Wednesday, June 12th 2024
AMD Says Ryzen 9000 Series Won't Beat 7000X3D Series at Gaming
AMD's upcoming Ryzen 9000 "Granite Ridge" desktop processors based on the "Zen 5" microarchitecture won't beat the Ryzen 7000X3D series at gaming workloads, said Donny Woligroski, the company's senior technical marketing manager, in an interview with Tom's Hardware. The new "Zen 5" chips, such as the Ryzen 7 9700X and Ryzen 9 9950X, will come close to the gaming performance of the 7800X3D and 7950X3D, but won't quite beat it. The new processors, however, will offer significant generational performance uplifts in productivity workloads, particularly multithreaded workloads that use vector extensions such as VNNI and AVX512. The Ryzen 7 7800X3D remains the fastest gaming desktop processor you can buy, it edges out even Intel's Core i9-14900KS, in our testing.
Given this, we expect the gaming performance of processors like the Ryzen 7 9700X and Ryzen 9 9950X to end up closer to those of the Intel Core i9-13900K or i9-14900K. Gamers with a 7000X3D series chip or even a 14th Gen Core i7 or Core i9 chip don't have much to look forward to. AMD confirmed that it's already working on a Ryzen 9000X3D series—that's "Zen 5" with 3D V-cache technology, and is sounds confident of holding on to the title of having the fastest gaming processors. This doesn't seem implausible.
Intel, in its recent "Lunar Lake" architecture reveal, went deep into the nuts and bolts of its "Lion Cove" P-core, where it claimed that the core posts a 14% IPC increase over the "Redwood Cove" P-core powering "Meteor Lake," which in turn has similar IPC to the "Raptor Cove" P-core powering the current 14th Gen Core processors. Intel intends to use "Lion Cove" P-cores in even its Core Ultra "Arrow Lake-S" desktop processors. Given that 3D V-cache gave "Zen 4" a 20-25% boost in gaming performance, a similar performance boost to "Zen 5" could make the 9000X3D series competitive with "Arrow Lake-S," if Intel's claims of a 14% IPC gain for the "Lion Cove" P-core holds up. That said, AMD in its interview stated that 3D V-cache may not add the kind of gaming performance gains to "Zen 5" that it did to "Zen 4."
AMD is building the "Zen 5" 8-core CCD on the 4 nm foundry process, which is expected to have the TSV foundation for stacked 3D V-cache memory, but there's an ace up AMD's sleeve. AMD hasn't ruled out the possibility of "Zen 5" having an expandable dedicated L2 cache. To a question by Tom's Hardware on whether the L2 cache is expandable on "Zen 5," AMD replied "Absolutely, if you get to finer-grain 3D interconnect. So we're at 9-micron through silicon via (TSV) pitches today. As you go down to, you know, 6-, 3-, 2- micron and even lower, the level of partitioning can become much finer-grained," It's important to note here, that this is not a confirmation on AMD's part. AMD didn't define the specific pitch required for an L2 cache.
If true, what this means is that in the 9000X3D series, the company could give the CCD a larger 3D V-cache chiplet, which not just expands the on-die L3 cache from 32 MB to 96 MB, but also increases the sizes of the dedicated L2 caches of each core. The "Zen 5" microarchitecture sees each core get 1 MB of dedicated L2 cache, which the new 3D V-cache chiplet could expand.
The L2 cache operates at a higher data-rate than the shared L3 cache, and uses a faster SRAM physical media. The next-gen 3D V-cache chiplet could hence feature two distinct kinds of SRAM—the 64 MB L3 SRAM that expands the on-die 32 MB L3 SRAM; and eight L2 cache SRAM units to expand each of the eight on-die L2 caches.
The L2 cache is expected to play a major role in gaming performance for next-gen processors, and Intel has significantly expanded it for "Lion Cove" P-cores with both "Lunar Lake" and the upcoming "Arrow Lake." On "Lunar Lake," the four P-cores each have a 2.5 MB of dedicated L2 cache. On "Arrow Lake," the same P-core is expected to get 3 MB of dedicated L2 cache. So AMD probably understands the importance of fattening not just the L3 cache, but also the L2.
The rumor mill is abuzz with reports of AMD bringing in the Ryzen 9000X3D series within 2024, with some sources pointing to a Q4-2024 debut, which should time them alongside Intel's launch of the Core Ultra "Arrow Lake-S" desktop processors.
Source:
Tom's Hardware
Given this, we expect the gaming performance of processors like the Ryzen 7 9700X and Ryzen 9 9950X to end up closer to those of the Intel Core i9-13900K or i9-14900K. Gamers with a 7000X3D series chip or even a 14th Gen Core i7 or Core i9 chip don't have much to look forward to. AMD confirmed that it's already working on a Ryzen 9000X3D series—that's "Zen 5" with 3D V-cache technology, and is sounds confident of holding on to the title of having the fastest gaming processors. This doesn't seem implausible.
AMD is building the "Zen 5" 8-core CCD on the 4 nm foundry process, which is expected to have the TSV foundation for stacked 3D V-cache memory, but there's an ace up AMD's sleeve. AMD hasn't ruled out the possibility of "Zen 5" having an expandable dedicated L2 cache. To a question by Tom's Hardware on whether the L2 cache is expandable on "Zen 5," AMD replied "Absolutely, if you get to finer-grain 3D interconnect. So we're at 9-micron through silicon via (TSV) pitches today. As you go down to, you know, 6-, 3-, 2- micron and even lower, the level of partitioning can become much finer-grained," It's important to note here, that this is not a confirmation on AMD's part. AMD didn't define the specific pitch required for an L2 cache.
If true, what this means is that in the 9000X3D series, the company could give the CCD a larger 3D V-cache chiplet, which not just expands the on-die L3 cache from 32 MB to 96 MB, but also increases the sizes of the dedicated L2 caches of each core. The "Zen 5" microarchitecture sees each core get 1 MB of dedicated L2 cache, which the new 3D V-cache chiplet could expand.
The L2 cache operates at a higher data-rate than the shared L3 cache, and uses a faster SRAM physical media. The next-gen 3D V-cache chiplet could hence feature two distinct kinds of SRAM—the 64 MB L3 SRAM that expands the on-die 32 MB L3 SRAM; and eight L2 cache SRAM units to expand each of the eight on-die L2 caches.
The L2 cache is expected to play a major role in gaming performance for next-gen processors, and Intel has significantly expanded it for "Lion Cove" P-cores with both "Lunar Lake" and the upcoming "Arrow Lake." On "Lunar Lake," the four P-cores each have a 2.5 MB of dedicated L2 cache. On "Arrow Lake," the same P-core is expected to get 3 MB of dedicated L2 cache. So AMD probably understands the importance of fattening not just the L3 cache, but also the L2.
The rumor mill is abuzz with reports of AMD bringing in the Ryzen 9000X3D series within 2024, with some sources pointing to a Q4-2024 debut, which should time them alongside Intel's launch of the Core Ultra "Arrow Lake-S" desktop processors.
141 Comments on AMD Says Ryzen 9000 Series Won't Beat 7000X3D Series at Gaming
How much faster is the 9700X compared to a 7700X? 16%?
There you have it. A 7800X3D is faster in gaming compared to a 9700X....
We won't know for sure until TPU has one in their hands and tested it.
The 2950X is a 16 core Threadripper, which will be absolutely decimated by a 16 core 9950X.
"AMD Says Ryzen non-X3D 9000 Series Won't Beat 7000X3D Series at Gaming"
Wait for the 9800X3D, duh!
The part about 9000X3D chips potentially using L2 cache on the 3D part is very interesting however. If true, they may see a higher performance boost from the 3D cache than previous processors that only used it for L3 cache.
I understand the motivation is to trounce whatever stronghold Intel has left in the HPC space but this will come at a big cost for desktop and mobile chips. Double pumped AVX512 is already quite the improvement on limited use cases like RPCS3 and other software. Unless they are planning to cut the number of Big cores and return to double pumping for Zen5C, these will be expensive chips for the performance provided.