NVIDIA Datacentre GPUs explained, every model in 2024 covered

No video

NVIDIA Datacentre GPUs explained, every model in 2024 covered

Рет қаралды 1,656

Күн бұрын

NVIDIA produces a wide range of datacentre GPUs in 2024, each optimised for specific workloads such as AI model training, inferencing, HPC and rendering to name just a few. We take a look through the range of Hopper GPUs, Ada Lovelace GPUs and Ampere GPUs, and explain how to pick the perfect NVIDIA GPU for your requirements.
Configure your perfect server with NVIDIA datacentre GPUs from Scan 3XS Systems www.scan.co.uk...
Find out more here www.scan.co.uk... - or get in touch with the Scan IT team here - business@scan.co.uk / 01204 474747
Follow us on social media here:
LinkedIn - / scan-computers-interna...
Facebook - / scanbusiness
Twitter - / scan_business
View our other Scan Business channels:
AI - / @scanaisolutions
Cloud - / @scancloud2500
Pro Graphics - / @scanprographics5094
Pro Video - / scanprovideotv
Digital Signage - / @scandigitalsignageana...

Пікірлер: 11

@jaffarbh 6 ай бұрын

As informative as always. I think it's worth considering whether the GPUs are required for training (large batch sizes) or inferencing (small batch sizes). Nvidia's big boys are effective for training but too costly and inefficient for inferencing. Apart from large tech companies, most users are likely to deploy pre-trained models (and may be a bit of finetuning on small datasets) and therefore shouldn't overspend on the most expensive hardware (and lengthy waiting time).

@scanitsolutions 6 ай бұрын

Thanks for the thoughtful comment. These less demanding inferencing workloads are one of the major reasons why NVIDIA has the L4 and A2 GPUs. We'll try and make that point clearer in future videos. - JG

@io9021 6 ай бұрын

Thanks for the video. I work mainly with A100 and A10 for ML workloads, but I didn't understand what the purpose of the L series is. Thanks for shedding light on this 💡 I'm wondering why aws still offers K and M series instances at fairly high prices. Are there legacy systems that can't switch to modern hardware?

@scanitsolutions 6 ай бұрын

Glad to hear you found it useful. Thanks for the feedback. K and M GPUs are indeed very much legacy models, from 2012 and 2015 respectively, so will be pretty hopeless for ML/AI workloads. We can't really say why AWS is charging so much for instances like this, but all cloud providers are bound to have some older GPUs kicking around. Scan will be launching a new and improved cloud service in a few weeks, watch this space for the announcement. - JG

@stananderson83 4 ай бұрын

Thanks a lot for your video. One question, can L4 be used for basic rendering tasks? On one of your tables this feature is crossed out for this particular GPU

@scanitsolutions 4 ай бұрын

Hi, there's nothing stopping you from using a L4 for rendering, it just isn't fast at it, so we wouldn't recommend buying one for that task. However, if you've got an L4 already there's no harm in sending some rendering it's way... - JG

@SosoAndMe 6 ай бұрын

Can they run crysis?

@scanitsolutions 6 ай бұрын

Yes, they can even run Crysis in 4K - JG

@idkxxx7532 6 ай бұрын

Can this run a Minecraft server?

@scanitsolutions 6 ай бұрын

That's more CPU-limited, than GPU-limited, but you could probably run dozens of Minecraft servers on the sorts of servers these GPUs are normally installed in. - JG