Happy Halloween! Like Dr. Frankenstein, our manager Doug has been toiling away at testing, validating, and testing again. The monster has finally come to life! (With the RGB colors set to orange for the upcoming holiday…)
What this build has to offer is the highest price-to-performance AI computing. For only slightly more than the cost of a single NVIDIA RTX PRO 6000 Blackwell, this system features:
16 CPU cores, all of which are Performance-oriented, with the AMD Ryzen™ 9 9950X
256GB of Crucial Pro DDR5 RAM, in the form of four 64GB modules
8TB of NVMe SSD storage, in the form of two 4TB NVMe SSDs configured in RAID0 for maximum throughput
DUAL NVIDIA® GeForce RTX™ 5090 graphics cards, featuring 64GB GDDR7 VRAM total and 6704 total AI TOPS (compare against a single RTX PRO 6000 Blackwell with 4000 AI TOPS)
Arctic Liquid Freezer III cooler for ideal CPU cooling, Fractal Design case with extreme airflow - the highest observed temperature in sustained AI compute on the RTX 5090 GPUs was approximately 82C on one card, and 77C on the other
In the ArsenalPC lab, we wanted to jump into the AI workstation market with both feet first, disrupting the entry-level with more compute than you can get from any other system in this price range! We have spent several months testing and verifying the maximum performance we can extract from this traditional Socket AM5 system, with amazing results!
Traditionally, our competitors are speccing out systems for AI compute using AMD’s new Ryzen™ Threadripper™ PRO 9000WX CPUs; at ArsenalPC, we’re finalizing these configurations as we speak and will be offering systems with up to four RTX 6000 Blackwell Max-Q GPUs, up to 14044 AI TOPS, up to 1TB of DDR5 RAM… stay tuned!
The MES2X gives users a solid machine for generative AI at home for a relatively reasonable price. A major step up from any home PC, and one that’s more than capable of gaming on the weekends! If you have questions about this system’s capabilities… comment down below!
Hey there Zac! While we don’t discuss our internal costs on components, you can check the pricing we charge for these beastly PCs by clicking through one of the links in the article.
Liquid cooling is completely unnecessary for this type of machine. The highest GPU temperature we observed during simultaneous load on both GPUs with OCCT… was only 82C. NVIDIA specifies the max GPU temp for an RTX 5090 as 95C, so there’s plenty of headroom. Liquid cooling would add well over a thousand dollars to the price between the blocks, tubing, fittings, pump, reservoir, distro block, and additional labor. Not to mention, a custom cooling solution would require the end-user to fill the system upon delivery, as there’s no good way to ship a filled open loop without significant risk - which we just aren’t willing to take on RTX 5090 cards. Adding custom cooling would also completely defeat the purpose of trying to offer the most price-competitive entry to AI that’s a significant leap over a system with a single RTX 5090, which is the entire point of this lineup of systems.
Memory temperatures were well within spec, partly due to the adept design of the case airflow. With the amount of air moving through the Fractal case, the RAM doesn’t have a chance to heat up even under sustained workloads.
Understood Looks like it instantly sold out here. What’s you’re opinion of the card so far?
BTW Just out of curiosity why only two front fans instead of three? It seems like both GPU’s would benefit from the direct airflow or moving one of front fans down a space?
We have no opinion of the Radeon AI Pro R9700 32GB yet as we haven’t been able to get one for testing. We’re interested to know whether it will top NVIDIA options in its projected $1200 price range.
The two front 140mm fans is how this Fractal case comes supplied. We found adding a third fan that jams air up against the back of the PSU accomplishes nothing.
I had asked a similar question. Like, i have to assume its on end of $8-10k.
I wonder if the design decision was based on Power Consumers who are interested or not cause at some point, I would have thought if a server board would given more memory for the AI.
It’s over 10k $10,260. I didn’t them list the board or see a clear pic of it. I would imagine they used a X870 or X870E chipset to help handle them beastly cards & CPU 256GB is the max for X870(E) 4 Slot boards if you want more you have to go to a server board which could have 8
