AMD CEO Lisa Su is highest paid CEO in S&P 500
fortune.com
fortune.com
Now, if they can just really complete with nVida.
AMD and Nvidia both supply to consoles: Lifetime sales, the Switch has sold 52 million, the PS4 almost 110 million, and the XBone almost 45.
Individual count, AMD has done three times the sales that Nvidia did over roughly the same time period; adjusted for silicon size, AMD made more money per sale; adjusted for CPU+GPU performance, AMD made more money per unit of performance. PS5 and XsX numbers are predicted to be in line with existing numbers, second gen Switches are predicted to be less (most users won't replace their Switch for the higher efficiency 12nm SoC versions).
On the other front, enterprise compute, they're about neck and neck.
Nvidia only has an edge because they took over academia with their CUDA library, and due to the lack of education of other APIs in colleges, they have a pretty nice vendor lock-in situation they're making bank on. However, more and more companies are switching to AMD because they get far more bang for their buck.
Nvidia bought Mellanox because they need a cache coherent DMA bus to allow GPUs to rip data straight off SSDs, or send data between GPUs, without involving CPUs at all. They did this in response to AMD.
AMD bought ATI because they already had a solution for that (Hypertransport, which they bought from the DEC sale the same time they bought the unpublished next gen Alpha CPU that became the K7 and K8 family), but they didn't have a GPU for it. They then sat on it for ten years because the previous CEO didn't understand how to capitalize on it, he waffled to shareholder whims too often and refused to allow innovation.
AMD recently has showed off Infinity Fabric-based external transports (IF is HTX protocol over PCI-E PHY, instead of HTX PHY; massively lowering the cost and allowing stuff like lanes switching jobs depending on how the the machine is configured), such as a laptop whose DGPU communicates over the traditional x16 GPU lanes but using IF instead of PCI-E, so it natively blends with the IGPU in the APU, showing one single coherent device to the OS; they also showed off that new super-computer they built that combines IF-native SSD controllers with IF-speaking Arcturus GPUs (they're using off the shelf PCI switches for this with new firmwares; HTX super-computers in the past were using modified Mellanox IB controllers and switches, expensive but it worked, and Nvidia bought Mellanox for the same idea since AMD proved it worked); they also showed off the storage sub-controller in the PS5 and XsX that can load texture pages directly off SSDs into the GPU, thus freeing a lot of CPU resources and lowering the chances of the dreaded texture pop-in issue.
AMD is unlikely to do to Nvidia what they did to Intel, but Intel got completely fucked because Intel's management is where AMD's previous management was: why innovate when rubes keep buying the same old shit with a slap of new paint on it; who needs more than four cores, anyways?
Nvidia's only strong point right now is the desktop experience for gamers. AMD's drivers, right now, are shit. They didn't use to be shit, they used to be just 'plain' and not do any special sort of optimizing for poorly performing games. Ever since the new branch of Adrenalin came out, they're trying to take on the Nvidia driver juggernaut, and they're just tripping all over the place.
Hence, I recently acquired a 1660 Super as a stopgap. Performs like a 2060, but doesn't have the RTX units. Enough for even Doom Eternal with everything on/high for 1080p60. I can't buy a new GPU because Series 30 and RDNA2 are going to be the new bare minimum (due to consoles having RDNA2, and console games leveraging raytracing), neither are out on the PC yet.
I went from drivers being not well optimized to outright buggy and crashy in the span of 3 months. Don't get me wrong, I find it amazing that every single GCN over the past 8 years is still supported and also gets all the new features (something Nvidia doesn't do), such as anti-lag and NN sharpening, but they should have delayed the RDNA1 launch to get the drivers ready instead of effecting all the GCN users too.
Also, that said, on my GCN, performance was starting to improve in badly performing games, so they've also been clearly trying to fix that front to catch up to Nvidia. You wouldn't expect an 8 year old 7970 to improve, but here we are.
> Nvidia's only strong point right now is the desktop experience for gamers
When using TF with AMD is as easy as nvidia I will belive it, but right now nobody I know is using AMD for machine learning and what I read on HN is that setting it up can be painful.
What a CEO! Respect!
...
Yes, I know that AMD is not one person. But this article mentions her in particular. I really want to congratulate everyone from AMD, a CEO is nothing without a good team!
Last time with Netburst Intel still had some talent (Core and successors started in Israel IIRC), but I keep reading about how all the "old guard" engineers that knew their stuff inside and out are retiring, and Intel isn't hiring or rewarding talent, and their corporate culture has collapsed into infighting and backstabbing.
Intel always had the Microsoft problem of launching things outside of the core x86 CPUs, from graphics cards to XPoint to the flash-on-motherboard cache, completely being flatfooted in ARM and mobile CPUs, but at least they always executed on their core moneymaker.
Now, they've not only lost their lead in process, they've fallen behind. I imagine their patent portfolio will be valuable for decades, but Intel just seems like a zombie company stumbling forward on inertia.
But AMD was dead a decade ago, so what do I know. Maybe Intel will hire an engineering CEO again.