Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
AMD announces Ryzen 3000 series: up to 12 cores, up to 4.6 GHz, up to 64 MB L3 cache, available 7/7
The full lineup is here:
https://www.anandtech.com/show/14407/amd-ryzen-3000-announced-five-cpus-12-cores-for-499-up-to-46-ghz-pcie-40-coming-77
AMD claims that their 8-core CPU with turbo up to 4.5 GHz beats Intel's 8-core Core i9-9900K outright at Cinebench, both in single-threaded and multi-threaded versions. That would point to AMD having a considerable IPC lead in Cinebench. Cinebench has been a very favorable benchmark to AMD ever since the launch of Ryzen, but this result still could point to third generation Ryzen roughly having IPC parity with Intel's Sky Lake and its many refreshes.
The dies that AMD is using have 8 CPU cores on them. The 12-core CPU has two CPU dies, presumably with six cores each enabled. This might well take some special binning, such as selectively disabling the two cores that couldn't clock that well or burned too much power, even though the cores actually worked. It also leaves open the possibility of AMD launching a 16-core CPU on the same platform in the future; my guess is that AMD didn't do so just yet because it wouldn't be able to clock as high.
As always, there's a considerable need for independent benchmarks to add more context. There are two ways at least two ways that the announced chips are outliers that could skew benchmark results. First, it takes two hops for the CPU to get to memory. One of those is inside the socket, going from a die with CPU cores to an I/O die. But more hops always adds latency, and high memory latency could be a serious problem for some purposes.
Second, that's 32 MB of L3 cache per CPU die, which is how you get to 64 MB of L3 cache for the two-die 12-core version. For comparison, as best as I can tell, there has never been an x86 CPU with 40 MB of L3 cache (or L2 cache) on a die. Intel's 28-core Xeon Platnium and 72-core Xeon Phi both come a little shy of that. So this is an enormous amount of cache on a tiny die. To a considerable degree, it's surely AMD's effort at making a ton of CPU cores useful even with only two channels of DDR4. But all of that L3 cache will surely benefit some benchmarks much more than others.
This does, however, mean that the rumors about 5 GHz turbo boost were simply wrong. Still, this is pretty good clock speeds for early parts on a new process node. It sounds as if when Intel moves to 10 nm, clock speeds are going to go way down, at least initially.
https://www.anandtech.com/show/14407/amd-ryzen-3000-announced-five-cpus-12-cores-for-499-up-to-46-ghz-pcie-40-coming-77
AMD claims that their 8-core CPU with turbo up to 4.5 GHz beats Intel's 8-core Core i9-9900K outright at Cinebench, both in single-threaded and multi-threaded versions. That would point to AMD having a considerable IPC lead in Cinebench. Cinebench has been a very favorable benchmark to AMD ever since the launch of Ryzen, but this result still could point to third generation Ryzen roughly having IPC parity with Intel's Sky Lake and its many refreshes.
The dies that AMD is using have 8 CPU cores on them. The 12-core CPU has two CPU dies, presumably with six cores each enabled. This might well take some special binning, such as selectively disabling the two cores that couldn't clock that well or burned too much power, even though the cores actually worked. It also leaves open the possibility of AMD launching a 16-core CPU on the same platform in the future; my guess is that AMD didn't do so just yet because it wouldn't be able to clock as high.
As always, there's a considerable need for independent benchmarks to add more context. There are two ways at least two ways that the announced chips are outliers that could skew benchmark results. First, it takes two hops for the CPU to get to memory. One of those is inside the socket, going from a die with CPU cores to an I/O die. But more hops always adds latency, and high memory latency could be a serious problem for some purposes.
Second, that's 32 MB of L3 cache per CPU die, which is how you get to 64 MB of L3 cache for the two-die 12-core version. For comparison, as best as I can tell, there has never been an x86 CPU with 40 MB of L3 cache (or L2 cache) on a die. Intel's 28-core Xeon Platnium and 72-core Xeon Phi both come a little shy of that. So this is an enormous amount of cache on a tiny die. To a considerable degree, it's surely AMD's effort at making a ton of CPU cores useful even with only two channels of DDR4. But all of that L3 cache will surely benefit some benchmarks much more than others.
This does, however, mean that the rumors about 5 GHz turbo boost were simply wrong. Still, this is pretty good clock speeds for early parts on a new process node. It sounds as if when Intel moves to 10 nm, clock speeds are going to go way down, at least initially.
![[Deleted User]](http://wa.vanillicon.com/v2/a5231d9a9017cdf5f82125ccaf807f8f.svg)
Comments
4.6, maybe a hair higher, is probably about the ceiling you can expect on all of these chips. That isn't shabby by any means. Sure, it's not 5.0, but it's not shabby either. And hey, backwards motherboard compatibility! And a decent air cooler in the box. And you don't have to disable hyperthreading over security concerns. There's a lot of good news here, even if it didn't quite reach the hype that was out there.
Benchmarks will tell the story though, haven't seen much other than a Cinebench number. I remain cautiously optimistic, and more excited about the R7/R9 @105W than I am the 8/16 5.0 i9 that was announced yesterday @???W.
Depending on benches, it may be enough to finally get me to consider upgrading some of my Ivy Bridge/Haswell machines.
Ryzen 5 3600: 6 cores, 12 threads, 3.6 GHz base, 4.2 GHz turbo, PCI-E 4.0, 65 W, $200
Ryzen 5 2600X: 6 cores, 12 threads, 3.6 GHz base, 4.2 GHz turbo, PCI-E 3.0, 95 W, $200
Ryzen 5 1600X: 6 cores, 12 threads, 3.6 GHz base, 4.0 GHz turbo, PCI-E 3.0, 95 W, $200
why would you want an older one? Especially when you can drop the new CPU into an old motherboard if you have it. The Ryzen 5 3600 will probably offer better IPC than the older parts, too.
Brenics ~ Just to point out I do believe Chris Roberts is going down as the man who cheated backers and took down crowdfunding for gaming.
As far as Navi, that's a GPU, so it's not even relevant to this thread.
I would be interested to see what the 3800X looks like delidded. Just to make sure it's 1 CPU complex instead of 1 and a dummy.
The CPUs above it give very little performance increase for normal gamer or office user.
https://images.anandtech.com/doci/13909/cpu44_678x452.jpg
"My Fantasy is having two men at once...
One Cooking and One Cleaning!"
---------------------------
"A good man can make you feel sexy,
strong and able to take on the whole world...
oh sorry...that's wine...wine does that..."
This time I am going to wait a while.
Gen2 Ryzen didn't provide any substiantial performance improvement to justify the upgrade and while we will need to wait how gen3 turns out, I wouldn't be expecting much of a difference there either.
Regardless, it is rather moot since AM4 is supposed to be discontinued next year.
If you were a Ryzen 1 buyer, the option to upgrade to Zen2 is great. If you are just jumping into Ryzen with Zen2, then there isn’t any great reason to get anything other than a X570 motherboard now
it sure beats the needing a new motherboard for each new revision of Skylake for no good reason other than $$$ that has been going on for the last five years.
https://adoredtv.com/amd-confirms-no-pcie-4-0-support-on-300-and-400-series-motherboards/
PCI-e 4.0 is not necessary right now. I don't think you can saturate a PCI-e 3.0 x8 with existing GPUs.