Analyzing performance: First-gen Threadripper versus 9th-gen Intel processors.
Analyzing performance: First-gen Threadripper versus 9th-gen Intel processors.
I'm just beginning to explore deep learning and TensorFlow, but my setup is... outdated. It's a 2012 Mac Mini with a Samsung EVO SSD and 16GB RAM. I'm far from where I want to be. After browsing Dell Outlet deals, I'm really thinking about rebuilding my own machine (my last build was in 2008 with an Intel Skulltrail) and I'm drawn to the AMD Threadripper. My budget is around $1000, so I can't afford the latest v2 or v3 models. The dilemma is whether I should build a system that can be upgraded later or stick with an i5/i7 now and wait for better options. More threads mean more processing power, but the base Threadripper lets me expand quite a bit (I don’t think I could reach the same level with four GPUs). What works best is finding a solid foundation to start a powerful deep learning or heavy computation rig while keeping costs reasonable. Should I choose Intel i5/i7 or try an older AMD Threadripper version?
Consider the Threadripper 1950x with 4x4/4x8 GB RAM plus a MoBo combo for roughly $600–700. However, I suggest opting for a Ryzen 3800x or 3900x processor paired with 2x8 GB of high-speed RAM (such as 3200 or even 3600 MHz). This setup offers greater upgrade flexibility compared to the Threadripper models and Intel options. You’ll also have the option for an 1660 Ti or RTX 2060 SUPER, which is ideal for deep learning and TensorFlow.
I really appreciated beginning with the Threadripper because it offers significantly more bandwidth from the GPU/memory in the future (meaning more lanes to add cores). On the other hand, I’m limited by the Ryzen line. My upgrade worries mainly involve the graphics card rather than the CPU. I just need a solid foundation.
What matters most are the core specifications, memory capacity, and CPU performance. If you need a lot of RAM, consider affordable DDR4 ECC options available online. For a deep learning setup, prioritize higher core counts and better IPC rather than top speeds. A used build around $1000 could work well if you're looking for cost savings, though it may lack the necessary threads or cores for optimal deep learning performance.
The issue with first and second generation Threadripper is that older MoBos on X399 TR4 connectors can't handle newer Threadripper models built for TRX40 chipset and sTRX4 sockets—same pin layout but completely different pin arrangements. The reverse is true. If you're looking to boost GPU performance, the 1950X could be a good option, particularly with Titan Vs or Turing cards featuring tensor cores.
Good decision on the Xeon/DDR ECC RAM. The key factor is likely the number of GPUs. That's why I chose the Threadripper with 64 lanes. An other option would be a completely different path with fewer lanes and Ryzen (16 lanes). I assembled: $160 Threadripper 1900x $280 Gigabyte X399 $140 32GB DDR4 3000
I picked up a Ryzen 3900x. It doesn’t offer enough lanes to drain my budget on graphics cards, but it’s better for the future than a Threadripper from the first or second generation. Now I’m supporting the red team!
I'm in a similar boat, with an R5 3600 and debating a 3950x or higher. One thing that isn't mentioned enough with Ryzen is the Intel MKL library crippling on non Intel. There are workarounds, but python and other commonly used ML/DL frameworks utilize the MKL by default. Performance is hugely crippled. 2-3 orders of magnitude in speed crippled vs the Intel chips, especially AVX. AVX-512 is even more of a difference. Good news is, at the hobbyist level doesn't matter and can be worked around. At the pro/getting paid level, it's one of the very few instances where a 10980xe or the other x299 10xxx chips are warranted-- out of the box avx (and, avx-512) support without any workarounds needed. Do some google on the MKL with respect to AMD, employ the workaround, and enjoy! Also, yea, GPU training is a thing but CPU running of trials before pushing to GPU is also important. As is running and evaluating the model after training.
It's good to hear about the details of this box. From the beginning, there are many aspects that could be exploited.