The 3970X offers better performance under thermal constraints compared to the 3960X.
The 3970X offers better performance under thermal constraints compared to the 3960X.
Hi everyone I am currently on an i7-4770 based build and I have been steadily pushing my system to the max these days especially given the age. I have an i7-2600 based old machine still around, and the last one year I have been using this one also as a secondary build machine to give me some headroom but my workflows have become complex for my needs managing two machines. Hence considering a new build and I am looking at Ryzen 3rd gen Threadripper, specifically one of 3960X vs 3970X. I am planning to purchase the parts over the next 3-4 months slowly based on deals that show up. I am going to be using air cooling (NH-U14S TR4 SP3) for my build. I am primarily looking to make an informed decision between 3960X vs 3970X for my use case (described below). 3970X has more cores/threads but a slightly lower base clock. I am trying to better understand if 3960X allows me to push the CPU much closer to its max as opposed to 3970X, before I get capped by the CPU/cooling TDP limits? Or is it the other way around. where 3970X is more efficient in allowing more parallelization for the same CPU/cooling TDP limits, thoughts? Basically, I don't want to get a 3970X if I am going to get the same CPU performance as I would have with the 3960X because of the Thermal limits, and this is where I would like inputs primarily. I also understand cooling is going to make a whole lot of difference. I researched cooling options, and I prefer sticking with air cooling approach that I am more comfortable with. AIOs would be acceptable, but I have found none that has a large enough baseplate that provides 3rd gen Threadripper coverage and have proven to be be reliable. Just to give an overview of my current workloads and future extensions: - My workstation is turned on and running 24x7, and running primarily Debian Linux. - It is my primary machine for code development and has development/continuous-integration builds running a good portion of the time (4-6 hours a day minimum). A lot of these are clean builds, and hence the time to finish turns out to be CPU and parallelism dependent. - A few different docker containers running for development, testing, CI, etc. and a few Virtual machines as well for deployment. A few times a month, I also use Windows on VM to run few programs that I can run only on Windows. - A decent amount of IO, but I am already using a NVMe SSD which I will be carrying over to this new build. The numbers I have pulled so far seems to indicate that I am not bottlenecked on the IO with the SSD. - I also use 4 hard drives in mdadm RAID-6 for cold storage. Again not bottlenecked on IO here as well so far since most of the heavily used data remains on the SSD, with only backup/long-term storage moving to this cold storage and also less frequently compared to the reads from the SSD. However I plan to be vastly (4 - 8x) scaling/expanding my development scenarios, and based on the calculations I did, I am most likely to start hitting bottlenecks here. This is where I am thinking that the TRX40 platform and available PCIe lanes might help. I might throw in one or two 4xM.2 PCIe adapter cards to add more NVMe storage. If that is overkill, I might take the route of adding few SATA SSDs to keep the costs down and extend my runway instead. - No gaming per-se on this machine, however I do need three 4k outputs. I already have a Radeon Pro WX 2100 GPU (comparable to an RX550) that is very much up to this task today and is quite energy efficient (Max TDP of 35W). - I also run Plex on the same server that streams/transcodes 4k content a few hours each day within my local home network while running the above workloads. - I am also planning to throw in additional new workloads like 4k/8k video editing (while my PC is handling few CI runs) for some of my hobby projects, but not currently. This is also one of the other reasons I am looking at high core count CPUs. - I also tend to keep these machines for 7-10 years, before going for the next upgrade. Thanks for the help.
Using more cores at a lower frequency tends to yield better efficiency and higher throughput. For tasks with fewer threads, the 3970X can match or exceed the performance of the 3960X. The concept of "max" might be unclear, but both will eventually face some limitations—either through forced overclocking or physical constraints like power and heat. These issues can be managed by adjusting clock settings manually. In any case, the 3970X should deliver a higher overall throughput.
Because you're handling many varied tasks that scale well across threads, I recommend the 3970x. The noctua cooler will help maintain stable performance, delivering the same single-thread speed as the 3960x while boosting turbo clocks.
Thanks for your feedback. I appreciate the clarification on throughput and scaling. Yes, I plan to use the cores/threads because the tasks will naturally run in parallel. @svmlegacy I'm sorry for any confusion earlier. A hypothetical scenario might help compare the two CPUs more clearly. If the workloads can maintain 64 threads active for a short period, would the 3970X more frequently reach thermal limits and experience back-off compared to the 3960X, which seems to stay within limits due to its lower thread count? In other words, do you think the shared thermal constraints are the main factor, or does having more cores give an advantage even if limits are similar? I'm trying to better grasp how Ryzen processors with many cores perform under comparable thermal conditions.
The CPU's temperature limits largely depend on the cooler, not the processor itself. Both the 3960X and 3970X share the same TDP of 280 W, yet they both experience a similar rise in temperature. Anandtech reported a maximum ΔT of 38.2°C for the Threadripper 1950X (180 W TDP). https://www.anandtech.com/show/12454/ana...ing-wins/5 For this chip, they also observed it reaching close to its TDP: 180 W. https://www.anandtech.com/show/11697/the...-review/19 Adjusting for a 280 W CPU under steady conditions, assuming a constant thermal resistance, we can scale the temperature change accordingly. Using the ratio 280/180, we find ΔT ≈ 59.4°C. This matches findings from adoredtv’s tests with the Noctua NH-U14s and NH-U12s: https://adoredtv.com/testing-noctuas-nh-...r-on-air/2 For the 3970X, testing showed temperatures around 85°C to 95°C in typical cases. Turbo performance on these chips is dynamic rather than fixed. Reports indicate both CPUs throttle between 80°C, which isn’t a complete shutdown but noticeable slowdown. Neither will completely stall; they’ll manage by reducing output. Regardless of the chip chosen, both will operate within the limits set by their thermal design. Reflecting on Anandtech’s analysis of these models, the 3970X still edges out the 3960X overall, even at full power draw of 24T. This is due to differences in core efficiency rather than just clock speed. Lowering voltage has a big impact because it drops exponentially, while increasing clocks only adds linearly.