Intel renames their DG2 discrete GPU the Alchemist

[magellan]

https://www.pcgamer.com/intel-dg2-gaming-graphics-card-release-date-specs-price-performance-xe-hpg/

The speculated release date is now Q1 2022.

Considering the continuing huge demand for GPU's I'd imagine as long as Intel picks a good price point for its performance I'd imagine they won't have any problems selling them.

I find it strange that this videocard is slated to release in Q1 2022 and there hasn't been any leaked indications of performance. Since Q1 2022 is only 5 months away, Intel should already have working prototype Alchemist videocards right?

Who is actually manufacturing the videocards or the actual GPU's for that matter? Intel doesn't have the facilities to manufacture videocards in house do they? Or fab GPU's?

 

[bmwbaxter]

Intel owns their own fabs. So I assume they will be making the GPU chips themselves.

 

[EarthDog]

There's been performance leaks...

Here's a couple...https://www.google.com/search?q=ijt...droid-verizon&sourceid=chrome-mobile&ie=UTF-8

Pretty sure these are fabbed with TSMC 7nm..

 

[mackerel]

TSMC N6 - no perf increase over N7 but more density.

When they ship next year, the Alchemist chips will be the first GPUs built on the N6 process. Though not TSMC’s leading process – that honor belongs to N5 – N6 is the next best choice to ship a GPU on – and likely the only economical one. N6 is essentially an improved version of TSMC’s N7 node, which replaces some DUV layers with EUV layers while retaining most of the same design rules and tools. TSMC hasn’t been promoting the half-node too much on a performance basis – performance is about equal to N7 – but it does take some important steps to simplifying the manufacturing process versus N7. And, especially important for a relatively large die product like a GPU, it offers around 18% improved density versus N7.

https://www.anandtech.com/show/16895/a-sneak-peek-at-intels-xe-hpg-gpu-architecture

Also, using the numbers from Anandtech top model would be about equal in perf to a 3060 Ti for FP32 workloads. This doesn't necessarily scale to gaming so give it a wide tolerance.