Interview: AMD Ryzen as a workstation
Today, we got an interview with drverzal about a brand new AMD Ryzen rig!
Is AMD Ryzen a good CPU? What can you do with it? What should you expect from it? We will have in a separate post a comparison against i7–7700K to check the performance difference!
We are going to ask some questions to our new AMD Ryzen owner…!
What is your new rig specs?
- CPU: AMD Ryzen 1700
- GPU: NVIDIA 1080 Ti
- Motherboard: ASUS ROG Crosshair VI Hero
- RAM: G.SKILL TridentZ RGB Series 32GB (4 x 8GB) 288-Pin DDR4 SDRAM DDR4 2400
- CPU Cooling: NZXT Kraken X62
- Hard Drive: M.2 Samsung EVO 960
What are you aiming to do with your rig?
I wanted my rig to be well-rounded. It’s not a monstrous computing machine, but it’s sufficient for exploring small to medium sized models in a timely manner.
Why choosing AMD instead of Intel for this specific rig?
From all the research that I did, the price-to-performance ratio of the Ryzen CPUs was astounding in multithreaded applications. Thus far, my benchmarks seem to agree.
See here for the detailed benchmarks: Benchmarking xgboost with and without virtualization
What would be equivalent pricing using Intel-only CPUs?
Grabbing the two chips nearest my scores come in at $613 and $660 whereas the Ryzen 1700 came in a $330.
AMD Ryzen 1800X + 32GB RAM with 1Gbps network setups are available online for renting for $79.00/month online. Is it better to own your own desktop than renting?
I utilize my machine for much more than only machine learning. I produce/write music where I utilize Ableton as my DAW (Digital Audio Workstation) and I enjoy a video game or two. To me, it was a better decision to buy.
Advantages/Disadvantages of choosing AMD instead of Intel?
The biggest advantage is undoubtedly the price-to-performance, with a large caveat, for multithreaded applications!
Intel still takes the cake in single threaded performance.
What was your previous rig?
- AMD: FX 8350
- GPU: RX 480
- RAM: 8GB
What would you prefer using a desktop or a laptop for data science?
Personally, I’d go with a desktop. There’s nothing better to me than having a nice home office. That and having a multi-monitor setup is always enjoyable!
How does AMD CPUs compare against Intel CPUs in raw performance?
- Singlethread: 127
- Multithread: 1378
Comparison: we are hitting with multithread “only” 1000 with a i7–7700K.
Did you overclock? How far did you go? Was it easy?
That was easy, 4.0 GHz on 8 cores!
Comparison: i7–7700K can reach 5.0 GHz on 4 cores, but many reports online are showing dead CPUs.
How does AMD CPUs compare to your previous rig for raw machine learning speed?
To compare scores: Linus Tech Tips Cinebench R15 score list
Looking through the list of Cinebench scores from above, the highest overclock FX 8350 came in at #348 with a score of 842. The stock 1700 came in at 1378.
There are two added benefits I’m feeling so far:
- One, small models are nearly instant and allow me to work without interruption;
- Two, the reduction in training time for larger models mean I’m not waiting on results as long.
Basically, I get results faster and more importantly, I can learn faster.
Opinion on AMD CPUs = too fast too furious?
I’m loving the Ryzen chip so far. As you might be able to tell from above, I’ve been supporting AMD for quite some time now.
I’m a big fan of keeping competition in there to keep Intel honest. It makes it a lot easier to support Team Red when they’re putting out killer products like the Ryzen lineup.
What about RAM? Are you a multiprocessing or multithreading user?
I upped to 32GB of RAM for training larger models. I’m typically not building models on anything much larger than this at home.
ANY COMPUTER PICS? FLASHY COLORS? VROOM VROOM NOISE? OVERCLOCKING MANIAC? WATERCOOLING MAGICIAN?
I’m still waiting on the mounting bracket for my water cooler :*(