The Intel EU is gone. The new block of its "gaming first" graphics card is the Xe-core

The intel eu is gone. The new block of its "gaming first" graphics card is the xe-core

In an attempt to change the branding of every part of its Xe graphics card architecture, Intel has killed the humble Execution Unit. The EU is the cornerstone of Intel's graphics architecture, but after years of use, Intel is making the term available to make way for a new drive, the Xe-core.

The name change occurs because Intel says there is too much within the same EU to classify it as the current EU.

However, this does not completely change what an EU / Xe core really does. Currently, an Intel UE contains several ALUs for floating-point and integer operations, and the same goes for an Xe kernel. However, the new Xe-core nomenclature plays into the significant incoming shock with Intel's first discrete generation of graphics cards, codenamed Intel Alchemist.

The new Xe-core of Alchemist GPUs includes 16 vector engines and 16 array engines. In fact, it's twice as much as the Ponte Vecchio GPU, which is built with the Xe-HPC architecture and is intended for the Aurora supercomputer.

Intel says this is a necessary step in expanding gaming graphics cards, and that this chip has been specially optimized to "play first".

If you lower the Xe-core tone, you'll find a Render Slice, which contains four of these Xe-core clusters, a fixed-function unit for DirectX 12 Ultimate support, and four new ray tracing acceleration units. A set of eight Render Slices will then share access to the L2 cache.

These new Intel ray tracing units are designed to accelerate ray traversal, box boundary intersection, and triangle intersection. It sounds a lot like the RT cores of Nvidia’s Ampere GPUs, and the AMD Ray Accelerators has within its RDNA 2 architecture. Although we don’t know how they deal with each other in terms of gross performance.

Image 1 of 4

Details of the intel xe-hpg gpu

(Image credit: Intel)
Image 2 of 4

Details of the intel xe-hpg gpu

(Image credit: Intel)
Image 3 of 4

Details of the intel xe-hpg gpu

(Image credit: Intel)
Image 4 of 4

Details of the intel xe-hpg gpu

(Image credit: Intel)

Again, this overall composition is different from Ponte Vecchio in the Xe-HPC architecture, which has fewer engines per Xe core, but a larger number of Xe cores per Render Slice. There are 16 Xe cores in each HPC segment with as many ray tracing units. A set of four rendering cuts share access to the L2 cache, HBM2e drivers, and a media engine. So on and so forth.

There are many levels of Ponte Vecchio that I dare not cover here.

Alchemist will be built on TSMC’s 6N process, however, while Ponte Vecchio will use a mix of TSMC’s N5, N7, and Intel 7 process nodes, depending on the tab you see in the Ponte Vecchio package. Basically, these are very different tokens.

However, what we are seeing is proof of the changes promised by Intel to the Intel Xe architecture, depending on how it is intended to be used.

The architecture of the games is substantially different from that of the data center, and both move away from the lower-spec Xe-LP chips of today’s laptops. This type of segmentation is not a completely new concept, both Nvidia and AMD data center cards look very different from gaming cards, but it's still exciting to see how another company's engineers approach design and using the GPU in a new way.

Intel also confirms that, beyond the configuration of an Xe-core or Render Slice, work has been done on optimizations and methodology to increase the energy efficiency of its upcoming graphics cards, which include adjustments to the architecture itself, memory and physical design of the card. .

(Image credit: Intel)

Taking into account all the optimizations performed by Intel and the disparities of the process nodes, the Xe-HPG architecture offers approximately 1.5 times the performance / watt and works approximately 1.5 times the frequency of Intel's Xe-LP chips .

While we believe we had good control over what was expected of DG2 even before Intel's Architecture Day, in terms of kernel counting and the specifics of the Xe-HPG architecture, maybe we didn’t have to take it as we suspected. After all, looking at what we know now, we compared apples to oranges among Intel Xe microarchitectures.

Intel’s Alchemist graphics cards will arrive early next year, and before then, we’re likely to hear more about the specifications of real cards and their performance compared to the competition. That's when we'll find out if this whole Xe architecture game really translates into frame rates.

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *