New Details Emerge About Intel Meteor Lake Graphics
Intel's Meteor Lake iGPU expected to feature 1024 stream processors.
Intel's next generation processor, codenamed Meteor Lake could offer iGPU compute performance comparable to that of the company's Arc A380/A370M discrete GPU, if newly leaked information about configuration of the upcoming CPUs is accurate.
The information comes from Golden Pig Upgrade Pack (hat tip to VideoCardz), which tends to be accurate. Still, this is an unofficial source, so everything should be taken with a grain of salt. Meanwhile, the new details seemingly corroborate previously leaked information.
Rumored Specifications of Intel's 14th Generation Core CPUs
PL1 TDP | Core Configuration | Total Cores | iGPU | Silicon | Package |
7W | 1P + 8E | 5 to 9 | 3 Xe - 4 Xe | Meteor Lake | M |
9W | 2P + 8E | 6 to 10 | 3 Xe - 4 Xe | Meteor Lake | M |
15W | 4P + 8E | 6 to 12 | 3 Xe - 8 Xe | Meteor Lake | M |
28W | 6P + 8E | 10 to 14 | 7 Xe - 8 Xe | Meteor Lake | P |
45W | 6P + 8E | 12 to 14 | 8 Xe | Meteor Lake | H |
55W | 8P + 16E | 14 to 24 | 32 EU | Raptor Lake | HX |
Qualification samples under test by Intel's partners feature up to six high-performance Redwood Cove cores operating at 4.80 GHz, eight energy-efficient Crestmont cores, two low-power Crestmont cores, and an integrated GPU based on the Xe-LPG architecture with up to 128 execution units (which equals to 1024 stream processors as well as 8 Xe clusters) that can boost to 2.20 GHz. The part features a configurable TDP between 20W and 65W. Meanwhile, the Core Ultra 9 part could feature a maximum CPU clock of 5.0 GHz or higher.
Based on what we know about Intel's Xe-LPG architecture, we can expect Intel's upcoming iGPU to feature compute performance of around 4.5 FP32 TFLOPS, which is higher than performance of Intel's standalone Arc A380/A370M GPU (around 4.2 FP32 TFLOPS). Of course, discrete GPUs have the advantage of memory bandwidth, so it remains to be seen how close Intel's new iGPU can get to the company's current dGPU. Meanwhile, the new iGPU promises to offer higher compute horsepower than AMD's integrated Radeon 780M, which is theoretically capable of 4.3 FP32 TFLOPS.
One of the things that multi-chiplet architecture promises with Intel client CPUs is to enable considerably higher performance of integrated GPU now that the company is not limited with iGPU die size. If the information about configuration and clocks of Meteor Lake's iGPU is correct, then Meteor Lake could indeed offer high performance in games. The CPU will naturally not be able to compete against the best graphics cards, but it will considerably increase the performance of integrated graphics when compared to Intel's existing iGPUs.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.
-
sherhi That would be cool, AMD's have been tested and it looked solid for older games...which is not particularly strong side of Intel GPUsReply -
ikjadoon The part features a configurable TDP between 20W and 65W.
Seems like a P CPU, which are 28W PL1 / 64W PL2. Sad to see high boost returning; I guess 15W+ per-core peak power & 5 GHz boost clocks are Intel's love affair it can't leave.
15W PL1 → 55W (holy <Mod Edit>) PL2
28W PL1 → 64W (seriously?) PL2
This is desktop heat output under heatsinks thinner than your finger. Judging from history, Windows OEMs ignore the 7W and 9W lineup because it's so slow.
Looks like "lifestyle company" Apple can design much higher 1T perf/W and we've not even seen Qualcomm's Oryon yet, which should ship close to MTL.
Bets that Intel will be firmly at the bottom of the 1T perf / W list?
Intel Meteor Lake - Intel 4
AMD Zen4 mobile / "Phoenix" - TSMC N4
Apple M3 - TSMC N3
Qualcomm Oryon - presumably N4 -
usertests The 1P + 8E configuration is new to me. I don't think the Celeron/Pentium 1P + 4E parts got much uptake.Reply -
bit_user No mention of the rumored L4 cache?Reply
https://www.tomshardware.com/news/intel-patent-reveals-meteor-lake-adamantine-l4-cache
That could make up for a lot of the missing bandwidth, relative to the dGPU equivalent (A380). -
bit_user
Yes, according to this:cknobman said:I thought the 780M, like the Z1 Extreme, was capable of up to 8.6tflops?
https://www.techpowerup.com/gpu-specs/radeon-780m.c4020 -
TerryLaze
The thing with intel is that they also factor in the power draw of the graphics, there is no having the CPU run slower when the iGPU runs or any of that crap going on.ikjadoon said:Seems like a P CPU, which are 28W PL1 / 64W PL2. Sad to see high boost returning; I guess 15W+ per-core peak power & 5 GHz boost clocks are Intel's love affair it can't leave.
15W PL1 → 55W (holy <Mod Edit>) PL2
28W PL1 → 64W (seriously?) PL2
The CPU alone will be using less than that.
You can already see from the desktop SKUs, the CPUs have a much higher PL2 rating than what they can achieve.
The 13400 for example is rated for 154W PL2 but doesn't even reach 80W in heavy loads, even with limits removed it only reaches 86 in blender.
https://www.techpowerup.com/review/intel-core-i5-13400f/21.html -
irish_adam
That has to be the most extreme cherry picking I have ever seen. It doesn't hit its PL2 because Intel is lazy and just copy/pastes specs.TerryLaze said:The thing with intel is that they also factor in the power draw of the graphics, there is no having the CPU run slower when the iGPU runs or any of that crap going on.
The CPU alone will be using less than that.
You can already see from the desktop SKUs, the CPUs have a much higher PL2 rating than what they can achieve.
The 13400 for example is rated for 154W PL2 but doesn't even reach 80W in heavy loads, even with limits removed it only reaches 86 in blender.
https://www.techpowerup.com/review/intel-core-i5-13400f/21.html
You'll find Intel CPU's do hit their PL2 and beyond even in CPU only benchmarks. You'd have to be completely new to the scene to think otherwise.
I'll just leave this here...
13700k review -
ikjadoon TerryLaze said:The thing with intel is that they also factor in the power draw of the graphics, there is no having the CPU run slower when the iGPU runs or any of that crap going on.
The CPU alone will be using less than that.
You can already see from the desktop SKUs, the CPUs have a much higher PL2 rating than what they can achieve.
The 13400 for example is rated for 154W PL2 but doesn't even reach 80W in heavy loads, even with limits removed it only reaches 86 in blender.
https://www.techpowerup.com/review/intel-core-i5-13400f/21.html
The iGPU consumes <2W; for TDP #s, Intel ensures the iGPU runs at very low clocks. It's easy to double-check: IGP power draw is independently measured by most software tools.
//
In desktop, yes: Intel sandbags its TDPs and has for many years. See the "35W" Haswell i3-4130T; it only consumed 26W in peak power. So you "saved" 9W or 26% power!
In mobile, it's a different story: Intel requires its entire PL2 budget to compete with AMD & Apple.
i7-1265U in Cinebench R23 Single15W - 1600 pts
20W - 1658 pts
28W - 1761 pts
This is just one core's power consumption. For +10% perf, Intel swallowed up to +86% more power.
//
Compare this to Apple, where the M2 scored 1585 pts. Intel requires +150% more power for the same performance.
Intel will need 2-3 more years maybe to reach the Apple M1 efficiency at high perf levels.
Under ~40W, Intel has struggled for 10+ years now. Today, it's Samsung Exynos bad. New nodes only help a little as they take 2-3 years but only offer incremental improvements over an already not-great node: it's gotta be a new uArch & design target. -
TerryLaze
But that was ikjadoon only argument, he took the copy/pasted PL2 specs and acted as if they where measured results.irish_adam said:It doesn't hit its PL2 because Intel is lazy and just copy/pastes specs.