Intel ARC GPUs, Xe Architecture for dGPUs [2018-2022]

[DavidGraham]

Built on 10nm and scalable for the entire market.

 

[Deleted member 2197]

Your IMG doesn't display.

 

[AlphaWolf]

Is that 'graph' actually supposed to be indicative of the performance of the various skus?

 

[giannhs]

a typical intel graph

literally no one knows wtf they are saying

 

[Dayman1225]

Probably better to talk about Gen11 graphics since they actually have given details on that like Tile based rendering and what not

 

[Dayman1225]

The Aurora Supercomputer, the USs first (and possibly worlds first) ExaFLOP/exascale system will be using Intels Xe Graphics Architecture aswell as Intels CPUs, Optane PM and Software (OneAPI)

Intel's Xe Graphics Architecture to Power DOE's First Exascale Supercomputer

Now we know why it’s callled Xe, for Xescale

 

[Alexko]

Or exponential growth? As in x^e.

 

[Alexko]

Or exponential growth? As in x^e.

Wait, no, that would be e^x, brain fart. For some reason I can't edit or delete my post, and now everyone will now how stupid I am until the end of time.

 

[Alexko]

Wait, no, that would be e^x, brain fart. For some reason I can't edit or delete my post, and now everyone will now how stupid I am until the end of time.

*Know.

Goddammit.

 

[JasonLD]

I assume both Xeon and Xe dGPU would be based on 7nm. Since they are shooting for at least 5 times of Summit's processing power, I don't think it would be achievable on 10nm, assuming similar total power target.

 

[Deleted member 13524]

I assume both Xeon and Xe dGPU would be based on 7nm. Since they are shooting for at least 5 times of Summit's processing power, I don't think it would be achievable on 10nm, assuming similar total power target.

I think Intel's 10nm is closer to TSMC's 7nm than their (or Samsung's) 10nm node.

 

[Entropy]

I think Intel's 10nm is closer to TSMC's 7nm than their (or Samsung's) 10nm node.

Initial information from Intel would suggest so. It is quite likely still the case, more or less. On the other hand, it is difficult to imagine that Intel would insist on a process that just hasn't worked out for them, so it is widely rumoured and assumed that the process we will finally see widely deployed will differ from the initial plans. In what ways, and to what extent, nobody outside intel knows or shares. It will be very interesting when a third party analysis goes public, to see whatever changes had to be made (if any) to make the process viable. (But by the way their PR spins things, we could see their 7nm products before their 10nm ever gets to broad volume production. )

 

[JasonLD]

I think Intel's 10nm is closer to TSMC's 7nm than their (or Samsung's) 10nm node.

Yeah, I know that. Since Aurora is expected to be completed in late 2021, I assumed it might be using Intel’s future 7nm Euv instead of their upcoming 10nm. Intel’s 7nm should be closer to TSMC’s and Samsung’s future 3nm.

 

[Dayman1225]

Yeah, I know that. Since Aurora is expected to be completed in late 2021, I assumed it might be using Intel’s future 7nm Euv instead of their upcoming 10nm. Intel’s 7nm should be closer to TSMC’s and Samsung’s future 3nm.

To be fair it’s being installed in 2021 so unless Intel is ramping new 7nm chips in late 2020 it’ll likely be 10nm based stuff. According to Toms it’s being “stood up” in early 2021 and will be fully operational by the end of 2021

 

[DavidGraham]

Intel's next dGPU to support hardware ray tracing.

https://wccftech.com/intel-xe-gpu-supports-ray-tracing/

 

[iroboto]

Intel's next dGPU to support hardware ray tracing.

https://wccftech.com/intel-xe-gpu-supports-ray-tracing/

it has begun, the ray tracing wars has

 

[Ike Turner]

This is some nice & tasty nothing burger...

GPU acceleration for Intel Embree & OSPRay (currently CPU only via AVX2 /SSE2) most probably via OpenCL or Vulkan.

The exact Intel quote

"Xe architecture roadmap for data center optimized rendering includes ray tracing hardware acceleration support for the Intel® Rendering Framework family of API’s and libraries."

Fanboy clickbait interpretation:

"Intel's next dGPU to support hardware ray tracing."

EDIT: Here's the full PR and it doesn't sound like anything more that Intel porting Embree & OSPRay to GPU and that they are treating RT as "as a general computational technique" (similar wording to Microsoft's DXR announcement btw). I don't see anything hinting at some "RT Core" like implementation...

Intel® architecture processors are the flexible, large memory capable, performance engines that drive the end-to-end creative process for visual effects and animated feature films. Today’s available GPUs have architecture challenges like memory size limitations and performance derived from years of honing for less sophisticated, “embarrassingly parallel” rasterized graphics use models. Studios continue to reach for maximum realism with complex physics processing for cloth, fluids, hair and more, plus modeling the physics of light with ray tracing. These algorithms benefit from mixed parallel and scalar computing while requiring ever growing memory footprints. The best solutions will include a holistic platform design where computational tasks are distributed to the most appropriate processing resources.

David Blythe’s recent blog provided initial insights into our exciting new Intel® Xe architecture currently under development. We are designing the Intel® Xe architecture as a cohesive acceleration companion to our continuing roadmap of Intel® Xeon® processors. As David closed his blog he mentioned, “We will look forward to sharing more details on the Intel® Xe architecture in the months ahead.” I’m pleased to share today that the Intel® Xe architecture roadmap for data center optimized rendering includes ray tracing hardware acceleration support for the Intel® Rendering Framework family of API’s and libraries.

Your existing investments in graphics and rendering solutions based on Intel® Rendering Framework open source products will seamlessly map to the exponential performance benefits of these flexible accelerated platforms. Further, ray tracing as a general computational technique for a variety of simulation computation beyond rendering is rapidly growing. To put it succinctly in my own words “Leave no transistor behind” by creating a holistic software and compute environment ready to maximize your workflow for exponential benefits.

 

[Malo]

Yeah that's basically saying the GPUs will be supported for the Intel framework API.

Fanboy clickbait interpretation:

Which WCCF specializes in and yet they're still spread everywhere.

 

[OCASM]

This is some nice & tasty nothing burger...

GPU acceleration for Intel Embree & OSPRay (currently CPU only via AVX2 /SSE2) most probably via OpenCL or Vulkan.

The exact Intel quote

"Xe architecture roadmap for data center optimized rendering includes ray tracing hardware acceleration support for the Intel® Rendering Framework family of API’s and libraries."

Fanboy clickbait interpretation:

"Intel's next dGPU to support hardware ray tracing."

EDIT: Here's the full PR and it doesn't sound like anything more that Intel porting Embree & OSPRay to GPU and that they are treating RT as "as a general computational technique" (similar wording to Microsoft's DXR announcement btw). I don't see anything hinting at some "RT Core" like implementation...

Well, NVIDIA mentioned that the Turing RT cores could also be used for physics and audio simulations so it can be interpreted either way

 

[keldor]

This is some nice & tasty nothing burger...

GPU acceleration for Intel Embree & OSPRay (currently CPU only via AVX2 /SSE2) most probably via OpenCL or Vulkan.

The exact Intel quote

"Xe architecture roadmap for data center optimized rendering includes ray tracing hardware acceleration support for the Intel® Rendering Framework family of API’s and libraries."

Fanboy clickbait interpretation:

"Intel's next dGPU to support hardware ray tracing."

EDIT: Here's the full PR and it doesn't sound like anything more that Intel porting Embree & OSPRay to GPU and that they are treating RT as "as a general computational technique" (similar wording to Microsoft's DXR announcement btw). I don't see anything hinting at some "RT Core" like implementation...

All I see is a bunch of PR drivel. And comparing their product due sometime in the next few years with the programming model of GPUs about a decade ago. I'm not impressed.