NVIDIA REFUSED To Send Us This - NVIDIA A100
0 up · 0 down · 0 ratings
Promos
We've experienced a lot of crazy, top-of-the-line graphics cards on LinusTechTips, but we've been unable to get our hands on one famed card - the NVIDIA A100..... until now. ;) Discuss on the forum: linustechtips.com Fan Duct 3D Print File: thingiverse.com Buy an NVIDIA A100: geni.us Buy an MSI RTX 3090 Suprim X: geni.us Purchases made through some store links may provide some compensation to Linus Media Group. ► GET MERCH: lttstore.com ► AFFILIATES, SPONSORS & REFERRALS: lmg.gg ► PODCAST GEAR: lmg.gg ► SUPPORT US ON FLOATPLANE: floatplane.com FOLLOW US ELSEWHERE --------------------------------------------------- Twitter: twitter.com Facebook: @LinusTech Instagram: @linustech TikTok: @linustech Twitch: twitch.tv MUSIC CREDIT --------------------------------------------------- Intro: Laszlo - Supernova Video Link: youtube.com iTunes Download Link: itunes.apple.com Artist Link: soundcloud.com Outro: Approaching Nirvana - Sugar High Video Link: youtube.com Listen on Spotify: spoti.fi Artist Link: youtube.com Intro animation by MBarek Abdelwassaa @mbarek_abdel Monitor And Keyboard by vadimmihalkevich / CC BY 4.0 geni.us Mechanical RGB Keyboard by BigBrotherECE / CC BY 4.0 geni.us Mouse Gamer free Model By Oscar Creativo / CC BY 4.0 geni.us CHAPTERS --------------------------------------------------- 0:00 Intro 1:06 How we got one, heh 1:58 Our contenders 2:38 A100 Specs 3:49 Teardown 8:38 Jake sucks at throwing 8:48 Buildup 9:43 Shroud 10:39 How to get it to boot 10:56 It works! 11:24 Blender 14:05 Can we trick windows into running games on it? & GPU-Z 15:27 Ethereum mining & afterburner options 16:45 Folding @Home 18:02 Resnet50 machine learning benchmark 18:59 Worlds most expensive lint roller 19:10 Resnet50 machine learning benchmark part 2 21:12 Closing thoughts
The video centers on Linus Tech Tips obtaining and examining an NVIDIA A100 GPU, a card Nvidia typically does not seed to reviewers due to its enterprise orientation and high price. The hosts compare the A100 to several other GPUs including the CMP 170 HX, RTX 3090, and a not-quite-official comparison with an A6000, highlighting the A100’s 40 GB memory, seven-nanometer GA100 GPU, and 250 W TDP. They document the unboxing, physical inspection, and teardown, noting the unusual single eight-pin power connector and the passive cooling approach, which necessitates external airflow to keep temperatures in check. The team goes through boot procedures, enabling 4G decoding, MMIO settings, and driver installation, then explores how the A100 behaves in Windows versus Linux environments, including GPU-Z, CUDA, OpenCL, and OpenGL support limitations. In performance tests, they run Blender classroom renders, CUDA vs OptiX comparisons, and a ResNet-50 benchmark under TensorFlow two on Ubuntu with Docker and Nvidia Docker, comparing power draw and efficiency against the RTX 3090. The results reveal striking efficiency and performance edges for the A100 in compute workloads, with substantially higher throughput per watt and, in some scenarios, parity or near-parity with consumer gaming GPUs on specialized tasks. The host reflections connect these findings to broader implications for Nvidia’s Ada Lovelace future and data-center GPU design, while acknowledging the practicalities of power, cooling, and total cost of ownership in enterprise deployments. Overall, the video blends hands-on hardware exploration with rigorous benchmarking to illuminate what makes the A100 tick and why it matters for AI and large-scale computation, even as it remains impractical for typical consumer use. Concluding remarks emphasize future Nvidia architectures and the ongoing tension between cutting-edge enterprise hardware and mainstream affordability, while also pointing viewers toward related content on the CMP 170 HX for further context.
Topics · hardware · gpu · ai · data-center · computing · energy-efficiency · tech-review · enterprise
Questions answered
- Waarom kreeg Linus toegang tot de NVIDIA A100 ondanks dat Nvidia geen revisie-samples stuurde?
- Linus kreeg de A100 doordat een tussenpersoon akkoord ging met een teruglevering plus een testopzet zoals Ethereum mining en warmteafzuiging, waardoor een tijdelijk uitleenmodel ontstond en de test kon plaatsvinden.
- Welke belangrijkste hardware-verschillen zitten er tussen de A100 en een RTX 3090?
- De A100 is een GA100 GPU gebouwd op zeven nanometer technologie met 40 GB memória en 250 W TDP, heeft geen standaard videoweergave outputs en gebruikt een afwijkende voeding en aansluiting, terwijl de RTX 3090 een consumentgerichte kaart is met videorendering- en DirectX-ondersteuning. OpenCL en CUDA zijn beschikbaar op beide, maar de functionaliteit en efficiëntie verschillen aanzienlijk in praktijkbenchmarks.
- Wat waren de grootste prestatiebevindingen van de A100 in deze video?
- In compute-workloads presteerde de A100 aanzienlijk efficiënter per watt dan de RTX 3090, met hoge omzettings- en doorvoersnelheden bij machine learning en bench-marking zoals ResNet-50, en aantoonbare voordelen in energieverbruik voor vergelijkbare taken.