Intel is rumored to be producing their next-generation Xe graphics structure based mostly DG2 GPU on TSMC’s 7nm course of node. Whereas Intel has already previewed their Xe Low Energy ‘DG1 GPU’ and the Xe HPC ‘Ponte Vecchio GPU’, the one factor that is lacking is the Xe Excessive-Efficiency GPU or extra generally identified to us because the DG2.
Intel’s Excessive-Finish DG2 GPU Rumored To Home 2nd Era Xe Graphics Structure, Based mostly on TSMC’s 7nm Course of Node
This may not be the primary time we’re listening to about Intel’s Xe DG2 GPU, it first appeared final 12 months in check drivers alongside the DG1 GPU. There have been no particulars aside from what gave the impression to be the EU depend of the three variants that received revealed and the truth that DG2 makes use of Xe HP (Excessive-Efficiency) microarchitecture as an alternative of the Xe LP (Low-Energy) which is extra suited to entry-tier designs such because the DG1 GPU.
Coming again to the rumor, AdoredTV experiences that their sources have gotten info that Intel can be introducing their DG2 GPU in 2022 and can be based mostly on the 7nm course of node. What’s extra attention-grabbing is that AdoredTV mentions that Intel can be counting on TSMC’s 7nm course of node for the manufacturing of DG2 GPU and never their very own in-house 7nm EUV expertise. It’s stated that Intel can be masking up the losses and the hit backs they needed to face with 10nm course of with their 7nm tech nevertheless, this is able to solely be related for CPUs, not GPUs.
Intel’s 7nm can be the driving power behind their next-generation CPUs (server, desktop, cellular). This sort of interferes with what Intel themselves have deliberate out for his or her 7nm course of node as they’d revealed that the 7nm lead product can be a Xe structure based mostly GP-GPU utilizing their Foveros packaging for datacenter AI & HPC market. We all know that chip now as Ponte Vecchio so it’s possible that whereas the lead 2021 product based mostly on Intel’s 7nm course of node would certainly be Xe HPC, they’d nonetheless be prioritizing a big portion of their fabs to CPUs because the Xe HPC class GPU is a really area of interest product which does not have to take up as a lot 7nm quantity as say a mainstream line of GPUs corresponding to Xe LP or Xe HP.
However right here comes the extra attention-grabbing half, Intel can be counting on TSMC’s 7nm course of, not the 7nm+ which has EUV. TSMC’s 7nm+ EUV course of hit quantity manufacturing within the second quarter of 2019. This is perhaps the opposite motive why Intel goes the usual 7nm route fairly than their very own 7nm (EUV) or TSMC’s 7nm+ node as the prices related to them is way greater and provide points within the first iteration of a brand new course of node from Intel are already obvious from the 10nm roadmap. So it makes extra sense to prioritize CPUs and low-volume datacenter chips to 7nm than a full-fledge mainstream and high-end line of GPUs proper now.
Only for comparability’s sake, Intel would proceed to supply optimizations of the 7nm course of node with 7nm+ in 2022 and 7nm++ in 2023. Similar to 10nm, 7nm would ship an excellent checklist of additives over 10nm which would come with:
- 2x density scaling vs 10nm
- Deliberate intra-node optimizations
- 4x discount in design guidelines
- Subsequent-Gen Foveros & EMIB Packaging
7nm Standing: Renduchintala offered first updates on Intel’s 7nm course of expertise that may ship 2 instances scaling and is predicted to offer roughly 20 % enhance in efficiency per watt with a four instances discount in design rule complexity. It’s going to mark the corporate’s first business use of utmost ultraviolet (EUV) lithography, a expertise that may assist drive scaling for a number of node generations.
The opposite necessary element talked about by AdoredTV is that DG2 goes to be powered by Intel’s 2nd era Xe graphics structure. And having a 2022 launch implies that it could be positioned in opposition to next-generation GPUs by NVIDIA and AMD that are aiming to make the most of extra superior nodes from TSMC. And each Ampere and 2nd Era Navi would have already been out there for months if not a complete 12 months by the point DG2 launches which implies that earlier experiences that Intel would goal the finances and entry-level phase may find yourself being true.
It’s possible that the event of the DG2 GPU continues to be not full as DG1, the primary era Xe graphics chip will begin delivery to ISV’s this quarter with a launch aimed on the year-end. However that is additionally the second report that touts Intel’s 7nm GPU launch in 2022. The earlier rumor acknowledged quite a lot of the explanation why Intel’s Xe GPUs will not make it on time for launch corresponding to:
- Progress Is not Going Effectively on Xe GPU
- Ponte Vecchio HPC GPU Might Not Launch Inside the Subsequent 2 Years (2022)
- The effectivity of Xe GPU is de facto low in comparison with rivals
- No AIB Assist on Launch, solely reference fashions for discrete graphics playing cards offered by Intel
- Drivers want loads of time to enhance
Following are the checklist of Xe DG* GPUs which have thus far appeared in Intel’s check drivers:
- iDG1LPDEV = “Intel(R) UHD Graphics, Gen12 LP DG1” “gfx-driver-ci-master-2624”
- iDG2HP512 = “Intel(R) UHD Graphics, Gen12 HP DG2” “gfx-driver-ci-master-2624”
- iDG2HP256 = “Intel(R) UHD Graphics, Gen12 HP DG2” “gfx-driver-ci-master-2624”
- iDG2HP128 = “Intel(R) UHD Graphics, Gen12 HP DG2” “gfx-driver-ci-master-2624”
As soon as once more, Intel has confirmed two merchandise for now based mostly on their Xe graphics structure, each of that are based mostly on two totally different generations of Xe structure. This rumor says loads however there’s nothing confirmed as of but. Intel is all set to unveil particulars of their Xe graphics structure at GDC 2020 in March so we’ll in all probability hear extra info from them quickly.
Share Tweet Submit