**Nội dung tóm tắt:** Groq/Cerebras alcançon tốc độ推理 ấn tượng nhưng производства ASIC mウンド jSubview Jensen formation không tin كثير. GPUs vẫn.version supotimes cho workloads AI do sacrifice linh hoạt. Tags: #AI #ASIC #GPU #Groq #Cerebras #TechNews #óng #m recession
(498 ký tự)

https://www.reddit.com/r/LocalLLaMA/comments/1o9z1jc/why_does_jensen_keep_telling_asics_arent_worth_it/

🎁 Dev Box DPU Upgrade 🎁

More ZFS PoC for vGPU + DOCA blah blah NVMe-oF offloads

Left–to–Right (lotta slots)
- Broadcom CNA 2x10G
- Optane P3608 2x 2TB NVMe AIC
- Nvidia T10 16GB GPU (w/ 90mm blower)
- blank
- Nvidia T10 16GB GPU (w/ 90mm blower)
- Nvidia Bluefield2 DPU 16GB RAM, 2x 25G
- Optane 4x 512GB NVDIMM
- Samsung 4x 32GB ECC Volatile
- Intel Xeon 8370C (Azure) 32C/64T 270W

Chassis cooled by 2x 180mm, 1x 140mm, 1x 120mm. Idles very comfortably below 40dB, wind-tunnel at full rpm is more like whoooooshing than the usual 28K-RPM 40mm axial fan screamers people expect from enterprise hardware. Choose wisely.

#gpu #dpu #ai #dev #llm #nvidia #intel #optane #zfs #linux #freebsd

Fast calculation of the distance to cubic Bezier curves on the GPU

**#TechTips: How to Read #ComputerPart Numbers for your #DIY #PC build**

Randomly up in the middle of the night thinking about this, so here goes...

#Computer parts these days are numbered in pretty much the same scheme, whether it's #AMD, #Intel, or #Nvidia, and whether its a #CPU, #GPU, or #Motherboard. Note that I'm only including Intel and Nvidia here because they are the same numbering: Always Buy AMD.

The first number is always the generation ("series"), the second number is the offering within that generation, and there might be 1 or 2 letters at the end if they did something special with it.

Example...
The Ryzen 7600X I just got is a 7th-generation Ryzen, while the intel 7700HQ I replaced was also 7th generation, but Ryzen is newer than core, this the CPUs were made about 5 years apart. Stick with AMD chips over 7000, because those use the new AM5 socket, and thus will be upgradable to CPUs AMD has yet to release. (intel changes the socket every generation, and can't be upgraded)

(Continued)

[Gamers Nexus] The problem with GPU benchmarks | Reality vs. numbers, animation error methodology white paper

[Gamers Nexus] The problem with GPU benchmarks | Reality vs.... #hardware #gamersnexus #gpu #graphics

https://fedia.io/m/hardware@lemmy.world/t/2875225

[Gamers Nexus] The problem with GPU benchmarks | Reality vs. numbers, animation error methodology white paper - Hardware - Fedia

> Benchmarking has long had a problem of ensuring numbers relate back to the reality of what players feel when gaming. The issue mostly comes to explaining precisely why stutters and hitching are happening in games, not just that they exist. For the most part, "frametime" testing (or more accurately, frame-to-frame interval...

"vLLM 0.11.0 với NVIDIA RTX Pro 6000 Blackwell (96GB VRAM) chỉnh BinghamTON F1/F16: Output 500-2k tokens/sг快. Lợi ích nổi bật: Speed 1,000+ tokens/s cho 2-5 user (32K context), latency chỉкого. 20 users vẫn ổn định ở 521 tokens/s (128K context). #vLLM #RTXPro6000 #AI #NVIDIA #GPU #BENCHMARK #Tecnology"

https://www.reddit.com/r/LocalLLaMA/comments/1o96gtu/vllm_performance_benchmark_openai_gptoss20b_on/