There is currently a lot of movement in the information jungle and so this week a lot of things were buzzing through the air like Matisse 2.0, Refresh, answer to Intel’s Comet-Lake-S, Ryzen 7 3750X and 3850X or 3700XT to 3900XT. First of all I deliberately kept out of it and rather asked questions, collected my own appetizers and then tried to put everything back together in a more or less plausible way. Thereby a quite clear picture results, whereby also new questions were raised.
Matisse Refresh is Matisse 2 (MTS 2) vs. Comet-Lake S
This with Matisse 2.0 is already saved and is abbreviated internally by AMD. It is only logical that the whole thing is now supposed to hold up with lower latencies and a higher clock rate until Vermeer comes along. The slight chaos that currently appears in the environment also shows that there is a slight actionism in front of a relaxed planned economy. Because already with the AGESA versions it becomes slightly confusing. There was only recently a ComboPi version without MTS 2 and this was then replaced in the meantime by a ComboPIv2 with a maximum FCLK of 2000 MHz, so there is now even ComboPI in the first version again, but for the first time with MTS 2 but lower FCLK.
There were also reports in some media that AMD would have the AGESA versions start at 188.8.131.52 again, but apparently this only referred to the ComboPIv2 variant, which has now been replaced by ComboPI 184.108.40.206 with MTS 2. It seems unlikely that the board partners would then manually convert these versions to a different naming scheme. But it would be more plausible to either put the v2 on hold for the time being, because MTS 2 was installed at short notice and can’t be compatible with ComboPIv2 in such a short time, or to hold ComboPIv2 back for the new chipsets and need some more time.
What can also be regarded as certain is the naming scheme of the new Ryzen 5 3600XT and probably also Ryzen 7 3700XT, although here the B0 stepping was also indicated several times as a revision. What is missing, but would be conclusive, is another Ryzen 3 3300XT, while everything about it might not be a real refresh, if the following assumption, which has now been mentioned several times by different sources, is correct.
Is Matisse 2 a Renoir with graphics disabled?
Let’s first look at Renoir and the block diagram. We see in dark red the areas that are used for the integrated graphics. And we remember that the Intel CPUs (almost) all come with an integrated graphics unit.
If you simply leave out these units, you would get a completely new CPU, which could offer even more performance with regard to the rather hindering latencies of current Matisse CPUs due to the CCD/CCX design including IOD. In addition, there would also be a possibly higher boost clock, because the full 65 watt TDP would then only be available for the CPU. A new entry in the BIOS disables AVFS (Adaptive Voltage & Frequency Scaling), which AMDs GPUs use since Polaris and the Bristol-Ridge-APUs. This would look like this in the block diagram:
The statements of some OEMs about Renoir also read quite interesting. For example, they think that Renoir (as CPU) would not bring any big performance advantages over the current Matisse processors, because they (AMD) have to put the GPU on the same package (in the same case). They will release it (Renoir as complete APU) only in the PRO version for PC assemblers/SI and channel distribution and a retail or BOX version is not planned (for the time being). Furthermore it was said again that the B450 mainboards will not support Renoir APUs.
Because this limitation could indicate that you want to annoy Intel again and prefer to bring the chips without graphics unit as faster Matisse CPUs on the market. Lower latencies, slightly higher boost clock. The only question mark would then be the obvious lack of PCIe 4.0 and the rather scarce number of lanes for the new desktop CPU, if one relies on the specs of the mobile variants. But maybe AMD still has unreleased reserves that have not yet been revealed ?