Last week at Computex, Intel released their Pro B-Series GPUs, the B50 and B60. On the surface it doesn’t seem like much but SemiAccurate thinks there is a lot of good here.
<rant>Lets start out with the elephant in the room, the Intel GPU roadmap. With the Battlemage/B GPUs, Intel finally has a hit on their hands. Sure it isn’t perfect but it is solid, has a great reputation, and demand is outstripping supply. This bodes well for the C-Series cards, right? Well talking with partners at the show, it seems like the C-Series has yet to get the green light for production. This may be due to management changes, market outlooks, or something SemiAccurate is not privy to. Our advice is to get the ball moving, Intel’s systemic inability to build on success of late needs to be put to rest once and for all. This is a golden opportunity to do so.</rant>
Baby, Mama, and Papa Battlemages
Back to the release of the pro-B GPUs which are basically the current Battlemage GPUs with a new driver and more memory, plus a few twists. The twists are a very large memory complement and a dual GPU card, both really useful for the intended purpose of the GPUs, basically AI. Yes you can say the pro drivers are useful for the traditional CAD and simulation software market, but lets face it, if you don’t lead with AI, no one cares any more. Yes this is sad but it is the reality of the market.
Two new Pro series Battlemage GPUs
` If you have been following the Intel GPU story to date you know that the current cards are firmly in the middle segment of the market, blame the long delays to market on that. This isn’t a slight, they are priced right, perform really well, have stable drivers, and are affordable. With AMD and Nvidia essentially cutting volume for enthusiast parts to near zero, Intel has a golden opportunity here (Authors note: *HINT*). Now in the pro/AI segment, they are repeating the same playbook.
Booth staff show off cards to Computex attendees
So if you have a reasonably priced card that outperforms it’s price by a noticeable amount, why should you buy it over an established player? Drivers and outsized memory specs. The B50 has 16GB and the B60 comes with 24GB, and since memory space is what determines a lot of AI performance, Battlemage has a lot going for it. Sure you can get an Nvidia card with larger memory space now but it also comes with an added bonus in the form of extra zeroes on the price and allocation time. Pick your poison.
Intel’s answer to having a mid-range device in a market where the high end dominates the headlines and price seems to be ignored is to show scaling. This is done in two ways, first the dual B60 card from Maxsun is touted as a 48GB AI device which it is. Sort of. 2x24GB not the same as 48GB of contiguous memory but given how tightly coupled a dual card is, it is about as close as you can get. And the price should still be way below an Nvidia or AMD card of similar performance. But dual cards are just a building block for a system which brings us to…
Put it all together and get a Battlematrix
Project Battlematrix is another in a line of halo products with a cool name slapped on. Take an Intel Xeon with PCIe5 slots and slap in four dual B60 cards. This is not a big trick since most Xeons of late support way more than the needed PCIe5 lane count and the B-Series cards only need x8 slots to perform. Intel has an optimized software stack for this configuration including drivers and a tuned Linux stack. So far so good. The pitch is that you can respond to more requests with lower response times than a single device of comparable performance not to mention price. So far so good and they have a bunch of graphs to prove it.
Intel AI assisted video debugging
One of the cooler use cases that we have seen of late is the video debug process. If you are developing hardware or software and there is an error, finding the error is hard enough but tying it back to a specific bit of code is painful and time consuming. The demo Intel had was to have AI ‘watch’ the video, see an error, hunt down the code that screwed up, and fix it. This is all done autonomously, at least in theory, and should be orders of magnitude faster than having meat sacks doing the same job. Even if the AI can only do some of the functions, it should be a massive benefit to productivity.
You can see a bunch of other use cases for the multi-GPU model, running different AIs with some subordinate to others, each with a specific use case for example. You can do this on a single big card but memory constraints and task switching/parameter reloading eventually bite. This is the long way of saying Intel is pointing out the use cases their cards work for but they are actually real use cases that have actual applicability in the market.
Overall the B50 and B60 look like solid players in the mid-market at a good price. The Battlemage consumer cards are all on allocation and demand far outstrips supply. This is a really good place for Intel graphics to be, they finally ‘made it’ with a real device that consumers like. The Pro cards should have the same reception, value, performance, and scaling. The real problem is that unless Intel green lights the next gen soon, and does so publicly and loudly, why would anyone invest the time and effort to bring their software to a dead end? If you don’t think this is a real problem, remember the Gaudi cards? Intel has a golden opportunity here, lets hope they don’t blow it.S|A
Charlie Demerjian
Latest posts by Charlie Demerjian (see all)
- AMD’s MI350 Is A Big Iterative Advance - Jun 12, 2025
- AMD Lays Out their MI350, MI400, and MI500 Roadmap - Jun 12, 2025
- Qualcomm Snapdragon Elite X2 CPU Spotted - Jun 2, 2025
- Acer Shows Toys And Laptops During Computex - Jun 2, 2025
- Qualcomm Teases Infrastructure Plans And More - May 28, 2025
