Intel Meteor Lake iGPU Reportedly Boosts up to 2.2 GHz

[T0@st]

Chinese tech tipster Golden Pig Upgrade has turned its attention to Intel's Meteor Lake processor series—they believe that hardware partners are putting qualification samples (QS) through the ringer. The short Bilibili social media post proposes that these laptop-oriented prototypes sport six high-performance Redwood Cove cores running at 4.80 GHz, eight energy-efficient Crestmont cores, as well as two low-power Crestmont cores. Apparently the unit's TDP can be adjusted—starting at 20 W, and going up to 65 W. Golden Pig Upgrade was also informed about a Core Ultra 9 model capable of hitting 5.0 GHz (or greater) maximum CPU clocks.

The qualification sample's integrated graphics processing unit is reportedly based on the Xe-LPG architecture—some experts reckon that this could offer performance (4.5 FP32 TFLOPS) comparable to Arc A380 or A370M discrete solutions. The sampled iGPU could feature up to 128 execution units (so equivalent to 1024 stream processors, as well as 8 Xe clusters)—insider sources allege that units can boost to a maximum of 2.2 GHz. It will be interesting to observe how this will stack up against AMD's Radeon 780M iGPU—which is theoretically in a similar ballpark (4.3 FP32 TFLOPS).



View at TechPowerUp Main Site | Source

 

[Dan.G]

I'm eager to see reviews! AMD might have the upper hand regarding drivers, though.

 

[tabascosauz]

some experts reckon that this could offer performance (4.5 FP32 TFLOPS) comparable to Arc A380 or A370M discrete solutions.

When has raw TFLOPS ever been relevant to iGPU performance? 680M and 780M still haven't caught up to the GTX 1650 or A380 yet.

On the Radeon side core config has always been more important the clocks (compare 680M and 780M). Remains to be seen how good the new arch is, how well EMIB handles bandwidth, and (like AMD) how generous Intel is with their 128EU config in not-flagship SKUs.

 

[SRB151]

Yeah, makes sense. The question I have is how much power does this thing draw. That's been their cpu solution, just keep overclocking.

 

[lemonadesoda]

Overclocking iGPU. Hot stuff.

 

[Nater]

Intel has some serious Chinese spying going on.

 

[Darmok N Jalad]

that these laptop-oriented prototypes sport six high-performance Redwood Cove cores running at 4.80 GHz, eight energy-efficient Crestmont cores, as well as two low-power Crestmont cores.

3 different core types?!

 

[tabascosauz]

3 different core types?!

What's there to be shocked about? All dynamiQ SoCs have been differentiating physically identical low/middle and middle/high cores like this since forever. Windows scheduler has also been unofficially trying this with 2CCD parts for a while. Intel doesn't put E-cores in their efficiency band for a lot of SKUs, so why not downclock a few to serve a low power role?

 

[Darmok N Jalad]

What's there to be shocked about? All dynamiQ SoCs have been differentiating physically identical low/middle and middle/high cores like this since forever. Windows scheduler has also been unofficially trying this with 2CCD parts for a while. Intel doesn't put E-cores in their efficiency band for a lot of SKUs, so why not downclock a few to serve a low power role?

It's the first I heard about it. Seems like it could get even more tricky for Windows for when to target each. There's already some complexity there with 2 core types, and I can't say I've ever seen 3 core types in a consumer space design.

 

[pressing on]

It's the first I heard about it. Seems like it could get even more tricky for Windows for when to target each. There's already some complexity there with 2 core types, and I can't say I've ever seen 3 core types in a consumer space design.

The existence of 2 LPE cores on the SOC of Meteor Lake first surfaced in a diagram published by Intel as part of a patent application for its L4 cache. The exact purpose of the LPE cores is not known at this stage. The position of these cores on the SOC suggests that they are not general purpose cores but could have a specialist function to do with power management or they could be linked to the Neural Processing Unit that is also located on the SOC.

 

[Darmok N Jalad]

The existence of 2 LPE cores on the SOC of Meteor Lake first surfaced in a diagram published by Intel as part of a patent application for its L4 cache. The exact purpose of the LPE cores is not known at this stage. The position of these cores on the SOC suggests that they are not general purpose cores but could have a specialist function to do with power management or they could be linked to the Neural Processing Unit that is also located on the SOC.

Thanks. I thought I had been keeping up, but this one escaped me. I know Apple's chips also have performance and efficiency cores, the iGPU, but also the Neural Engine. The NE is very handy in some applications, but developers have to target it in order to make use of it, I believe. I suppose this is what we'll also start seeing from AMD and Intel, and it can be a really good way to get real performance gains. I can think of one personal example, where a RAW "AI Denoise" process took 20 seconds on the M1 GPU, but when devs targeted the M1's NE, it dropped to 8 seconds (which is about what my RX 480 can do for the same process). It's the most demanding thing I do these days, so I welcome any advancement there were it uses way less power and takes way less time than more traditional means. If that's what these 2 extra cores help serve, that makes good sense, maybe translating and setting up the code?

 

[Minus Infinity]

I think Meteor Lake mobile will put AMD Phoenix in a large spot of bother assuming the leaks about improved efficiency and IPC uplifts of the new Redwood cove (p-cores) and Crestmont (e-cores). Combine that with the 128EU Alchemist+ iGPU in top tier models and it should be very nice laptop APU. I'm not in the market yet for a new laptop, but even if I were I would not rush out and buy Phoenix and would definitely wait for Meteor Lake reviews. We are led to believe ML will use 40% less power than Raptor Lake. Fingers crossed Intel can actually deliver on the hype and timetable. The other thing is their iGPU drivers are trash so they need a quantum leap in that area.

 

[kapone32]

I think Meteor Lake mobile will put AMD Phoenix in a large spot of bother assuming the leaks about improved efficiency and IPC uplifts of the new Redwood cove (p-cores) and Crestmont (e-cores). Combine that with the 128EU Alchemist+ iGPU in top tier models and it should be very nice laptop APU. I'm not in the market yet for a new laptop, but even if I were I would not rush out and buy Phoenix and would definitely wait for Meteor Lake reviews. We are led to believe ML will use 40% less power than Raptor Lake. Fingers crossed Intel can actually deliver on the hype and timetable. The other thing is their iGPU drivers are trash so they need a quantum leap in that area.

Somehow I cannot see that. Right now the whole internet is awash with gushing on that APU from AMD. The USB 4 support will also be a popular update. Hopefully USB 4 E GPU enclosures will not be as expensive as TB3 ones.

 

[R0H1T]

We are led to believe ML will use 40% less power than Raptor Lake.

It's always "up to" with those claims, there's also rumors of another AMD APU with RDNA 3+ IGP & they could achieve similar efficiency gains in a handful of tasks. No one's getting that kind of a gain in a single gen unless you're comparing something like a 13900HX with 14xxxU or something.

 

[Minus Infinity]

It's always "up to" with those claims, there's also rumors of another AMD APU with RDNA 3+ IGP & they could achieve similar efficiency gains in a handful of tasks. No one's getting that kind of a gain in a single gen unless you're comparing something like a 13900HX with 14xxxU or something.

Well sure something always better is coming, but Meteor Lake is being talked about from Intel and we have plenty of real information. We have little details of next years Strix APU, so all one can do is compare to current barely themselves just released, Phoenix 7x40/45 APUs. Those next gen AMD APU's will be also fighting whatever Arrow Lake brings. I think Meteor Lake will give Intel 6 months or so of dominance before we see Strix.

 

[R0H1T]

I'd be slightly surprised if their first chiplet tiled implementation lives up to most people's expectations, but Intel's done better than that in the (distant) past so it can certainly happen.

 

[Dan.G]

It's nice to see iGPUs getting better, but they're still limited by RAM bandwidth. I always thought that's what's holding them back and it's, probably, why AMD and Intel don't bother too much with iGPUs. They could easily release a 24 CUs / 256 EUs iGPU, but it wouldn't be able to deliver its' full potential. Also, would like to see nVidia graphics card without dedicated VRAM - similar to an iGPU. It could be solely powered through PCI-Ex and would use system RAM. A 2048 CUDA cores AD108 (let's call it) that draws less than 75 W and would look like a "Founder's Edition" GT 640 (well, without the VRAM chips). I'd like that. You would have as much "VRAM" as you'd want. That's what I like about iGPUs: you never run out of "VRAM". Got 16 GB of system RAM and it's not enough on your iGPU desktop? No problem: just slap 16 GB more and you're done!

 

[TheinsanegamerN]

Somehow I cannot see that. Right now the whole internet is awash with gushing on that APU from AMD. The USB 4 support will also be a popular update. Hopefully USB 4 E GPU enclosures will not be as expensive as TB3 ones.

USB4 is TB3, so IDK why you would think enclosures would be cheaper.

 

[persondb]

That would roughly be in the ballpark of the 780M, except in the(very rare) scenario where it can dual issue, then the compute throughput would be greater(for what's worth).

 

[Darmok N Jalad]

It's nice to see iGPUs getting better, but they're still limited by RAM bandwidth. I always thought that's what's holding them back and it's, probably, why AMD and Intel don't bother too much with iGPUs. They could easily release a 24 CUs / 256 EUs iGPU, but it wouldn't be able to deliver its' full potential. Also, would like to see nVidia graphics card without dedicated VRAM - similar to an iGPU. It could be solely powered through PCI-Ex and would use system RAM. A 2048 CUDA cores AD108 (let's call it) that draws less than 75 W and would look like a "Founder's Edition" GT 640 (well, without the VRAM chips). I'd like that. You would have as much "VRAM" as you'd want. That's what I like about iGPUs: you never run out of "VRAM". Got 16 GB of system RAM and it's not enough on your iGPU desktop? No problem: just slap 16 GB more and you're done!

I'm waiting for AMD to drop a good chunk of Infinity Cache on their monolithic chips. It would probably deliver really good performance. The challenge is that since both AMD and Intel also hope to sell GPUs, they probably don't want to make the iGPU too good, as it might start robbing sales. Intel has less to lose than AMD. I suppose pricing could fix that, but then they won't get as many design wins.

 

[Pizdarenkowitch]

Why people till these days still keeps buying all the POS from Intel,especially in this period ???
Isn't a wiser choice to wait until 2024 to finally ride RibbonFet+PowerVia tech ???

P.S LOL peeps i envy you,because your pockets is pretty fatty if you like to spend your money like this.

 

[TheinsanegamerN]

Why people till these days still keeps buying all the POS from Intel,especially in this period ???

Because they are perfectly functional, widely available, and computers are bought for support form their OEM, not the brand inside. That only matters to enthusiasts like us.

Isn't a wiser choice to wait until 2024 to finally ride RibbonFet+PowerVia tech ???

Just Wait. For whatever that is supposed to be.

P.S LOL peeps i envy you,because your pockets is pretty fatty if you like to spend your money like this.

Believe it or not there are people who need a replacement PC this year. So many people bring me 10+ year old machines and want them fixed up.

 

[SRB151]

I think Meteor Lake mobile will put AMD Phoenix in a large spot of bother assuming the leaks about improved efficiency and IPC uplifts of the new Redwood cove (p-cores) and Crestmont (e-cores). Combine that with the 128EU Alchemist+ iGPU in top tier models and it should be very nice laptop APU. I'm not in the market yet for a new laptop, but even if I were I would not rush out and buy Phoenix and would definitely wait for Meteor Lake reviews. We are led to believe ML will use 40% less power than Raptor Lake. Fingers crossed Intel can actually deliver on the hype and timetable. The other thing is their iGPU drivers are trash so they need a quantum leap in that area.

I guess Intel finally got their foundry straight a few years early. 20% uplift in multi core performance, more cores, and a 40% reduction in power usage all in one generation. TSMC can't touch it. Just wait until the next gen. Probably do it again and really take off. Why is it rumored that they'll use TSMC for their next cpu? They have this cold by their own numbers.

 

[R0H1T]

Oh sure 20% more performance & 40% less power, so 2x as efficient? I guess unicorns must be real as well, just as Barbie is

Spoiler

 

[hs4]

Redwood Cove and Crestmont will not improve IPC. This can be inferred from the Intel Vision 2022 exhibits. There will be only one of the following: 20% performance improvement with the same number of cores and the same power consumption, 40% more cores with the same power, or 40% less power with the same number of cores and the same clocks.

Lunar lake will be the true unicorn we can touch; how Arrow lake will be treated remains to be seen.