AMD Ryzen 9 7950X3D Chiplet Boost Frequencies, Voltages, Thermal, and Throttling Benchmarks
We turned our eyes to testing each chiplet individually to tease out the difference between the chiplets. The 7950X3D has two core compute die (CCD) chiplets lurking under the hood, and the CCD chiplet with the 3D V-Cache will run at lower clock speeds than the bare chiplet. This occurs because of the silicon shim stacked atop the CPU cores on the chiplet with 3D V-Cache — this shim transfers heat from the cores to the integrated heat spreader (IHS), but inevitably reduces the efficiency of the thermal transfer from the cores.
In effect, the shim traps a small amount of heat. As such, AMD sets a 1.4V limit for the entire processor, but further limits the 3D V-Cache CCD chiplet to ~1.1V to keep heat within safe boundaries. We put this to the test to generate hard numbers.
As you can see in the images above, we put the voltage and frequencies to the test by disabling one CCD and then running through a spate of standard heavily threaded applications (Cinebench, HandBrake, AVX-heavy y-cruncher) to measure power and thermals in multi-threaded work, and then another series of lightly-threaded apps to check performance in lighter fare. Then we tested the other chiplet using the same methodology.
Tom's Hardware | Single-Threaded Peak | Multi-Threaded Sustained | nT Power | Voltage (peak) |
CCD 0 (3D V-Cache) | 5.25 GHz | 4.85 GHz | 86W | 1.152 |
CCD 1 (No extra cache) | 5.75 GHz | 5.3 GHz | 140W | 1.384 |
The tests above allowed us to generate this table with the frequencies and voltages for both CCDs. As you can see, the chiplet with 3D V-Cache (CCD 0) has single-core and multi-core clock rates of 5.25 and 4.85 GHz, respectively, significantly lower than the 5.75 and 5.3 GHz we measured with the chiplet without cache (CCD 1). We conducted these same tests with both CCDs active and easily achieved the rated 5.7 GHz boost, indicating AMD’s thread targeting implementation works correctly.
The 3D V-Cache chiplet also peaks at 1.152V during threaded work, while the bare chiplet peaks at 1.384V. This translates into much more power consumption — the bare chiplet pulls 140W when run alone, while the 3D V-Cache chiplet tops out at 86W. The bare CCD probably doesn’t sustain that level of power consumption often during normal usage due to both the additional heat from the other chiplet and power limits — we recorded a peak of 132W during a Prime95 workload for the entire chip with both CCDs active. That’s significantly lower than the rated 162W PPT limit.
Ryzen 9 7950X3D Thermal Throttling Tests
AMD specs an 89C limit for the Ryzen 9 7950X, lower than the 95C spec for the vanilla 7950X. We were curious if that limit applies to both chiplets, so we turned to a Prime95 test to take a closer look. We often don't include Prime95 power measurements in our standard CPU reviews, largely because there is a massive disconnect between this extremely rigorous stress test and most real-world applications' power consumption and thermal load. Still, we want to push the chips to their throttle point for this test.
To push the processor's limits even harder, we unplugged the fans on the Corsair H115i cooler but left the pump running (unplugging the pump caused clocks to drop too rapidly for our logging to provide any granularity). We then kicked off a Prime95 run with small FFTs, but with AVX instructions disabled. This type of Prime 95 stress test is brutal, and you won't see this type of stress during even the heaviest normal use. As such, remember that we're doing this for science, not as an indicator of how the chip would function in your PC.
The results show that the bare chiplet without the additional cache (CCD 1) reaches 94C before dropping to 89C when we remove the fan power. That 94C peak is actually above the rated 89C for the chip. Under normal circumstances, the chip would aggressively throttle power, and thus clock rates, to reduce thermals and stay under the target 89C. However, we can see that AMD allows this chiplet to run at a higher thermal threshold that matches the vanilla 7950X’s 95C limit.
In contrast, the chiplet with 3D V-Cache (CCD 0) never exceeds 89C, showing that it is programmed for the lower 89C threshold we see on the spec sheet.
You'll also notice that the bare chiplet begins the test around 140W and finishes at 90W when the chip is aggressively thermal throttling. In contrast, the 3D V-Cache CCD chiplet never exceeds 88W and hits a low of 78W at the end of the test. The 3D V-Cache CCD also drops to slightly lower clocks at the end of the test. That difference in power draw at the end of the test is largely due to the challenges associated with dissipating heat from the CPU cores covered by a silicon shim on the 3D V-Cache-equipped chiplet.
To be clear, you would never encounter these conditions during normal use, and the Ryzen 9 7950X3D runs perfectly fine within its specifications. Thermal dissipation has been one of the major sticking points that have prevented high-performance 3D-stacked chips from going mainstream, but AMD has done an amazing engineering job in bringing thermals under control enough to deliver a chip that provides excellent performance within an acceptable TDP threshold. Combining both a bare chiplet and a 3D-stacked chiplet is an ingenious design that helps provide the best of both worlds in a diverse range of workloads.
- MORE: Best CPU for gaming
- MORE: CPU Benchmark Hierarchy
- MORE: Intel vs AMD
- MORE: How to Overclock a CPU