Intel Gen 11 Graphics Architecture

Dayman1225

Newcomer
Features:
  • >1 Teraflop Integrated graphics
  • Tile Based Rendering
  • 'Improved' Memory Subsystem
  • Support for Coarse Pixel Shading
  • Redesigned FPU
  • 2xFP16
  • 'Improved' Efficiency
  • Parallel Decoders
  • Ground up HEVC encoder design
  • HDR Tone Mapping
  • HDR Support
  • Adaptive Sync Support
  • 2 Pixels/Clock Pipeline
AnandTech article
 
With 1/12th (8%) the performance of the Radeon RX Vega 64 and 1/6th (17%) of the Radeon Pro Vega 20.
The number against Radeon Pro Vega 20 is wrong (should be 1/3).
But yes it obviously doesn't compete with that - however on paper it pretty much corresponds to a Vega 8 (with same tflops / clock) (whereas the current GT2 is close to Vega 3 there). Now of course if in practice it will perform similar remains to be seen (the current GT2 isn't too far off from Vega 3, though Gen11 will need improvements in bandwidth efficiency if it wants to keep up against Vega 8 - and of course AMD should have something newer out by then).
 
should be 1/3
Acknowledged.

on paper it pretty much corresponds to a Vega 8
No doubt Teraflop-scale processing power is a milestone for Intel, but that's been available on low-end AMD Radeon Rx x50/x60 discrete graphics from 5+ years ago - which also have 100 GByte/s dedicated GDDR5 memory.
 
No doubt Teraflop-scale processing power is a milestone for Intel,
Intel has broke the teraflop barrier before with their GT4e configuration on Gen9, which had the benefit of eDRAM, never in their much more common GT2 configuration with no eDRAM however which is why its significant for Intel.
 
Interesting tidbit from the Tech Report:

TechReport said:
Blythe claimed that implementing a Gen9 EU and a Gen11 EU on the same process would put the Gen11 EU at 75% of the area of its predecessor, partially explaining how it was able to pack so many more of those units into the undisclosed area allocated for GT2 configs of Gen11 on Ice Lake.

So a Gen 11 EU would be 75% of the size of a Gen 9 EU on the same process node.
 
So a Gen 11 EU would be 75% of the size of a Gen 9 EU on the same process node.
Intel's IGP never really were area-efficient compared to the competition as far as I can tell, so it's not surprising to see improvement there. And I bet they improved power-efficiency, which was subpar too, along with it (that is, power efficiency with 3d load - power management seemed great, as well as efficiency of video decode and display blocks).
I am curious though if these improvements were really first realized with Gen 11, or not already Gen 10 (could of course be partly Gen 10, partly Gen 11). But of course nowadays intel won't even acknowledge Gen 10 exists, even though yesterdays tune was that the chips which actually have Gen 10 graphics are in great shape and shipping in volume since a year :).
Although it's quite possible Gen 11 is the bigger improvement than Gen 10 would have been (after all the GT2 GPU on Cannonlake increased EUs only to 40).
 
Any and all improvements are welcome, there is huge amounts of users working with integrated GPUs.
Really hope that in future they will bring lessons learned from discrete graphics to integrated.
 
Intel Buys Indian Startup To Gain Discrete GPU Tech Expertise
Intel has acquired part of the India-based start-up Ineda Systems. The Hyderabad-based company develops technology for autonomous driving, but the acquisition would actually have an emphasis on the GPU-related skills of about a hundred technicians.
...
Intel acquired engineering resources from Ineda Systems, a silicon and platform services provider based in Hyderabad. This transaction provides Intel with an experienced SOC (system on chip) team to help build a world-class discrete GPU business,” an Intel spokesperson told TOI. Ineda Systems, which was set up in 2010-11 by Hyderabad based serial entrepreneur Dasaradha Gude, operates in the sphere of autonomous driving, artificial intelligence and IoT.
https://www.guru3d.com/news-story/intel-buys-indian-startup-to-gain-discrete-gpu-tech-expertise.html
 
So is this the first tangible improvement since Skylake? What happened to Gen 10? It's about time that we see a ubiquitous integrated chip completely surpass the ancient G80 on paper.
 
Being the first new core since Skylake, yes.
The wording here is a bit imprecise, if you mention core and skylake that sounds like cpu core.
The gpu generation is not necessarily tied to the cpu generation - it just so happens that indeed neither the cpu nor gpu parts really differ since skylake.
Cannonlake yields were funny.
The really funny thing is, the driver is ready for ages (I'm talking about the open source mesa driver, but I bet it's the same for the windows one), and it is apparently running in the lab just fine. So gen 10 is very real, but will not appear in the market (the atoms are skipping gen 10 too, but that might be coincidence). And because of that, intel never mentions gen 10 when comparing gen 11 to previous solutions.
 
Back
Top