bugTraditionally, AMD was 1/10th the size of Intel. AMD demolished Netburst, introduced AMD64 (and effectively killed off Itanium), built multi-cores that were not glued together, was the first to integrate the memory controller into the CPU... I don't know, that's not exactly the image of a mere follower.

They had a good stride in the early 2000s, yes and by the end of the Netburst's lifespan Athlons were overall better chips. Even so, they didn't really demolish them (in some tasks P4s were still better), definitely nowhere near the demolition AMD received during the Bulldozer era (end honestly Zen1 as well). But here's the thing - the reason they were able to do so back then, was the well known and documented plan, that Netburst would eventually reach well over 5Ghz (which at its inception seemed solid, given the rapid silicon technology development at the time), but the process nodes never delivered. And what do you know, AMD's next breakthrough once again came during Intel's struggles with 10nm and based on development and revenues from an entirely different industry (TSMC/mobile). If all they had was their (once own) Global foundries, they would still be hobbling along with Zen+ (or maybe by now +++) just like they did previously with barely perceptible improvements of Bulldozer for 6+ years.

HenrySomeoneThey had a good stride in the early 2000s, yes and by the end of the Netburst's lifespan Athlons were overall better chips. Even so, they didn't really demolish them (in some tasks P4s were still better), definitely nowhere near the demolition AMD received during the Bulldozer era (end honestly Zen1 as well). But here's the thing - the reason they were able to do so back then, was the well known and documented plan, that Netburst would eventually reach well over 5Ghz (which at its inception seemed solid, given the rapid silicon technology development at the time), but the process nodes never delivered. And what do you know, AMD's next breakthrough once again came during Intel's struggles with 10nm and based on development and revenues from an entirely different industry (TSMC/mobile). If all they had was their (once own) Global foundries, they would still be hobbling along with Zen+ (or maybe by now +++) just like they did previously with barely perceptible improvements of Bulldozer for 6+ years.

So is Ryzen development entirely down to AMD or are TSMC responsible like they are for the 3D cache?

HenrySomeoneThey had a good stride in the early 2000s, yes and by the end of the Netburst's lifespan Athlons were overall better chips. Even so, they didn't really demolish them (in some tasks P4s were still better), definitely nowhere near the demolition AMD received during the Bulldozer era (end honestly Zen1 as well). But here's the thing - the reason they were able to do so back then, was the well known and documented plan, that Netburst would eventually reach well over 5Ghz (which at its inception seemed solid, given the rapid silicon technology development at the time), but the process nodes never delivered. And what do you know, AMD's next breakthrough once again came during Intel's struggles with 10nm and based on development and revenues from an entirely different industry (TSMC/mobile). If all they had was their (once own) Global foundries, they would still be hobbling along with Zen+ (or maybe by now +++) just like they did previously with barely perceptible improvements of Bulldozer for 6+ years.

Yeah, saying Neburst was better in some tasks is like saying a dragster is a good car for the road, because in some "tasks" it fares better. Netburst's wins were very few and usually came because of Intel's compiler.
Regardless, AMD's win wasn't on performance (performance was in the same ballpark), it was on cost and perf/W.

The thing right now and in reality is, we actually do not have more than 8 P cores on anything that is on a single CCD or ring bus. You have to go back to COmet Lake and Broadwell E where I think hey had a maximum of 10 on same ring/CCD.

AMD even with Zen 3 and now Zen 4 still have 8 cores maximum per CCD. And the latency to cross CCDs is far higher (3-4X than inter core communication within a CCD or in Intel's case the ring. That could cause problems or slight pause in games if they take more than 6 cores on 5900X or 7900X which just have 2 6 core chiplets. In that case games are better off with a 5800X as they have 8 cores in one CCD or Intel 12th gen 8 p cores with e-cores off.

www.anandtech.com/show/16214/amd-zen-3-ryzen-deep-dive-review-5950x-5900x-5800x-and-5700x-tested/5

Zen 5 is supposed to have 16 core CCDs so that will be exciting as it will be first true more than 8 core CPU with modern IPC.

For now though games actually are better off on one CCD/ring chiplets as I found out latency to cross CCDs or a ring is much much higher than I thought. In theory that i bad for something like game threads that need to always communicate with each other as fast as possible.

So Zen 5 looks exciting.

Best thing now as I just found out would be for a 10 core Intel ring bus Golden Cove or Raptor Cove CPU, or an 8 core with lots more cache and no e-cores. Or buy a 12700K and shut off e-cores and you get the best gaming CPU tied with 5800X3D, but far better in other things as well with e-cores off of course.

E-cores are just not good right now as hybrid arch causes more trouble than its worth and it drags the ring clock down a lot which hampers performance.

Though games fortunately do not really even need more than 6 cores 12 threads with excellent IPC, but 8 cores 16 threads is definitely more comfortable for headroom. Would like a couple more cores for overprovision, but unnecessary at least now and unfortunately such product with more than 8 cores in a single CCD/ring bus does not exist yet and will not till Zen 5 or maybe Meteor Lake as Intel seems to be changing the design on it where it appeared it was going to be still