Ingabe i-CMP 170HX ilungele i-training?

Ilungele kakhulu i-inference nemisebenzi encike kumemori. Nge-12.1 TFLOPS FP32 ihamba kancane kwi-training edinga amandla amakhulu okubala kunamakhadi afana ne-A40 noma i-RTX 3090, ngakho hlola umsebenzi wakho ngaphambi kokuhlela ukuwusebenzisa isikhathi eside.

Kungani kufanele ngihlole umsebenzi wami kuqala?

I-CMP 170HX yikhadi le-compute kuphela kanti ezinye izindlela zesoftware zisebenza ngendlela ehlukile kuyo kunakuma-GPU ajwayelekile. Iseshini emfushane yehora ikutshela ukuthi i-framework yakho, ama-driver kanye nemodeli kusebenza ngendlela elindelekile yini.

Yikuphi ukucaciswa okunamandla kakhulu kwe-CMP 170HX?

I-Memory bandwidth: ebalelwa ku-1516 GB/s, ephakeme kune-1313 GB/s ebhaliwe ye-A100 PCIE, okusiza noma yimuphi umsebenzi olinda imemori kakhulu kunokubala kwezibalo.

Kufanele ngigcine iseva ye-CMP 170HX isikhathi esingakanani ku-Cloudj?

Isikhathi osidingayo kuphela. Inkokhelo ibalwa ngehora ngenkathi iseva isasebenza, futhi ungayicisha noma uyisuse ngokushesha nje lapho isivivinyo noma umsebenzi uphela.

GPU Cloud · NVIDIA · Ampere

Qasha i-NVIDIA CMP 170HX GPU ene-64 GB VRAM

Renta amaseva e-cloud GPU e-CMP 170HX ngehora ku-Cloudj: i-64 GB VRAM ebhaliwe, i-Ampere architecture kanye ne-memory bandwidth ebalelwa ku-1516 GB/s yokubala okuncike kumemori (memory-bound compute). Leli yikhadi le-compute elingenasibonisi (headless) labathuthukisi abakhathalela kakhulu i-bandwidth nomthamo kunejubane le-FP32 (12.1 TFLOPS, 250 W, compute capability 8.0). Qala ngokwenza isivivinyo esifushane kuthempulethi ye-PyTorch noma ye-Ollama, bese ukhulisa umsebenzi wakho kalula.

Qasha i-CMP 170HX Wonke ama-GPU

Imininingwane ye-CMP 170HX

VRAM nge-GPU ngayinye

64 GB

Architecture

NVIDIA Ampere

Compute capability

Ukubala kwe-FP32

≈ 12.1 TFLOPS

I-bandwidth yememori

≈ 1516 GB/s

Umkhawulo wamandla

250 W

Amanani abikwa umhlinzeki wengqalasizinda yemishini ebalwe ohlwini njengamanje; ukusebenza kwangempela kuyehluka ngomshini ngamunye.

Izilungiselelo zamanje nezintengo ze-CMP 170HX

Isitoko esibukhoma nezintengo, kubuyekezwa ngokuqhubekayo. Intengo yokugcina iqinisekiswa ngaphambi kokuba u-ode.

Qasha manje

Ongakusebenzisa ku-cloud GPU ye-CMP 170HX

I-inference encike ku-bandwidth

Ukwenziwa kwama-token e-LLM kuncishiswa ikakhulukazi yi-memory bandwidth, kanti i-1516 GB/s isezingeni elifanayo namakhadi asesigabeni se-A100.

Ukusingatha ama-quantized LLM

Layisha amamodeli e-quantized ku-Ollama noma ku-llama.cpp bese uwasebenzisa nge-API yokuhlola noma ngamathuluzi angaphakathi.

Izivivinyo zama-CUDA kernel

Hlaziya ama-kernel adla imemori kakhulu ekhadini le-Ampere eline-compute capability 8.0 ngaphambi kokuwayisa kwi-hardware enkulu kakhulu.

Ukucubungula idatha ngeqoqo

Sakaza ama-array amakhulu ngememori ye-GPU ukwenzela izinguquko, i-hashing noma imisebenzi ye-analytics encike ekufinyeleleni kwimemori.

Ingabe i-CMP 170HX iyi-GPU efanele umsebenzi wakho?

Khetha i-CMP 170HX uma umsebenzi wakho ubambeka ngenxa yomthamo wememori kanye ne-bandwidth kunokubala okulula okusezingeni elilodwa (raw single-precision compute). Khetha i-GPU enokubala okunamandla kakhulu uma udinga i-training esheshayo, noma ukhethe ikhadi labathengi elincane uma amamodeli akho engadingi i-64 GB ye-VRAM.

CMP 170HX uma iqhathaniswa nama-cloud GPU afanayo

GPU	VRAM nge-GPU ngayinye	Ukubala kwe-FP32	Umkhawulo wamandla
NVIDIA CMP 170HX	64 GB	≈ 12.1 TFLOPS	250 W
NVIDIA A800 PCIE	80 GB	≈ 15.6 TFLOPS	300 W	Imininingwane →
NVIDIA H100 PCIE	80 GB	≈ 41 TFLOPS	350 W	Imininingwane →
NVIDIA H100 SXM	80 GB	≈ 53.5 TFLOPS	700 W	Imininingwane →
NVIDIA A40	45 GB	≈ 29.9 TFLOPS	300 W	Imininingwane →

I-cloud GPU ye-CMP 170HX: imibuzo evame ukubuzwa

Qasha i-cloud GPU ye-CMP 170HX ngehora

Khetha ithempulethi, khetha izilungiselelo zamanje bese uxhuma nge-SSH noma itheminali yesiphequluli. Ukukhokhisa kuyama uma ususa i-GPU.

Qasha i-CMP 170HX Wonke ama-GPU