usertests vanadiel007 said: Let's see AI come up with a solution to this problem. The more you buy, the more TOPS you have. Reply
shady28 TBH from what I have seen the Ryzen 9 AI+ 395, when paired with a healthy dose of high performance memory, matches the DGX at roughly 1/2 the cost. The main problem for the Ryzen is software / API compatibility. While that problem is significant, it illustrates how overpriced the DGX is. Reply
SkyBill40 S58_is_the_goat said: I'm sure when they said 240w there was an asterisk and at the bottom it said 240w peak and 100w sustained. Yeah… the fine, fine print. Reply
bit_user S58_is_the_goat said: I'm sure when they said 240w there was an asterisk and at the bottom it said 240w peak and 100w sustained. The power figure is something people are latching onto as a potential cause, but the real issues are whether it can achieved the sustained compute rates that Nvidia claimed and: The article said: several threads on Nvidia’s developer forums now include reports of GPU crashes and unexpected shutdowns under sustained load. vanadiel007 said: Let's see AI come up with a solution to this problem. AI would probably design a custom waterblock for the Spark, to be used with an external radiator + pump. Reply
bit_user shady28 said: TBH from what I have seen the Ryzen 9 AI+ 395, when paired with a healthy dose of high performance memory, Nvidia said Spark was good for 1000 TOPS, which is a lot more than 126 TOPS that AMD quoted for combined performance of Strix Halo's CPU + GPU + NPU cores. TBH, I'm not sure if AMD's figure includes sparsity, but I think not. So, it's probably more like a quarter of Nvidia's figure than an 8th. Of course, that's on paper. What's achievable in the real world, on real models is another matter. And if one of the machines isn't even stable, then it doesn't really matter how fast it is. Reply
shady28 bit_user said: Nvidia said Spark was good for 1000 TOPS, which is a lot more than 126 TOPS that AMD quoted for combined performance of Strix Halo's CPU + GPU + NPU cores. TBH, I'm not sure if AMD's figure includes sparsity, but I think not. So, it's probably more like a quarter of Nvidia's figure than an 8th. Of course, that's on paper. What's achievable in the real world, on real models is another matter. And if one of the machines isn't even stable, then it doesn't really matter how fast it is. Perhaps in some areas that is worth something, but for what I have been looking into those metrics are pretty worthless. Specifically, this is for using local LLM AI for development. This can be somewhat of a big deal, since such AI costs $$ Apple is actually the best, but the cost of entry is also quite expensive, and the software support isn't there either for a serious developer. Another big advantage of the Spark is the ability to daisy chain them together via its built in 200 Gbit interface. But just the hardware, yeah, it's not really there. Except for that daisy chaining. For the cost of a decked out Mac Studio, you could buy two Sparks. At that point, the performance tables would probably turn. a/SOHozLp View: https://imgur.com/a/SOHozLp 82SyOtc9flA:5 View: https://www.youtube.com/watch?v=82SyOtc9flA&t=5s Reply
Key considerations
- Investor positioning can change fast
- Volatility remains possible near catalysts
- Macro rates and liquidity can dominate flows
Reference reading
- https://www.tomshardware.com/tech-industry/semiconductors/SPONSORED_LINK_URL
- https://www.tomshardware.com/tech-industry/semiconductors/users-question-dgx-spark-performance#main
- https://www.tomshardware.com
- NVIDIA Contributes to Open Frameworks for Next-Generation Robotics Development
- SanDisk Extreme Pro With USB4 (2TB) review: Bursty speed, but not great for pros
- NVIDIA Launches Open Models and Data to Accelerate AI Innovation Across Language, Biology and Robotics
- New Windows 11 feature aims to diagnose crashes — will check RAM after BSODs to look for problems
- Nvidia unveils Vera Rubin supercomputers for Los Alamos National Laboratory — announcement comes on heels of AMD's recent supercomputer wins
Informational only. No financial advice. Do your own research.