The highlights include turbo speeds of at least 5 GHz, single-threaded performance increases of at least 15%, PCI Express 5.0 support for both a video card and an SSD, DDR5 support, and an integrated GPU in the I/O die.
That the parts used for the single-threaded performance increase are unspecified makes the claim meaningless, though AMD hasn't normally been aggressive about cherry-picking corner cases to make wildly inflated claims, so it's likely to be more or less accurate. It sounds like much of the gain will come from higher clock speeds.
The new AM5 socket will support up to 170 W, which likely means that Zen 4 parts will have a turbo power of 170 W, up from the current 142 W for Zen 3 parts. More performance coming at the expense of more power seems to be a common theme in computer parts in recent years, though at least 170 W would remain far below the 241 W of Alder Lake.
The CPU core chiplets move from TSMC 7 nm to 5 nm. AMD claims that the L2 cache per core doubles to 1 MB, though they didn't say anything about L3 cache.
The I/O die moves from Global Foundries 14 nm to TSMC 6 nm. An I/O-heavy chip can have its minimum size dictated by the need to have enough space for pins for the I/O. It's possible that that left AMD with a bunch of free transistor space to use, so they decided to put in an integrated GPU. It's also possible that it's a very small integrated GPU with only two or three compute units or some such, which is still plenty enough to display the desktop. We'll learn more about the chip when it launches in several months.
Zen 4 parts will be DDR5 only. That's in contrast to Intel's strategy for Alder Lake offering both DDR4 and DDR5 support. The more time passes, the less need there is for DDR4 support, and it's likely that Intel's next generation won't support DDR4, either.
Comments
Not that 15% is a bad number. It's just not what I would expect when you see the combination of all 3 - TDP bump, clock bump, and process node jump.
Uh oh. That 170 W might be the nominal (and mostly meaningless) TDP, up from 105 W on current generations. The turbo power that is currently 142 W goes up to 230 W. That's still a little shy of the 241 W of Intel's Alder Lake. But it's also still a lot.
Those rumors of next generation high end video cards needing 600 W are looking more credible.
Ill always be a Intel man though. And Ill always be a Geforce guy.
Brand loyalty seems to be a rare breed. Cost and % increase of said device seems to be most important.
Have not yet though. Maybe there is hope?
Nah there is really none for me or the human race.
The workstation Threadripper Pro 3000 and 5000 series both go down to 12 cores. But they also have significantly lower turbo clock speeds than the mainstream consumer Ryzen line. When all cores are in use, the higher TDP does allow the cores to clock higher. But if that's the use case, more cores are going to help far more than a higher TDP.
Rather, the workstation Threadripper Pro parts are more about memory and I/O. They have 128 PCI-E 4.0 lanes, and Socket AM5 is going to top out at 28 lanes, albeit PCI-E 5.0. They have eight 64-bit DDR4 channels and support up to 2 TB of memory, as compared to two channels and probably either 64 or 128 GB for the mainstream consumer parts.
it's either having the massive core count, because they have something that can use that, and Ryzen isn't going to replace that - even if the new high-end Ryzen 9 caps out past 16 cores, it's still going to be a long way from 64.
Or
It's having the massive amount of PCI lanes available, because they have a lot of SSDs or NICs or other hardware that they want to drive, and this isn't going to replace that either.
I think you are right, that for the most part, the TR models under 24 cores seem rather pointless -- unless you need all those PCI lanes for something (and I can't imagine what, but I won't discount that someone might be able to do something with it)?
These same folks are also seriously pissed that TR 5000 so far is OEM only, and doubly pissed that it's only Lenovo. It makes me wonder if AMD is going to kill it off and just make anyone that wants either of those two criteria to move to a proper Epyc server architecture.
If you are holding out for the perfect game, the only game you play will be the waiting one.
Even if they don't release them immediately, I'd find it surprising if they haven't already plans to increase Ryzen's max core count. At which point they need their socket to be able to deliver enough power.