Die size of Intel’s DG2-512 GPU based on the Xe-HPG graphics architecture has been measured by Twitter fellow, Locuza. The die shot was posted by Intel’s Chief Architect of graphics, Raja Koduri, in a tweet, yesterday.
Intel DG2-512 GPU Die Size Measured, Xe-HPG Architecture Based Chip Is Bigger Than AMD RDNA 2 & NVIDIA Ampere GPUs
According to Locuza, the Intel DG2-512 GPU should measure about 396mm2 which makes it bigger than the AMD RDNA 2 and NVIDIA Ampere offerings. The Xe-HPG architecture-based chip might be the flagship but in terms of performance, it should be competing against the RTX 3080 and Radeon RX 6800 XT. It’s definitely not as big as the flagship GPU offerings from either company but if we are to compare the high-end GPU category (Navi 22 / GA104), the DG2-512 is slightly bigger.
DG2-512 die is about 396mm² large.
DG2-512: 4096 FP32 Units, 256-Bit GDDR6
Navi21: 5120 FP32 Units, 256-Bit GDDR6 + 128MiB L3$
Navi22: 2560 FP32 Units, 192-Bit GDDR6 + 96MiB L3$
GA104: 6144 FP32 Units, 256-Bit GDDR6
— Locuza (@Locuza_) June 2, 2021
The DG2-512 GPU will come in the BGA-2660 package which measures 37.5mm x 43mm. NVIDIA’s Ampere GA104 measures 392mm2 which means that the DG2 chip is comparable in size while the Navi 22 GPU measures 336mm2 or around 60mm2 less. This isn’t the final die size of the chip but it should be very close. NVIDIA packs in tensor cores and much bigger RT/FP32 cores in its chips while AMD RDNA 2 chips pack a single ray accelerator unit per CU and Infinity Cache. Intel is expected to have hardware-accelerated ray-tracing capabilities onboard it’s Xe-HPG GPUs & a recent tweet from Raja also seems to point out hardware DL/ML capabilities that would assist in supersampling tech.
Intel Xe-HPG DG2 512 EU Discrete Gaming Graphics Cards Specifications
Each Xe-HPG based DG2 GPU SKU will come in various configurations which will range from the full-fat chip to several cut-down variants. This is similar to NVIDIA’s Ampere GA102-400, GA102-200 naming schemes, or AMD’s Navi 21 XTX, Navi 21 XT, Navi 21 XL naming conventions. The top DG2 512 EU variant has just one configuration listed so far and that utilizes the full die with 4096 cores, 256-bit bus interface, and up to 16 GB GDDR6 memory (8 GB GDDR6 listed too). Based on demand and yields, Intel could produce more variants of this flagship chip but we can’t say for sure right now.
The Xe-HPG DG2 512 EU chip is suggested to feature clocks of up to 2.2 GHz though we don’t know if these are the average clocks or the maximum boost clocks. Also, it is stated that Intel’s initial TDP target was 225-250W but that’s been upped to around 275W now. We can expect a 300W variant with dual 8-pin connectors too if Intel wants to push its clocks even further. We have also already seen leaked PCB and pictures of an ES Xe-HPG DG2 based graphics card which you can see here.
Intel Xe-HPG DG2 GPU Based Discrete Gaming Graphics Card Specs:
|GPU Variant||GPU SKU||Execution Units||Shading Units (Cores)||Memory Capacity||Memory Bus||TGP|
|Xe-HPG 512EU||DG2-512EU||512 EUs||4096||16/8 GB GDDR6||256-bit||~275W|
|Xe-HPG 384EU||DG2-384EU||384 EUs||3072||12/6 GB GDDR6||192-bit||TBC|
|Xe-HPG 256EU||DG2-384EU||256 EUs||2048||8/4 GB GDDR6||128-bit||TBC|
|Xe-HPG 192EU||DG2-384EU||192 EUs||1536||4 GB GDDR6||128-bit||TBC|
|Xe-HPG 128EU||DG2-128EU||128 EUs||1024||4 GB GDDR6||64-bit||TBC|
|Xe-HPG 96EU||DG2-128EU||86 EUs||768||4 GB GDDR6||64-bit||~120W|
Intel Xe-HPG DG2 512 EU Discrete Gaming Graphics Cards Performance & Features
Once again, these are just a few of several variants that will be utilizing the Xe-HPG architecture. The Intel DG2 GPUs will be coming to both desktop and mobility designs later this year and are rumored to be based on an external foundry process node, most probably TSMC and its 6nm (N6) node.
Besides that, some performance targets are also mentioned and the flagship Xe-HPG DG2 discrete graphics card is hinted to be close to the NVIDIA GeForce RTX 3080 which should put it close to the AMD Radeon RX 6800 XT too. The 3DMark TimeSpy scores vary a lot based on various samples and end up somewhere between the RTX 2080 and the RTX 3090. The performance metrics could be for differently configured DG2 SKUs.
It is stated that the company will have very good drivers but that depends on the timing of the cards itself as Intel could go two routes, either push the line back for drivers to mature or release them early but still offer good support but lack in terms of other features, similar to AMD’s RDNA 2. On the same front, Intel is said to offer its own ‘XeSS’ solution to rival DLSS and FSR. The encoding and prosumer capabilities are also going to be very impressive for Xe-HPG graphics cards. Intel has already confirmed support for hardware-accelerated ray-tracing, sampler feedback, & other DX12 Ultimate features on its Xe-HPG architecture.
— _rogame (@_rogame) June 2, 2021
Intel Xe-HPG DG2 512 EU Discrete Gaming Graphics Cards Availability & Pricing
The leaker states that the card is nowhere ready for a launch until Q4 2021 or even early next year (proper availability). No AIBs have received proper information about the cards yet as the cards are still receiving design tweaks by Intel themselves. Prosumer variants are also expected but they will not be ready till 2022. It looks like the 512 EU SKU will be heading out to the consumer segment-first followed by 128 EU SKUs shortly after. The leaker also states that the successor to DG2 GPUs will be known as Elasti ‘DG3’ and has a release date scheduled for 2023. A Q1 2022 release is hinted and good quantities are expected at launch.
As for pricing, the mid-range cards are expected to be priced at around $200-$300 US which is what Raja Koduri considers to be the sweet spot for gaming. There’s no mention of pricing for the higher-end variants but the lineup is expected to land in several variants with aggressive pricing starting at $349 and $up to $499 US. We can get a glimpse of the Xe-HPG DG2 GPU-powered cards at Intel’s Computex keynote next week.
Intel Xe-HPG DG2 GPU Specifications (Credits: Igor’s Lab)
|SKU 1||SKU 2||SKU 3||SKU 4||SKU 5|
|Supported Memory Technology||GDDR6||GDDR6||GDDR6||GDDR6||GDDR6|
|Memory speed||16 Gbps||16 Gbps||16 Gbps||16 Gbps||16 Gbps|
|Interface / bus||256-bit||192-bit||128-bit||64-bit||64-bit|
|Memory Size (Max)||16 GB||12 GB||8 GB||4 GB||4 GB|
|Smart cache size||16 MB||16 MB||8 MB||TBC||TBC|
|Graphics Execution Units (EUs)||512||384||256||196||128|
|Graphics Frequency (High) Mobile||1.1 GHz||600 MHz||450 MHz||TBC||TBC|
|Graphics Frequency (Turbo) Mobile||1.8 GHz||1.8 GHz||1.4 GHz||TBC||TBC|
|TDP Mobile (Chip Only)