Intel can offer Xe-HP server GPU products, shifts focus to Xe-HPC and Xe-HPG

In a tweet posted yesterday afternoon by Raja Koduri, Intel’s SVP and CEO of Accelerated Computing Systems and Graphics (AXG) Group, the GPU frontman revealed that Intel has given up on its plans to bring out its Xe-HP series of server GPUs on the commercial market. Stating that Xe-HP has evolved into the Xe-HPC (Ponte Vecchio) and Xe-HPG (Intel Arc) families within Intel’s GPU group, the company apparently no longer sees the need to release a second set of server GPUs – at least, it’s not something based on Xe-HP as we know it.

Also codenamed Arctic Sound, Intel’s initial family of server GPUs was the most visible product in development from Intel’s reborn GPU group. Koduri often showed chips containing silicon while on display at Intel’s labs. And, apart from the Xe-LP / DG1, this was the first high-performance Xe silicone to be developed by Intel. Notably, it was also the only high-performance Xe silicon to be produced by Intel; Xe-HPC computer boards and Xe-HPG matrices are manufactured by TSMC.

We haven’t heard much about Xe-HP this year, and in retrospect it was a sign that something was wrong. However, since the year Intel has shown the Xe-HP demo with performance of as much as 42 TFLOPS FP32 streams. In November, the company announced that Xe-HP was sampling selected customers.

But it seems Xe-HP just isn’t doomed. For 2021, Intel focused on assembling the Ponte Vecchia for the Aurora supercomputer (and possibly for other customers), as well as introducing the Xe-HPG Alchemist GPU family for Q1 2022. According to Kodura, the Xe-HP was used as a development vehicle for the Aurora and Intel’s oneAPI – so it didn’t go unused – but that’s what Xe-HP did.

For now, the Xe-HP cancellation raises some big questions about Intel’s server GPU plans. Xe-HP was to be the backbone of their server efforts, using a scalable design that could range from one to four cores to meet data center needs ranging from computing to media processing. Between Xe-HP and Ponte Vecchia covering a very high market overall (e.g. HPC), Intel needed to develop a powerful group of parallel processors to compete with market-leading NVIDIA and offer traditional Intel customers a GPU option that allows them to remain in Intel’s ecosystem.

It is unclear at this time what will fill the gap left by Xe-HP in Intel’s product group. The Ponte Vecchio is now in production, and according to Intel’s revised Aurora data, it is doing better than expected. But a massive chip is expensive to build – at least in its current configuration. And while Xe-HPG could be invited to use the server next year, unless Intel is able to set it up as Xe-HP, it won’t be able to offer the performance Xe-HP was supposed to deliver.

Equally unclear is a full understanding of why Intel decided to cancel Xe-HP. Since the silicone is already up and running, its cancellation certainly restores their server GPU plans. But as AMD has already begun introducing its new server GPU products based on the CDNA2 architecture, and NVIDIA is likely to target some sort of in-house refresh in 2022, the question is whether Xe-HP is simply too late and / or too slow to compete. in the server market. Along with that, it’s the only line of high-performance Xe parts that Intel has made itself, using the 10nm Enhanced Superfin process (now called Intel 7).

In any case, Intel is clearly not giving up on its plans to break into the server GPU market, even if parts of that plan now need to be rewritten. We contacted Intel for further details and will continue to update this story if Intel publishes a more detailed statement about its server GPU plans.

Source link

Naveen Kumar

Friendly communicator. Music maven. Explorer. Pop culture trailblazer. Social media practitioner.

Related Articles

Leave a Reply

Your email address will not be published.

Back to top button