CPU Benchmark Performance: Science

Our Science section covers all the tests that typically resemble more scientific-based workloads and instruction sets. For our 2023 CPU suite, we've also added SciMark 2.0 which measures numerical kernels and various computational routines found in numeric coding.

We are using DDR5 memory on the Ryzen 9 7950X3D and the other Ryzen 7000 series we've tested. This also includes Intel's 13th and 12th Gen processors. We tested the aforementioned platforms with the following settings:

  • DDR5-5600B CL46 - Intel 13th Gen
  • DDR5-5200 CL44 - Ryzen 7000
  • DDR5-4800 (B) CL40 - Intel 12th Gen

All other CPUs such as Ryzen 5000 and 3000 were tested at the relevant JEDEC settings as per the processor's individual memory support with DDR4.

Science

(2-1) 3D Particle Movement v2.1 (non-AVX)

(2-2) 3D Particle Movement v2.1 (Peak AVX)

(2-3) yCruncher 0.78.9506 ST (250m Pi)

(2-4) yCruncher 0.78.9506 MT (2.5b Pi)

(2-4b) yCruncher 0.78.9506 MT (250m Pi)

(2-5) SciMark 2.0: Composite

(2-5b) SciMark 2.0: Monte Carlo

(2-5c) SciMark 2.0: Fast Fourier Transform

(2-5d) SciMark 2.0: Sparse Matrix Multiply

(2-5e) SciMark 2.0: Dense LU Matrix Factorization

(2-5f) SciMark 2.0: Jacobi Successive Over-Relaxation

(2-6) Primesieve 1.9.0: High Core Count

Our science tests focus on scientific-level workloads, and from the above figures, the Ryzen 9 7950X3D performs respectably, as we would expect. The additional 64 MB (128 MB in total across both CCXs) to one of the CCXs does actually provide some benefit in these types of workloads, at least compared directly to the 7950X.

CPU Benchmark Performance: Power, Office And Web CPU Benchmark Performance: Simulation
Comments Locked

122 Comments

View All Comments

  • cruiseliu - Monday, February 27, 2023 - link

    Thanks, looking forward to see updates.
  • Otritus - Monday, February 27, 2023 - link

    AMD does not have an intelligent scheduler built into their CPUs to handle the heterogeneous design like Intel. As a result that have a driver that is basically just: if a game is detected then only schedule on the V-Cache CCD. This driver is imperfect as is for games as seen with Factorio, and clearly not optimized for other workloads. The Windows 11 scheduler will typically park threads on the higher frequency CCD because the driver is supposed to tell it not to. As a result, the driver isn’t activating causing threads to be scheduled on the higher frequency CCD and performance is worse than the 7950X because the X3D has a lower power limit and all-core turbo.
  • cruiseliu - Monday, February 27, 2023 - link

    Yeah I know the scheduler is basically "if game then cache else frequency".
    But Dwarf Fortress and Factorio are also games, though not 3D games. So it seems hard to say which CCD will be used.
  • Gavin Bonshor - Monday, February 27, 2023 - link

    Yeah, that's right about the 'scheduler.' It hinges on Xbox Game Bar feeding the information to the drivers if A. If it's a game, park one CCD and enable V-Cache CCD or B. Not a game, then act as normal.

    In relation to Factorio, I know our version in our suite is without the UI, which is possibly why it wasn't flagged. As this is more of a CPU benchmark in our suite than a game, it is reflected in the method of data.

    That's why I did a side test with Factorio on the CCD with V-Cache, which, as expected, boosted performance by nearly double.

    I'm in the process of doing additional testing, so bare with me
  • Bruzzone - Monday, February 27, 2023 - link

    7X 3D 1P + 1P offload engine seems underwhelming on initial assessment. mb
  • nightbird321 - Monday, February 27, 2023 - link

    Wow AnandTech, pulled out the slowest DDR5 money can buy for this review and didn't pair the Ryzen 7000 with "sweetspot" 6000 speed ram, and put 5 pages of productivity to 4 pages of gaming benchmarks for a "Gaming" CPU. Just WOW. The other site known for biased Intel reviews did a flip flop as well, what is the world coming to.
  • Gavin Bonshor - Monday, February 27, 2023 - link

    We test at JEDEC settings; that's just how we do it. Check ANY of our CPU review content going back multiple years.
  • eloyard - Friday, March 3, 2023 - link

    Isn't the point of testing missed if the testing is done in wildly suboptimal fashion that does not reflect real world configuration? On one hand you have overpowered cooling, PSU, MB and GPU, but on the other the real bottleneck for some core designs: RAM - is artificially kept slow? Either you run top of the line enthusiast spec or some basic discount shop bad config. The excuse "we've always been doing that" feels hollow. If you've been doing it wrong, you should correct that, not try to make that point of virtue. You've might as well tested all with 300W PSU and basic air cooler and claim the same, and in same vane such "CPU testing" wouldn't show what people are expected to see - how CPU will behave in a real life config.
  • cruiseliu - Monday, February 27, 2023 - link

    Theoretically EXPO and XMP are overclocking.
    Though most of us treat them as standards nowadays...
  • Ryan Smith - Monday, February 27, 2023 - link

    More specifically, they will invalidate your CPU warranty. That is our single largest issue with XMP/EXPO right now.

Log in

Don't have an account? Sign up now