Aceitamos criptomoeda — para que as moedas que detém possam comprar material informático real.Pague em cripto, gaste-a em material informático real. Envio seguro gratuito para todo o mundo acima de 399 $. Caixas neutras, sem marca, expedidas a partir da Alemanha.Envio seguro gratuito acima de 399 $. A sua taxa fica bloqueada durante 30 minutos assim que abre a finalização da compra.Taxa bloqueada 30 minutos. Cada artigo é selado, o seu número de série é verificado, e está coberto por uma garantia de 24 meses.Selado, garantido 24 meses.

Artigos guardados Iniciar sessão
Full-length passive GPU card with silver heatsink and single 8-pin power connector

NVIDIA

NVIDIA Tesla M40 24GB GDDR5 graphics card

SKU CH-4M8S-2AN8R

A 24 GB GDDR5 GPU with 3072 CUDA cores and 288 Gb/s memory bandwidth for machine-learning workloads

  • Memory TypeGDDR5
  • 24 GB GDDR5 memory enables large-model training
  • 3072 CUDA cores accelerate parallel compute workloads
  • 288 Gb/s bandwidth feeds data to the GPU fast
  • Passive 250 W thermal design suits dense server racks
  • PCI Express 3.0 x16 interface fits standard accelerator slots

NVIDIA Tesla M40 24GB GPU

The NVIDIA Tesla M40 is a 24GB GDDR5 accelerator built for machine-learning workloads. It delivers 3072 CUDA cores and a 384-bit memory interface running at 6GHz. The card connects through a PCI Express 3.0 x16 slot and has a thermal design power of 250W.

Passive cooling and power checks

This model uses a passive heatsink so your chassis must supply strong directed airflow across the board. Verify that the power supply can provide a dedicated 8-pin CPU connector and sustain the 250W board draw. The double-width bracket occupies two full-height expansion slots in the server.

Sustained 7 TFLOPS single precision

Under continuous load the GPU maintains its ~1140MHz boost clock to deliver 7 TFLOPS of FP32 throughput. Double-precision performance runs at 0.21 TFLOPS with a 1/32 ratio. The 288 Gb/s memory bandwidth and 24GB frame buffer keep large datasets resident during long training runs.

Pontos fortes

  • 24 GB GDDR5 memory enables large-model training
  • 3072 CUDA cores accelerate parallel compute workloads
  • 288 Gb/s bandwidth feeds data to the GPU fast
  • Passive 250 W thermal design suits dense server racks
  • PCI Express 3.0 x16 interface fits standard accelerator slots

Especificações

Memory TypeGDDR5

Perguntas sobre este artigo

Does the Tesla M40 24GB card ship with any cables or brackets, or must I buy them separately?

The card is supplied as shown in the picture; no cables or mounting brackets are included, so you will need to source an 8-pin CPU power cable and a compatible bracket for your chassis.

Should I choose this 24GB model or a smaller capacity for machine-learning workloads?

The 24GB GDDR5 frame buffer and 3072 CUDA cores give the headroom needed for larger models and batches that would not fit on a smaller card.

What does the 7 TFLOPS single-precision rating mean when the card is running in a server?

At ~1140 MHz boost clock the GM200 GPU can deliver up to 7 TFLOPS FP32 throughput, limited by the 250 W TDP and passive cooling in your airflow.

Também nesta secção

Os clientes também compararam

Mesma prateleira, mesma finalização da compra — oito moedas e uma taxa bloqueada durante 30 minutos.