AMD has officially announced the Ryzen 9000 chips for desktop PCs. The Granite Ridge processors are designed for the Socket AM5 desktop platform and will work with existing motherboards after a BIOS update. Like the previous CPUs, the new Ryzen 9000 models use a chiplet layout in 16, 12, 8, and 6-core configurations, but the new Zen 5 computing architecture allows for a significant increase in relative performance per clock (IPC).
As expected, the “initial wave” of Ryzen 9000 series chips will include four models. At first glance, the processors of the new line structurally resemble the corresponding previous Ryzen 7000 models. The flagship of the series, the Ryzen 9 9950X, has a layout with 16 computing cores and SMT support, which allows processing up to 32 data streams simultaneously. The CPU accelerates up to 5.7 GHz, has a total of 80 MB of cache (L2+L3), and a TDP of 170 W.
The 12-core Ryzen 9 9900X has a peak operating frequency of 5.6 GHz and a combination of 76 MB of cache. At the same time, the power package of this model is 120 W, while the predecessor Ryzen 9 7900X had a TDP of 170 W.
The Ryzen 7 9700X model has a functional configuration of 8/16 and, compared to its predecessor (Ryzen 7 7700X), has added +100 MHz to the maximum operating frequency of 5.5 GHz. The structure of the cache buffers remained unchanged – 1 MB L2 for each computing core and 32 MB of total L3 cache memory. The total is 40 MB (8×1 MB + 32 MB). At the same time, the declared TDP for an 8-core model with one CCD is pleasantly surprising – 65 watts. For Ryzen 7 7700X, the package was 105 watts.
The junior model Ryzen 5 9600X has 6 computing cores and can process up to 12 threads simultaneously. The frequency formula is 3.9/5.4 GHz. Again, the maximum acceleration is 100 MHz higher than that of Ryzen 7 7600X. The cache capacity remained unchanged at 38 MB (6×1 MB + 32 MB), but the TDP here is also 65 W instead of 105 W for its predecessor.
AMD has not yet disclosed all the details of the Zen 5 computing architecture, which is likely to happen before the start of sales of new CPUs. However, the developers note that in this case we are talking about a number of improvements. These include improvements to the transition prediction unit, increased pipeline capabilities, and changes in parallelism efficiency.
The Zen 5 core also has a doubled instruction bandwidth for the Front-End part, data bandwidth from L2 to L1 and from L1 to FP, and an upgraded FPU unit to speed up AVX512 instruction processing and increase AI performance.
What practical results can we expect from Zen 5? Of course, independent tests will show this, but AMD has already given some guidance, assuring that compared to their predecessors on Ze4, chips with the new architecture have an average performance gain per clock cycle of ~16%. The chart shows that the actual difference will vary depending on the situation. We have from +10% in the game Far Cry 6 to +23% in Blender and even +35% in the GeekBench 5.4 AES XTS subtest.
The presentation also included a slide comparing the performance of the Ryzen 9 9950X with that of the Core i9-14900K. The results show that the new AMD flagship has a 7-56% advantage in multi-threaded work applications, while in certain games the new CPU was 4-23% faster.
The marketing department probably chose the most favorable situations for the presentation, but even the available data provides information for an initial assessment of the Ryzen 9000’s capabilities.
AMD has also pointed out the advantage of its platform when it comes to using PCI-E 5.0 drives that are connected directly to the CPU. In this case, Intel solutions have to divide the available CPU bus lines in the proportion of 8+8 (4+4), while the top Socket AM5 boards retain the number of PCI-E 5.0 x16 lines for the graphics adapter. For the most part, there is not much practical difference here, but visually the chart looks convincing.
The developers also claim that architectural changes in Zen 5 allow us to talk about a 20% advantage when it comes to AI acceleration. At least a similar difference is recorded when using Mistral’s LLM model running on CPU cores. By the way, Ryzen 9000 desktop chips did not receive a separate NPU unit, so in such cases, the processors of the line can rely on the capabilities of the main computers. But for sure, platforms based on these CPUs will not meet the requirements of the Copilot+ PC concept.
As we have already noted, Ryzen 9000 uses the familiar chiplet layout with several crystals on a single substrate. In the top configuration, there are two 8-core CCDs (Core Complex Die), as well as a separate I/O Die. This time, the computing modules are manufactured by TSMC according to 4 nm standards, while for Ryzen 7000, CCD blocks were produced using Taiwanese 5-nm technology. It is still difficult to predict how exactly this has affected the frequency potential of the processors, especially given that the regular maximum frequencies of the older Ryzen 9000 models have not increased. However, the decrease in TDP for 12/8/6-core CPUs may be the result of the transition to a more advanced crystal manufacturing process.
As for the I/O chip, it has remained virtually unchanged. The chip includes integrated graphics with two computing clusters based on the RDNA2 architecture. So there is no competition with APUs here. However, some improvements are still expected. AMD has officially announced support for DDR5-5600 RAM instead of DDR5-5200 in Ryzen 7000. In addition, we can talk about the acceleration of the Infinity Fabric bus, which will allow the use of faster RAM kits. According to previous leaks, the new AMD X870/X870E motherboards will allow the use of EXPO/XMP profiles up to DDR5-8000.
Ryzen 9000 processors are scheduled to start selling at the end of July. AMD hasn’t yet specified the recommended price tags. We expect to see values identical to those that were relevant at the time of the Ryzen 7000 launch. However, we expect an adjustment based on the actual retail price of Raphael chips.
Loading comments …