It's been four years since AMD launched RDNA, the successor to the venerable GCN graphics architecture. We take a look through the tech and numbers to see...
Starting to really dislike this 100% or 0% mentality, even if it’s just being used as the vehicle to carry the article. It’s what’s so wrong with the world today. It’s a gross oversimplification of reality.
That aside, I agree with the premise. With RDNA3 AMD established that it wasn’t willing to compete on price, 4000-series pricing left the door wide open for AMD to undercut and wholly outsell NVIDIA hardware by offering better price/performance value. But AMD choose to not take advantage of it, and oddly instead copied NVIDIA’s upselling strategy with its own pricing choices.
Strangely AMD made that decision despite realizing cost-savings from migrating to a chiplet design. If AMD is no longer willing (or able) to compete on best value, then as the article implies AMD needs to start competing for the performance crown again. AMD needs to at least pick a strategy and do something, because it simply doesn’t have any spare graphics marketshare left to waste.
I’ve heard that, and it’s plausible. But keep in mind the MCD’s are made on 6nm which AMD isn’t using for CPUs. Also AMD is fabricating some APUs and EPYCs on TSMC 4nm already.
If AMD is suffering from limited wafer supply at 5nm then it absolutely has to adopt a lower volume, higher performance crown targeting approach to its GPU designs. Regardless of whatever the supply is today, wafer supply in the future is not going to be improving as everyone and everything funnels into the same future TSMC node groups.
Then they should have made a fatter GCD 250mm^2 vs 150mm^2 with 140 or something CU’s, and added 3d stacked cache to feed it on the MCDs. They might have even had a plan to but flaked out. Beat 4090 on raster, and compete with 4080 on RT. I was hoping RDNA4 would do just that, but the rumors of AMD only doing mainstream for RDNA4 seems the opposite.
Starting to really dislike this 100% or 0% mentality, even if it’s just being used as the vehicle to carry the article. It’s what’s so wrong with the world today. It’s a gross oversimplification of reality.
That aside, I agree with the premise. With RDNA3 AMD established that it wasn’t willing to compete on price, 4000-series pricing left the door wide open for AMD to undercut and wholly outsell NVIDIA hardware by offering better price/performance value. But AMD choose to not take advantage of it, and oddly instead copied NVIDIA’s upselling strategy with its own pricing choices.
Strangely AMD made that decision despite realizing cost-savings from migrating to a chiplet design. If AMD is no longer willing (or able) to compete on best value, then as the article implies AMD needs to start competing for the performance crown again. AMD needs to at least pick a strategy and do something, because it simply doesn’t have any spare graphics marketshare left to waste.
From what I heard they don’t want high volume low margin gpus to take up fab capacity when they could instead use that for cpus
I’ve heard that, and it’s plausible. But keep in mind the MCD’s are made on 6nm which AMD isn’t using for CPUs. Also AMD is fabricating some APUs and EPYCs on TSMC 4nm already.
If AMD is suffering from limited wafer supply at 5nm then it absolutely has to adopt a lower volume, higher performance crown targeting approach to its GPU designs. Regardless of whatever the supply is today, wafer supply in the future is not going to be improving as everyone and everything funnels into the same future TSMC node groups.
Then they should have made a fatter GCD 250mm^2 vs 150mm^2 with 140 or something CU’s, and added 3d stacked cache to feed it on the MCDs. They might have even had a plan to but flaked out. Beat 4090 on raster, and compete with 4080 on RT. I was hoping RDNA4 would do just that, but the rumors of AMD only doing mainstream for RDNA4 seems the opposite.