853.77 Cats Per Second

Fresh Setup

Overview

Purpose for the build is to work with CUDA for data science seperate from my gaming machine in compact build as I have limited desk space.

Parts List

PCPartPicker Part List

Hardware Choices

• I had $2k pre-tax/s&h budget in mind. Initially opting for 3080 due to VRAM but budget and availability had me step down to a 3070 Ti. I was sucessful at this goal
• Why not 3070 non-Ti? Availability and marginal difference in “real” price had me opt for this Ti version. GDDR6X upgrade is worth it for me. Most reviews only did gaming related benchmarks but I would like my use case tested as well
• I am big fan of EVGA cooler design and it shows in the temps I see. IMO EVGA has the best warranty and support in the industry
• Asrock Mobo is basic but functional and ticks all the boxes for me for this build (from size to chipset RAM OC capabilities to headers to vrm to back I/O)
• RAM kit is great for price to perf ratio and overclock ability and is minimum size for my use case
• NVMe might not be what it was before WD nerfed the write speeds but it is still better than SATA speeds for not much of a price difference
• PSU is well reviewed and by reputable manufacturer and satisfies the hungry GPU and it fits my budget

OS

Running Pop_OS! 20.04 (nvidia based installer). Why? It just works (most of the time)

Build Notes

• Booted right up on the first try!
• Overall impressed with this case, they thought about all the little things like push-fit side panels, PWM case fans, bracket to keep front panel headers together and modularity for the PSU bracket that make building a joy! The price point is good as well
• Had no issues fitting the tall deepcool cooler in this compact case, no clearance issues with the RAM. Mounting mechanism wasn’t difficult to work with
• PSU cables were the perfect size for this build and made cable management easy

Issues

• Couldn’t fit to top 25mm thick case fan with this Asrock motherboard due to the position of the CPU socket. It was really millimeters away from fitting right in (Yes I tried removing the fan grills). I am confident that cooling won’t be an issue. Not a deal breaker as I knew this going into the build. The ITX motherboard was the right price. April 2022 Edit: I was clearning the case and noticed that the PSU can be mounted a bit lower in the case and I was able to fit the top fan above the PSU
• Slight GPU sag and it took a bit of manoeuvring to get the GPU in the case but it is again not really a deal breaker

Temps

• I didn’t see temps go above 65C (in Room Temprature) for both GPU and CPU in my limited testing
• Idle temps were 30C for CPU and 39C for GPU

TensorFlow resnet50 benchmark 128 batch size with FP16: 853.77 Cats Per Second!!!

Build Pics