Intel has just lately made the whitepaper for its subsequent era of processor graphics freely accessible. The doc, masking the architectural particulars of the GPU portion of its upcoming 10nm Ice Lake CPUs – sure, they’re coming this time, trustworthy – has arrived alongside the primary Intel Odyssey occasion at GDC this yr.
The two are form of associated as Intel has said that the Gen11 GPU design incorporates the fundamental essential building blocks of the upcoming Intel Xe discrete graphics card set to launch subsequent yr. That’s to not say it will likely be operating the very same slice/subslice design, however Intel isn’t throwing all its newest GPU efforts out with the bathwater when it shifts focus onto Xe. With as much as 1TFLOP of compute energy, the Gen11 silicon is a real enchancment on earlier iterations of Intel built-in graphics.
“Gen 11 is a great step forward for us,” stated Intel’s Gregory Bryant at an investor discussion board final yr. “You’ll see us do it again with Gen 12 graphics for 2020 and Gen 12 graphics IP is the basis for that discrete graphics portfolio that Raja is architecting and building.”
You can obtain the Gen11 whitepaper your self (PDF) however the important thing spotlight is that it’s an enormous enchancment over the Gen9 GPU utilized in all of the last-gen 14nm processors we’ve been utilizing since Skylake.
Read extra: These are the best GPUs round at the moment
But what about Gen10? Well, Intel struggles with that quantity – it was the designation given to the graphics element of the Cannon Lake CPUs, besides not one of the truly launched Cannon Lake chips got here with built-in graphics enabled. So we’re skipping over Gen10 and going straight to Gen11.
| Intel Gen11 | Intel Gen9 | |
| Slices | 1 | 1 |
| Subslices | 8 | 3 |
| Cores (EUs) | 64 | 24 |
| FP32 FLOPS | 1,024 | 384 |
| FP16 FLOPS | 2,048 | 768 |
| Local cache | 512KB | 192KB |
| L3 cache | 3,072KB | 768KB |
The full spec Gen11 GPU element will include 64 execution items – these wee cores able to doing all that integer and floating level maths. The strongest Gen9 silicon managed simply 24. That provides the brand new era of processor graphics as much as 1,024 floating level operations per second (FLOPS) of FP32 compute energy the place the last-gen may muster simply 384 FLOPS.
Intel can be engendering the Gen11 silicon with its Coarse Pixel Shading (CPS) characteristic. This is roughly analogous to the Variable Rate Shading (VRS) Nvidia has been speaking about as regards to its Turing GPU cores, however extra importantly that is what’s permitting Intel to be within the dialog when Microsoft announced its DirectX 12 VRS help this week.
The Gen11 spec additionally consists of Tile Based Rendering too. This permits the GPU to interrupt down a scene into particular person tiles relatively than having to render the complete scene as a complete. This breaks down the quantity of knowledge that must be saved within the reminiscence of the GPU, lowering reminiscence bandwidth calls for.
Intel can be constructing adaptive sync help into the Gen11 GPU too, giving screens that help the VESA normal, and connect with their PC through DisplayPort, the power to immediately synchronise with the graphics silicon. This will do away with display screen tearing and the micro-judder that’s symptomatic of V-Sync, very similar to AMD’s FreeSync and Nvidia’s G-Sync do.
The primary Gen11 design is inbuilt a approach that shall be acquainted to anybody who has frolicked trying over Nvidia’s GPU layouts. I’m such an attention-grabbing fella that unsurprisingly describes me… Intel divides its graphics processors into slices and subslices, that are related in structure to Nvidia’s basic processing cluster (GPC) and streaming multiprocessor (SM) design.
Each Intel slice incorporates as much as eight totally different subslices, every with eight executions items inside it. This provides up the complete 64EU complete. On a Gen9 chip, nevertheless, Intel solely provided Three subslices for a most of 24EU. Intel will supply totally different ranges of Gen11 graphics, in order that they gained’t all have the identical eight subslices, some may solely comprise six, for instance, for a complete of 48EU.
The subslices every have their very own set of components; their very own instruction cache and thread dispatch items, in addition to media and texture samplers. They need to share a certain quantity of assets, nevertheless, together with the L3 cache and rasterizer items. This all permits the subslices to have a certain quantity of autonomy, which gives the parallelism required of a GPU, whereas not losing logic house on extraneous components.
While historically Intel’s processor graphics haven’t actually set the gaming world alight, the efficiency uptick of the Gen11 GPUs ought to supply a good stage of low-end gaming for PCs and laptops with out the necessity for devoted graphics playing cards. That ought to imply skinny and light-weight 10nm machines may ship fairly first rate 720p, and perhaps even 1080p, gaming body charges.
And that features the upcoming Intel Lakefield processor, utilizing the progressive Foveros packaging know-how to create 12mm2 SoCs with unprecedented gaming energy. Game on, certainly.
Source
