The story of Intel’s inability to deliver a 7nm process chip and a supercomputer called Aurora to the Argonne National Laboratory opened new chapters yesterday with statements from Intel CEO Bob Swan that the 7nm “Ponte Vecchio” GPU The company’s an integral part of the Aurora Exascale System due to ship next year will be delayed at least six months.
More than two years ago, Intel announced that plans to build a pre-exascale system called the Aurora (A18) for Argonne had been abandoned – a system due for delivery in 2018 that includes Cray’s supercomputer architecture “Shasta” and is a top performance of 180 petaflops.
Instead, Intel switched as the prime contractor to a new version of Aurora (A21) with an exact computing power (one billion billion operations per second), the Intel 10 nm CPUs and 7 nm GPUs from “Ponte Vecchio” together with the Cray Shasta combines architecture for 2021, a core part of the US exascale strategy.
However, yesterday’s news is not just another delay in a long line of delays in Intel chips (both 10nm and 7nm) in recent years – which translates into an immediate fall in Intel stock prices of more than 15 percent despite an otherwise stellar earnings report – includes Intel’s first admission that 7nm chip production can be outsourced to a third-party semiconductor foundry. This would likely be TSMC (Samsung is another option), which is already making 7nm CPUs and GPUs for AMD, whose processor innovations and overall value for money have allowed Intel to take market share of HPC and data center processors .
Among the Wall Street analysts looking to downgrade Intel stock was Matthew Ramsay of Cowen, whose Intel comment, titled “Yes, It’s That Bad,” includes this statement, as reported by MarketWatch: The Outsourcing Manufacturing Strategy Is Going frustrating Intel customers and investors while ensuring that the company cannot match the current leadership of silicon processes from TSMC-enabled competitors (AMD and Nvidia) in a limited time frame. “
But is Ramsay too gloomy and does Intel’s renewed 7 nm trouble necessarily mean a repeat of the first Aurora (A18) error?
HPC industry analyst Steve Conway, senior advisor on HPC market dynamics at Hyperion Research, sees the latest developments as a relatively short-term hiatus rather than a crippling blow.
“This would move it (Aurora) to late 2021, but possibly move it to early 2022, so it’s not a huge delay – but a delay nonetheless,” Conway said. “I don’t know what that has to do with the terms of the contract, but it looks like they figured out the design pretty well, but there are some flaws that they need to correct before it can go into full production – don’t just this GPU, but all of its seven nanometer parts. “
“It looks a lot like Intel is basically on top of it with design and so on,” added Conway. “You only have to correct one defect. These are very, very complicated parts, and it’s not uncommon to have a defect that is probably not enough. “
Swan said that yesterday.
“We identified a defect mode in our 7-nanometer process that led to a decrease in yield,” said Swan. “We created the problem and believe that there are no fundamental obstacles, but we have also invested in contingency plans to protect ourselves against further uncertainties in the schedule.”
… Yesterday’s news is not just another delay in a long line of delays in Intel chips (both 10nm and 7nm) in recent years – resulting in an immediate fall in the price of Intel shares by more than 15 Percent despite an otherwise stellar earnings report – it includes Intel’s first admission that the production of 7nm chips can be outsourced to a third-party semiconductor foundry. This would likely be TSMC (Samsung is another option) which is already making 7nm CPUs and GPUs for AMD. Thanks to its 7nm chips and its total price / performance, Intel was able to take the market share of HPC and data center processors.
Conway said it would most likely be a temporary measure until Intel can iron out its own manufacturing capabilities.
“Intel will probably make some of the output and then outsource the rest,” he said. “And that’s only for an initial period.”
However, Conway sees this as a strengthening of AMD’s position over Intel in the HPC market as well as efforts by AMD, the Ela Capitan system with two exaflops with HPE / Cray for Lawrence Livermore National Lab and the Frontier system with 1.5 exaflops also to be built in collaboration with HPE / Cray for the Oak Ridge National Lab.
“It’s another silver dollar that falls into AMD’s hat,” said Conway. “At that point, AMD was doing really well with Epyc (7nm CPU) partly because of the memory bandwidth, but it’s also a combination of the pricing for similar SKUs. AMD prices are often lower. So the combination of these gives AMD a boost. “
Both Intel and Argonne National Lab had yet to respond to inquiries at the time of going to press.