Jizz_Beard_thePirate said: AMD launches Instinct MI300X and MI300A CDNA3 accelerators https://videocardz.com/press-release/amd-launches-instinct-mi300x-and-mi300a-cdna3-accelerators Certainly a lot of claims by AMD but once you read the finer details, you do see some interesting quirks. The jist of it stems from two key areas, one of which is that AMD did not use TensorRT LLM with H100 which means it's not performing to it's full potential and the other is that using their benchmarks, it largely sounds like it will be 20% faster than H100 until H100 hits a vram limitation at which point MI300X will be much faster. Keep in mind that H100s released back in March and H200s will be releasing Q2 next year. It is interesting because despite MI300X having 30% higher theoretical compute performance, we aren't really seeing the results which shows that AMD has a long ways away to optimize their software stack in the data center. You can read more about it here: https://www.semianalysis.com/p/amd-mi300-performance-faster-than With all that being said, I think AMD is still very much ahead of Intel in the Ai race which in itself is an achievement AMD announces Ryzen 8045HS, 8040HS and 8040U “Hawk Point” series powered by Zen4, RDNA3 and XDNA AMD next-gen “Strix Point” series with XDNA2 NPU to offer 3x performance boost for generative AI I wonder if having hardware whether it be a CPU or GPU will be required as a minimum when it comes to installing Windows 12 Intel 144 Core Sierra Forest Xeon CPU Leaks Out: Powered By Sierra Glen E-Core Architecture, 172 MB Cache https://wccftech.com/intel-144-core-sierra-forest-xeon-cpu-leak-e-core-architecture-172-mb-cache/ ARM-Based Chips From Qualcomm & Apple To Give AMD & Intel Tough Time In AI PC Segment In 2024 https://wccftech.com/arm-chips-qualcomm-apple-give-amd-intel-tough-time-in-ai-pc-segment-2024/
|
The Core Truth marketing is dumb because Intel, sans the 14700, just launched an entire lineup of rebadged processors. And their lower end CPUs like the 13100 last gen was refresh of the 12100. It's serious pot meet kettle situation. With them contradicting themselves throughout and basically just explaining to consumers how to read AMD's part numbers better than AMD.