What is going on with Intel’s GPU program? The chatter is negative but what is really happening? SemiAccurate dug in to the program and found out why things went bump in the night.
For long time readers of SemiAccurate, please sit down and stop drinking anything before you keep reading. We say this for your safety because we are going to defend Intel for the second time in a row. The current Intel GPU program is in pretty awful shape but we are going to point out that the current malaise isn’t their fault. Really. OK, mostly not but this latest delay is truly something they could not have foreseen.
If you recall, DG2/Arc Alchemist, was supposed to debut late last year as a holiday sales item. this would have been roughly 18 months late, something we won’t defend. What was meant to be a device that nipped at the heels of the high end market was now a solid mid-range device. Silicon ages far worse than fish but it was finally coming out.
That holiday release was delayed 4-6 weeks because the factory making the boards was hit by Covid and things obviously slowed down or stopped. SemiAccurate has confirmed this issue. If you are going to launch for holiday sales and you get delayed, it is probably a better idea to time it with the next obvious sales uplift than launch it between, oh say Christmas and New Years Day. So that pushed DG2/AA into mid/late Q1. Fair enough.
During the Q2/22 analyst call, Intel pointed out that the standalone cards were delayed again and the program wasn’t exactly doing well. While the card is out now, the reports of drivers being, lets be kind and say sub-optimal, abounded. The power/performance ratio was way off too, but there aren’t many saying the price is way off unless you are looking at Intel’s margins to determine what to buy the kiddies.
On the power front, that is more understandable given the now ~2 year delay and the fact that it was meant to be a high end card. While Intel hasn’t sampled the GPUs, it looks like they were downclocked a bit to save energy at the cost of performance. If you wanted to be charitable you could point out that these devices probably have a lot of headroom for enthusiasts. That said between the driver mess and the delays, the performance is not something to set new records.
So why the delay and the mess? Intel is usually pretty good at drivers but this time around things are quite uncharacteristic. Intel offered a few reasons for this on their Q2/22 analyst call which boiled down to, ‘this is harder than we thought’ but that isn’t actually the reason. If that was it, the SemiAccurate blamethrower would have been used and refueled several times already so what really caused this mess?
The short version is to look where the drivers are being developed. In this case Intel is literally developing the DG2 drivers all over the world as they do for many things, hardware and software. The problem this time is that key parts of the drivers for this GPU, specifically the shader compiler and related key performance pieces, were being done by the team in Russia. On February 24, Russia invaded Ukraine and the west put some rather stiff sanctions on the aggressor and essentially cut off the ability to do business in the country.
Even if businesses decided to stick with Russia, it would have been nearly impossible to pay the wages of their workers due to sanctions on financial institutions and related uses of foreign currencies. In short Intel had a key development team cut off almost overnight with no warning. This is why SemiAccurate say it isn’t their fault, even if they saw the war coming, they probably didn’t see the sanctions coming.
Note: One tech CEO SemiAccurate talked to recently said they should have seen it coming but we didn’t have time to dig in to why they had that viewpoint.
Note 2: SemiAccurate has known about this problem for several weeks and did not write this story, especially the following portion, until we knew certain things were completed and no one would be put in harm’s way by this disclosure. We are satisfied that our story will not cause harm to innocent people at this point, something we could not have said with certainty a month ago. That said we will still omit some details, sorry.
Intel did undertake a large and rather extraordinary effort to get their staff, and some say their families, out of Russia. Multiple sources say this was a bit of a mess, especially early on when the effects of the war were not clear to anyone and the pundits were all proclaiming it would be over soon. Many engineers were relocated, some chose not to go, and some changed their minds during the process. Things like this are never easy and some of the tales SemiAccurate heard only enhance that impression.
The next problem was where to move these engineers to? Several sites were said to be considered but at least one country balked at having a large contingent of Russian engineers moved onto their soil while the country was waging war with their neighbors. Once again SemiAccurate is told this problem is behind Intel now and all of the engineers that wanted to leave are in a safe location.
Even with the process being ‘done’, you had a massive disruption to key parts of Intel’s GPU driver program. The mess you see is largely the result of this disruption. Don’t forget that some of the engineers stayed behind in Russia and are effectively cut out of the process. What else was lost? How long will it take to get the programs up and running at full speed? We can’t say but the effect of this disruption will undoubtedly be long lived. It was a mess and from what we gather, Intel did right by their people.
So there you have it. DG2/Arc Alchemist was going to be late and silicon doesn’t age well. The first final release was delayed by Covid and then delayed again by a war in Europe. Those last two are things we won’t fault Intel for but the first is not really excusable. What was meant to be the lead TSMC 6nm product launched after TSMC 5nm debuted. As SemiAccurate has been saying for years now, DG3/Arc Battlemage is the real ‘first’ Intel GPU but these delays are unlikely to help it.
All this said, SemiAccurate picked up two tangential data points from this research that we thought you might enjoy. So enjoy.
Note: The following is for professional and student level subscribers.
Disclosures: Charlie Demerjian and Stone Arch Networking Services, Inc. have no consulting relationships, investment relationships, or hold any investment positions with any of the companies mentioned in this report.
Latest posts by Charlie Demerjian (see all)
- Raja Koduri and Dr. Randhir Thakur out at Intel - Mar 21, 2023
- What does Intel’s Emerald Rapids bring to the fight - Mar 21, 2023
- A new ARM code name for a new market pops up - Mar 20, 2023
- A big ARM server project shut down in silence - Mar 14, 2023
- A bit more on AMD’s Genoa memory issues - Mar 13, 2023