Always good with new stuff but no matter how much AMD claims that performance is "looking good" - we won't see the jump we saw from 2000 to 3000 again.
This was purely because of GloFo 12nm to 7nm TSMC and the XT refreshes shows that clockspeed is close to maxed, with 0-1% performance gain and same all-core speeds. I expect a refinement and not much else.
1000 and 3000 = big leaps.
2000 and 4000 = small refinements.
The best thing about 4000 series will probably be cheaper 3000 series.
Now, if just AMD could be competitive in the GPU market too...
With Cyberpunk 2077 featuring DLSS 2.0 I think AMD GPUs will have a hard time tho. This is probably the most anticipated game in years. Can't wait to buy a 3080 or better later this year and immerse myself in this game, on my OLED TV at 1440p or 2160p at 120 Hz native with Gsync !! This game is going to look absoutely insane in full HDR glory !!
Zen 3 is a rather major improvement. And while the IPC may only be 10-20% faster, the gains in other areas may be much more than that. There are limitation in the design that hinder performance when you have a lot of talk between CCX's. Each Chiplet (CCD) has two groups of 4 cores that share L3 cache (CCX). Unlike Intel where all cores share the higher level cache, it is broken up on Zen. This means when a CCX needs data from another CCX the latency is much higher than if that data was on its own CCX. This is the primary reason why Zen is sometimes slower than Intel on some games, while beating them at nearly every other benchmark.
Zen3 changes up the CCD/CCX setup. Now all 8 cores in a CCD share the L3 cache. Pretty much making the CCD and CCX one and the same, but full details on this are not 100% clear. But the change to 8 cores sharing L3 cache is going to be a big move. For applications where latency between CCX's caused performance slowdowns, we will see big gains. So games where Zen fell behind, should see Zen3 pull ahead. This is a big deal. This isn't even factored in to the IPC gains, this is just gains from removing slowdowns between cores. This is one of two big changes for the Zen Arch that Zen3 will bring to the table.
The Second big change for the Zen Arch Zen3 is being to the table is a much improved Floating Point Unit design. AMD has said it is upto 50% faster than the old one. Again this is a big boost for the Arch, as the old FPU wasn't exactly lacking either. While it doesn't mean much for normal workloads, a faster FPU is never a bad thing. It will do wonders for the server world. But videogames do also take advantage of the FPU, but it will largely be a case to case basis for what games get the most out of it.
The IPC gains for the integer unit (What most apps run on) are estimated to be in the 10-20% range depending on the application. Not sure if this also factors in the gains for the more effective node it will be on or not. But these are solid gains, much like from Zen1 to Zen2. Along with the IPC gains we should see 100-200mhz higher clocks on avg across the board. Zen2 already has higher IPC than Skylake, but Skylake has the benefit of higher clock speeds. And Skylake having all the cores sharing the same L3 cache, latency slowdowns were never the problem. Which is why it has held the gaming crown. This is about to change.
There are also a number of other changes to the Zen3 Arch that will speed things up and make it more efficient. It is going to be a massive improvement over Zen2. In many ways it will be a bigger leap than Zen1 to Zen2 was. It should be the first time AMD beats Intel pretty much across the board, including gaming. Zen3 won't need 5ghz+ clock speeds to beat Intel. Intel will be stuck on Skylake based CPUs till mid-late 2021 on the desktop. Which then we will get the 10nm Sunny Cove++ based Arch from Intel. At this point in time we will be talking about the upcoming Zen4 and AM5 motherboards. According to Intel's roadmap, they will not be able to match AMD in IPC performance till 2023-2024 or later. Which is around the time they plan to catch back up to TSMC Node wise. But we'll see how this works out for them. And while it may change in the future, for the next couple years Intel is going to lag behind AMD in Core count on their large server grade chips. Intel is already prepared to lose market share to AMD in this regard. And I wouldn't be surprised if this isn't the first area Intel plans to improve itself. Either with stacked cores or going with a chiplet design like AMD. All I know is AMD having more cores on a single Socket is going to hurt Intel in the server market. Especially when AMD is giving you better Power/Performance. Being able to pack more into a smaller space, while having the crown in Power/Performance is a big deal. AMD Already does this with Zen2.
AMD will also no doubt only get better in the mobile space. AMD is also aiming to bring Arm Based mobile chips to market. And their RDNA2 GPU Arch is looking to bring AMD's GPU Arch to being on Par or nearly on Par with Nvidia. For a company that had such a rocky last decade, I can't wait to see how they perform over the next 5 years. 2021 is going to be a good year for them.
