I think it’s more that they’re unwilling. AMD goes after low hanging fruit and targets the mass market. In essence, they’re willing to let NVIDIA invest in all of the new tech, and then they implement whatever gets popular.
So unless they decide to truly prioritize their GPU business, they’ll be happy to target the quiet majority who care mostly about price to performance while focusing on innovating on the CPU side of the business where they make their real money.
I’m sure they could compete on the GPU side if they threw money at the problem, but they don’t see a need to when it’s decently popular and they’re seeing a lot more growth and profit on the CPU side.
And that’s how it has been for a long time at AMD.
Look at CPUs, they were in a comfortable second place as the economy option for many years, and when they tried something new, it blew up in their face (Bulldozer).
Ryzen was all about the chiplet design first, and architecture improvements second. They didn’t go for the most innovative core design or smallest process (they didn’t even have a fab), they went for the economical option (chiplets have better yields). They were able to catch up with Intel with IPC gains, but Ryzen was pretty uninteresting aside from that. Even today, Zen 4 is just an iteration on the chiplet design, and they’re beating Intel because Intel struggled with lithography issues, and Intel is also trying novel things that haven’t resulted in a clear win vs AMD. So AMD is happy to attack yields (chiplets) and innovate by extension (add-on cache) instead of trying something radical with core design.
Their GPUs are going the same way. NVIDIA is trying hard with RT cores, whereas AMD mostly reused regular shader cores initially. NVIDIA is building a huge model for DLSS, AMD just applies a simple, one-size fits most filter on top. NVIDIA goes for the best experience for the high end, AMD just goes for a pretty good experience for most.
I don’t see that changing, that has been AMD’s main playbook since Intel overtook them after the x64 transition.
AMD seems incapable of competing with Nvidia at the high end, they can’t make FSR as good as DLSS and they are still far behind in RT
I think it’s more that they’re unwilling. AMD goes after low hanging fruit and targets the mass market. In essence, they’re willing to let NVIDIA invest in all of the new tech, and then they implement whatever gets popular.
So unless they decide to truly prioritize their GPU business, they’ll be happy to target the quiet majority who care mostly about price to performance while focusing on innovating on the CPU side of the business where they make their real money.
I’m sure they could compete on the GPU side if they threw money at the problem, but they don’t see a need to when it’s decently popular and they’re seeing a lot more growth and profit on the CPU side.
If you look at the die sizes, it becomes clear AMD are not targeting the super top end.
https://www.eatyourbytes.com/gpus-by-die-size/
And that’s how it has been for a long time at AMD.
Look at CPUs, they were in a comfortable second place as the economy option for many years, and when they tried something new, it blew up in their face (Bulldozer).
Ryzen was all about the chiplet design first, and architecture improvements second. They didn’t go for the most innovative core design or smallest process (they didn’t even have a fab), they went for the economical option (chiplets have better yields). They were able to catch up with Intel with IPC gains, but Ryzen was pretty uninteresting aside from that. Even today, Zen 4 is just an iteration on the chiplet design, and they’re beating Intel because Intel struggled with lithography issues, and Intel is also trying novel things that haven’t resulted in a clear win vs AMD. So AMD is happy to attack yields (chiplets) and innovate by extension (add-on cache) instead of trying something radical with core design.
Their GPUs are going the same way. NVIDIA is trying hard with RT cores, whereas AMD mostly reused regular shader cores initially. NVIDIA is building a huge model for DLSS, AMD just applies a simple, one-size fits most filter on top. NVIDIA goes for the best experience for the high end, AMD just goes for a pretty good experience for most.
I don’t see that changing, that has been AMD’s main playbook since Intel overtook them after the x64 transition.
If what MLID saying is true then yeah that’s possible
And AI/ML workloads. Nvidia gets lots of shit and is more expensive but you get a better ecosystem with their cards.
A good portion of this though is the CUDA stranglehold nvidia has. Good luck getting a neural net accelerated on OpenCL or Vulkan Compute.
AMD do seem to be taking steps in the right direction here, still a while away from a more balanced landscape.