Intel vs AMD for a new build in 2025

Hello everyone,

I’m planning a new PC build and am currently undecided about which CPU to choose.

I understand that AMD CPUs have faced some slower performance in MATLAB due to Intel MKL but I understand it was solved with the AMD AOCL? Bench instruction results also seem to show a performance difference between Intel and AMD CPUs (but couldn't check too many).

Intel CPUs seem to have some issues which probably have been solved for 14th gen with the new BIOS updates, and 15th gen is not quite mature. Their power dissipation is quite high.

For MATLAB, I plan to use the Parallel Computing and Optimization Toolboxes.

Thank you!

Answer 1

[Mike Croucher]

I am in the market for a new personal machine myself. I don't have any answers as to what is best but I can tell you that MATLAB now ships with both Intel MKL (Currently the default) and AMD AOCL. You can switch to AOCL as dicussed in this blog post Linear Algebra in MATLAB: Trying out AMD’s AOCL » The MATLAB Blog - MATLAB & Simulink

When I looked at it back in 2022, performance was mixed but I hope to have another look in the near future.

Anyway....If worrying about Intel MKL is holding you back, my point is 'your don't have to worry' since we now ship both.

Answer 2

[Umar]

Hi @Dan Hulea,

When selecting a CPU for your needs, particularly for MATLAB and demanding applications like ANSYS, both Intel and AMD offer compelling options. The AMD Ryzen 7 9800X3D and Ryzen 7 9950X are excellent choices, especially with the recent improvements in AMD's AOCL, which enhances performance in MATLAB. The Ryzen 7 9800X3D, priced at $474, provides a strong balance of performance and cost, making it suitable for parallel computing tasks.

On the Intel side, the i9 14900K is a powerful option, but its higher power dissipation and cost ($390) may not justify the performance gains for your specific use case. The Ultra 7 265K is more budget-friendly at $230, but it may not meet the demands of your applications.

Given your requirements for both MATLAB and gaming, the AMD Ryzen 7 9800X3D stands out as a well-rounded choice and I do concur with @Star Strider’s comments mentioned above, it does offer excellent performance without excessive power consumption. Pairing it with your RTX 3070 will ensure a robust system for both computational tasks and gaming.

Hope this helps.

[Austin Fite]

For any future readers, here is a benchmark of the 9800X3D. PC info:

• 9800x3d undervolted -30 all core with PBO auto (~5.3 GHz boost, ~90W under load)
• 64GB 6000MHz CL30 RAM
• 3080 Ti
• Windows 11, R2024b

I've been happy with its performance in MATLAB. I think the large cache and fast cores make it an excellent choice for everything but the 1% most demanding multithreaded workflows.

0.1951    0.1078    0.0794    0.3332    0.3059    0.0953

Answer 3

[Star Strider]

My preference would be a high-end Ryzen CPU, at least a 1TB SSD, and as much memory as you can manage. (I have the laptop version of that in one I bought a couple months ago.)

[Dan]

What Ryzen CPU do you have?

also, could you run the "bench" instruciton in Matlab?

For a mobile Intel I5-13450HX I get this result:

0.4636 0.2105 0.1083 0.3200 0.4589 0.6478

[Star Strider]

My desktop (this computer) is reasonably old, running a Ryzen 9 3900 in Wiin 11 with 32GB RAM and stock settings for the clock speed (not overclocked). I don’t remember when I bought it, however Tom’s Hardware has a review for it dated 12 November 2019, so it’s been a while (at least four years).

The MATLAB benchmark results for it —

I also ran this on my new laptop, and it compares reasonably well with other laptop CPUs (relative speed about 30) although not equivalent to a desktop processor.

Previoously (I don’t remember exactly when), the bench documenttation page had reference values for a number of different CPUs. It doesn’t currently.

.

[Dan]

Thank you!

Looks pretty good for the Ryzen 9 3900

[Star Strider]

My pleasure!

Tthe newer Ryzen processors have faster clock speeds and more cores, so they would be faster. Also consider the AMD Threadripper processors.

There are several threads on Answers that discuss processors. Use the search facility (the magnifying glass icon in the upper right corner of this page) and use the More option to the right of MATLAB FAQs (not the one in the top toolbar) so More -> Help for documentation on searching with it. Just searching on AMD would probably work.

I generally prefer AMD products, and have since the Athlon days in the early 2000s.

.

Answer 4

[Jeno Boka]

Matworks just posted that they are aware of the MATLAB stability issues on the 13th and 14th gen intel processors.

I myself am using an Intel i9-14900k, and I experience random crashes frequently. I don't recommend buying a 13th or 14th gen intel CPU!

Source:

https://www.mathworks.com/matlabcentral/answers/2174514-why-does-matlab-sporadically-crash-on-my-13th-or-14th-generation-intel-processor

Answer 5

[Walter Roberson]

The best Intel CPUs have individual cores that peak at higher performance than comparable AMD devices.

The best AMD CPUs have lower power disapation and more cores than Intel devices.

Thus, the best to select depends on your workload. If your workload is primarily things that are more or less single-threaded, then you want the Intel for its higher peak performance. If your workload is easily multi-threaded or benefits from having a large parpool, then you want the AMD.

Note: there tends to be a peak performance on multi-threaded work, at approximately the 4 to 8 core range. Beyond that, it is common for the overhead of distributing work to cores and getting results back outweights the benefits. But this does not apply to parfor situations in which the individual processes last relatively long.

Note: "performance" cores versus "efficiency" cores modify the analysis a bit. It may be overall more efficient to stay with performance cores, by reducing parpool size to the same number of performance cores. Otherwise if you have automatically threaded work out to all of the cores, the efficiency cores end up idling waiting for the efficiency cores to finish the batch of work.

[Walter Roberson]

Hmmm... I realized I don't know anything about automatic migration of work from efficiency cores to performance cores, or vice-versa.