other games are already fine using their Intel IGP. It forces some compromises: low settings or resolution. But it's certainly feasible. And when you play on a 14" notebook, there's hardly any difference between 1080p and 720p.

If we wanted to point a competing product that could have pushed Intel into boosting the performance now, it is not the Ryzen APU, but the sudden popularity of MX chips in laptops.
Almost every mid/high-end ultrabook is available with an MX130 or MX150. This has to be one of the more popular product lines they have.
I have no idea how much Nvidia makes from each chip. $20 maybe?
This is a clear sign for Intel: if they made a more powerful SoC or a separate GPU chip, these extra $20 would have gone to them, not their main competitor.

 

[Aquinus]

I always assumed the smaller Die is the CPU and the bigger die is the GPU with the 128MB of eDRam.

The smaller die was just the Crystalwell eDRAM cache.

I believe you are correct.

I believe you're not.

As we’ve been talking about for a while now, the highest end Haswell graphics configuration includes 128MB of eDRAM on-package. The eDRAM itself is a custom design by Intel and it’s built on a variant of Intel’s P1271 22nm SoC process (not P1270, the CPU process). Intel needed a set of low leakage 22nm transistors rather than the ability to drive very high frequencies which is why it’s using the mobile SoC 22nm process variant here.

Despite its name, the eDRAM silicon is actually separate from the main microprocessor die - it’s simply housed on the same package. Intel’s reasoning here is obvious. By making Crystalwell (the codename for the eDRAM silicon) a discrete die, it’s easier to respond to changes in demand. If Crystalwell demand is lower than expected, Intel still has a lot of quad-core GT3 Haswell die that it can sell and vice versa.

https://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/3

Crystal Well is the codename for the L4 cache, a discrete eDRAM silicon die, which is featured in the high-end Iris Pro-equipped Intel Haswell microprocessors. The eDRAM silicon die is separate from the main Haswell die but is packaged together with it.

https://en.wikichip.org/wiki/intel/crystal_well

The extra on package die is the eDRAM

https://www.anandtech.com/show/10281/intel-adds-crystal-well-skylake-processors-65w-edram

 

[Aerpoweron]

Thanks a lot Aquinus

I really assumed the smaller die was just a die shrink from the haswell die with small upgrades.

Do you know by any chance a GT3 or GT4 without eDRAM? I only know 128MB and 64MB ones.

 

[john_]

I believe it will be the complete opposite, dedicated graphics chips are becoming more and more expensive (to manufacture and to sell), I think APUs will slowly become more and more relevant.

Those millions that play Fortnite don't need a thousand dollar 2080ti, a 150$ APU is enough. The reason why this didn't happened yet is thanks to the years of atrociously crappy GPUs Intel stuffed inside their chips. Most people aren't even aware of the fact that video cards are no longer an absolute necessity even when it comes to light gaming.

Slowly but surely these systems will get integrated over time.

You are saying what I said, not the opposite.

The reason why it haven't happened yet, is because GPUs are really complex chips. So manufacturing had to really evolve to make them fit on a CPU socket and at the same time offer enough performance to make them useful. And of course system RAM had to become much faster to offer enough bandwidth. Even if Intel was at Nvidia's level in GPU performance, we wouldn't have been much better than what we are today with AMD's APUs. In my opinion, probably +30-50% more performance from an Intel APU with Nvidia Turing architecture for the GPU part, compared to a 2400G.

 

[Vya Domus]

You are saying what I said, not the opposite.

Then why did you said we will see the death of everything under 150$ ? Wont that include most APUs ?

 

[john_]

Then why did you said we will see the death of everything under 150$ ? Wont that include most APUs ?

You obviously misunderstood. I was talking about discrete graphic cards. Cards under $150 will be selling at much lower numbers if integrated graphics get much faster. And probably we will see cards under $200 to not offer such a great performance, making them a bad buy. Look at cards that cost $50-$100 today. 10 years ago a $50-$100 card was enough for low-mid settings gaming. Today a $50-$100 card is just a bad buy and money thrown in the trash for no reason.