To begin, the MTL-H and MTL-P variants are for high performance. The MTL-U version is low-power, high efficiency. That bracketing is unchanged from Alder Lake. The more powerful H-and-P series will max out at 14 cores, with the U-series gaining two more cores over Alder Lake for a maximum of 12. One thing that seems new is the addition of LP-E cores. On Alder Lake there were just “big” P-cores and “little” E-cores. Now there appears to be a new version of E-cores, with increased efficiency.
It’s not clear if these different types of E-core share the same design, with the LP-E type being binned or not. However, our guess is that they are just binned versions of the regular E-cores. Also, Igor says Intel will put the GPU on top of the CPU with its Foveros stacking technology. However, the diagrams Intel has published previously shows them side-by-side, so we’re not sure where Igor is getting that from.
Stacking the GPU on top of the CPU may be possible, but it depends on the specifics of the circuit layout. While AMD’s 5800X3D stacks a large vertical L3 cache, AMD has positioned the cache so it doesn’t sit directly over any CPU hot spots. Intel would need to do the same, while accounting for the fact that an on-die GPU can be a significant source of heat in its own right. This isn’t impossible, but if true, it’s a change compared to what Intel has previously disclosed.
Another notable item is the lack of support for DDR4 memory. It was previously leaked that Intel’s 13th generation Raptor Lake will support both DDR4 and DDR5. It would appear that will be the final stop for the aging memory standard, with Meteor Lake being DDR5 only. Along those lines, it says it supports LPDDR5X-7467, which is faster than the DDR5-5600 it will offer with Raptor Lake. The diagram notes it will also support that slower memory type too. When using LPDDR5 it supports up to 64GB of RAM. That’s bumped up to 96GB when using DDR5-5600.
Igor’s Lab also posted some spec sheets as well as the block diagram. They show the GPU will feature up to 128 execution units, probably using Intel’s Battlemage architecture. That’s quite a lot of EUs, but it will be going up against AMD’s Dragon Range. AMD claims its upcoming mobile platform will allow it to achieve “gaming dominance” over Intel, so that will be a cage match for sure. The GPU connects to the SoC with a single x8 PCIe 5.0 channel, which is interesting. Intel is probably not using a x16 channel to conserve energy. For storage it’s using 12 lanes of PCIe 4.0. Again, that could be a power-saving move. WiFi 7 hardware might be available by 2023, but Intel is sticking with WiFi 6E with Raptor Lake (assuming these slides are true).
There’s also mention of an Extended Gaming Mode,but no particular mention of what it might be. This could be a new iteration of power-shifting technology between the CPU and GPU, or another type of power consumption optimization intended to distribute resources more efficiently.
Under the section labeled ‘AI’, there’s mention of an onboard VPU. This is a term Intel currently uses to refer to Movidius-branded AI processors. It is not clear if the VPU mentioned here would be a further iteration of the company’s further Movidius products, or an altogether new processor. Intel hasn’t had much to say about Movidius for several years, and it’s not entirely clear if the “Keem Bay” third generation Movidius architecture ever actually shipped.
Meteor Lake is due to arrive in the second half of 2023. Given that we’ve already seen pictures of it (top of article) and that these slides exist, we can surmise it’s quite far along in development. Also, Intel insiders recently confirmed the company is on track with its Intel 4 process. This is what it’s using for Meteor Lake, and which is supposed to go into high volume production later this year.
Now Read:
- Intel Rumored to Cut Alder Lake Prices as Recession Fears Increase
- Intel Flashes Arc Desktop Graphics Card at Texas Gaming Tournament
- Intel Shows Off Tile-Based Meteor Lake, Ponte Vecchio CPUs at Vision Event
No comments:
Post a Comment