After one year of Ryzen 3000 (“Matisse”), the replacement of the current Ryzen 5 3600X, Ryzen 7 3800X and Ryzen 9 3900X is approaching, by adding MTS2 (internal for “Matisse Refresh”), an improved XT replacement, which should make waiting for Ryzen 4000 (“Vermeer”) a little easier. And probably also a little bit the wallet, because you are aiming for the same EIA as a year ago. Changing of the guard sure, but not a real changing of the guard either. More likely it’s the hastily-convened late shift, while Intel’s 10. Generation on night shift awaits the dawn.
Today I’m testing the Ryzen 9 3900XT, which was kindly provided by AMD. The reason why I didn’t sample any smaller CPUs this time must have been the amount of CPUs that were available at all. So I have to design this test a little bit differently and adjust its content. In today’s test, I logically position the new XT model against the old one, against a Ryzen 9 3950X and also fairly against an Intel Core i9-10900K. All other CPUs in the test, including the motherboards, are from my own stock, so there are no dependencies here.
A total of three CPUs are actually ennobled to the XT version, whereby the old X versions are to continue running. In terms of price, the XT models are based on the EIA of one year ago, which creates a little shadow. Because the street prices speak a very clear language and the distance to the new, old EIA does not really make the slightly improved XT Troika more attractive. If the customer will be presented with an XT for an X, whose surcharge is questionable, I can only answer this question for the Ryzen 9 3900XT. But I’ll do my best, I promise!
Series | Name | Cores | Threads | Base MHz | Boost MHz | L3 | PCIe 4 | TDP |
Ryzen 9 | 3950X | 16C | 32T | 3.5 | 4.7 | 4×16 MB | 16+4+4 | 105W |
Ryzen 9 | 3900XT | 12C | 24T | 3.8 | 4.7 | 4×16 MB | 16+4+4 | 105W |
Ryzen 9 | 3900X | 12C | 24T | 3.8 | 4.6 | 4×16 MB | 16+4+4 | 105W |
Ryzen 9 | 3900 | 12C | 24T | 3.1 | 4.3 | 4×16 MB | 16+4+4 | 65W |
Ryzen 7 | 3800XT | 8C | 16T | 3.9 | 4.7 | 2×16 MB | 16+4+4 | 105W |
Ryzen 7 | 3800X | 8C | 16T | 3.9 | 4.5 | 2×16 MB | 16+4+4 | 105W |
Ryzen 7 | 3700X | 8C | 16T | 3.6 | 4.4 | 2×16 MB | 16+4+4 | 65W |
Ryzen 5 | 3600XT | 6C | 12T | 3.8 | 4.5 | 2×16 MB | 16+4+4 | 95W |
Ryzen 5 | 3600X | 6C | 12T | 3.8 | 4.4 | 2×16 MB | 16+4+4 | 95W |
Ryzen 5 | 3600 | 6C | 12T | 3.6 | 4.2 | 2×16 MB | 16+4+4 | 65W |
Ryzen 5 | 3500X | 6C | 6T | 3.6 | 4.1 | 2×16 MB | 16+4+4 | 65W |
Ryzen 3 | 3300X | 4C | 8T | 3.8 | 4.3 | 1×16 MB | 16+4+4 | 65W |
Ryzen 3 | 3100 | 4C | 8T | 3.6 | 3.9 | 2×8 MB | 16+4+4 | 65W |
Binning or a little more?
For the new 3000XT processor family AMD advertises an increase of the turbo frequency by 100-200 MHz at the same power consumption and with up to approx. 4% more performance. This can be confirmed in a general way, without spoiling it already. AMD states that this is due to the use of an optimized 7nm manufacturing process, but leaves the specific answers open. One can only assume that the improvement could be due to a minor BKM or PDK update, which now allows TSMC to tune the entire process to a better voltage/frequency curve.
I have tested that it is not only a pure binning with my own manually selected Ryzen 9 3950X, which is said to have the best factory binning currently available. I used this CPU in 3+3 mode as a “real” twelve-core CPU and also lowered the respective voltages slightly. So this synthetic 3900XT copy had about the same power consumption as the later tested original and also looked very similar in clock speed. Similar, but not anymore. Because in the course of the very extensive benchmarks, there were reproducible outliers again and again, where up to 10% and more additional performance was created, which can hardly be explained in this way. But I will come to that in a moment with the benchmarks.
Test system and task definition
In order not to limit the CPUs thermally by the test setup and to enable boosting algorithms such as Thermal Velocity Boost, which are fair to Intel’s improved boosting algorithms, the Alphacool Chiller Ice Age 2000 comes into play again together with the XPX Pro water block with the necessary brackets, because water at a constant 20 °C is a solid basis for consistent results.
The motherboard used is the old acquaintances who already have their measurement conversions behind them. I have run all current CPUs, as the normal buyer probably does, with a set of Patriot Viper DDR4 3600 (PC4-28800) in the stored XMP profile. The Ryzen 3xxx can’t synchronize more than that anyway, and it doesn’t really do any good anymore. Unfortunately I could not afford to play games like RAM to specs or slow down motherboards due to lack of time. So far, everything is default and thus exactly as the normal customer would probably screw it together.
Here once again the tabular compilation of the test setup:
Test System and Equipment |
|
---|---|
Hardware: |
AMD Ryzen 9 3900XT, 3900X, 3950XT Intel Core i9-10900K Intel Core i9-9900 K, i9-9900KS, i7-9700K Intel Core i9-9980XE, i9-9960X General |
Cooling: |
Alphacool ice block XPX Pro (1151, 1200, 2066, AM4) Alphacool Ice Age 200 Chiller (modified) Thermal Grizzly Kryonaut |
Case: |
Open Benchtable |
Monitor: | BenQ PD3220U |
Power Consumption: |
Non-contact direct current measurement on PCIe slot (riser card) MCU-based shunt measuring (all rails and VRM) |
Thermal imager: |
1x Optris PI640 + 2x Xi400 Thermal Imagers Pix Connect Software Type K Class 1 thermal sensors (up to 4 channels) |
Acoustics: |
NTI Audio M2211 (with calibration file) Steinberg UR12 (with phantom power for the microphones) Creative X7, Smaart v.7 Own anechoic chamber, 3.5 x 1.8 x 2.2 m (LxDxH) Axial measurements, perpendicular to the centre of the sound source(s), measuring distance 50 cm Noise emission in dBA (slow) as RTA measurement Frequency spectrum as graphic |
OS: | Windows 10 Pro (all Updates) |
AMD Ryzen 9 3900XT, 12C/24T, 3.80-4.70GHz, boxed ohne Kühler (100-100000277WOF)
AMD Ryzen 9 3900X, 12C/24T, 3.80-4.60GHz, boxed (100-100000023BOX)
AMD Ryzen 9 3950X, 16C/32T, 3.50-4.70GHz, boxed ohne Kühler (100-100000051WOF / 100-100000051BOX)
Intel Core i9-10900K, 10C/20T, 3.70-5.30GHz, boxed ohne Kühler (BX8070110900K)
Kommentieren