NVIDIA CMP 70HX vs NVIDIA Tesla T4

Comparison of NVIDIA CMP 70HX with 8 GB GDDR6X and 3,840 cores vs NVIDIA Tesla T4 with 16 GB GDDR6 and 2,560 cores.

Loading...

Performance Rating

NVIDIA CMP 70HX outperforms NVIDIA Tesla T4 by 19.1% in the overall GPU ARK performance rating

A100 A100
H200 H200
MI325X MI325X

NVIDIA CMP 70HX

7.7

NVIDIA CMP 70HX

7.7
RX 7900 XTX RX 7900 XTX
MI250 MI250
Instinct MI300X Instinct MI300X

NVIDIA Tesla T4

6.4

NVIDIA Tesla T4

6.4

Expert Comparison

NVIDIA CMP 70HX лучше по многим параметрам: имеет больше ядер (3840 против 2560), большую пропускную способность памяти (608.3 GB/s против 320.0 GB/s) и более высокий GPI (7.7 против 6.4). Она также предназначена для более современных задач, особенно в облачных вычислениях и AI. NVIDIA Tesla T4 подходит для задач машинного обучения и глубокого обучения благодаря своей памяти объемом 16 ГБ, но уступает по другим техническим характеристикам.

Contents:

Memory ML Performance Compute Power Architecture & Compatibility ML Software Support Clocks & Performance Power Consumption Rendering Benchmarks Additional

Memory

Memory Size

🔥 8 ГБ
16 ГБ

Memory Type

GDDR6X GDDR6

Memory Bandwidth

608.3 GB/s 320.0 GB/s

Memory Bus Width

256 бит 256 бит

ML Performance

FP16 (Half Precision)

🔥 10.71 TFLOPS
65.13 TFLOPS

BF16 (Brain Float)

No No

TF32 (TensorFloat)

No No

Compute Power

FP32 (Single Precision)

🔥 +32% 10.71 TFLOPS
8.141 TFLOPS

FP64 (Double Precision)

🔥 0.1674 TFLOPS
0.2544 TFLOPS

CUDA Cores

🔥 +50% 3,840
2,560

RT Cores

🔥 30
40

Architecture & Compatibility

GPU Architecture

Ampere Turing

SM (Streaming Multiprocessor)

🔥 30
40

PCIe Version

PCIe 1.0 x4 PCIe 3.0 x16

ML Software Support

CUDA Version

🔥 8.6
7.5

Clocks & Performance

Base Clock

🔥 +133% 1,365
585

Boost Clock

🔥 1,395
1,590

Memory Clock

🔥 1,188
1,250

Power Consumption

TDP/TGP

unknown 70 W

Recommended PSU

🔥 -20% 200 W
250 W

Power Connector

1x 12-pin None

Rendering

Texture Units (TMU)

🔥 120
160

ROP

🔥 30
40

L2 Cache

🔥 4 MB
4 MB

Benchmarks

LLM Perf, Llama-2-7b-hf (bnb)

8.55 tokens/s

LLM Perf, Llama-2-7b-hf (gptq)

33.6 tokens/s

LLM Perf, Llama-2-7b-hf (unquantized)

16.7 tokens/s

LLM Perf, Llama-3.1-8B-Instruct (awq)

22.0 tokens/s

LLM Perf, Llama-3.1-8B-Instruct (bnb)

7.68 tokens/s

LLM Perf, Llama-3.1-8B-Instruct (gptq)

28.8 tokens/s

LLM Perf, Meta-Llama-3-8B (awq)

22.4 tokens/s

LLM Perf, Meta-Llama-3-8B (bnb)

7.96 tokens/s

LLM Perf, Meta-Llama-3-8B (gptq)

28.8 tokens/s

LLM Perf, Meta-Llama-3-8B-Instruct (awq)

22.3 tokens/s

LLM Perf, Meta-Llama-3-8B-Instruct (bnb)

7.76 tokens/s

LLM Perf, Meta-Llama-3-8B-Instruct (gptq)

29.1 tokens/s

llama.cpp, llama-2-7b-Q4_0

46.4 tokens/s

LLM Perf, Mistral-7B-Instruct-v0.2 (gptq)

26.4 tokens/s

LLM Perf, Mistral-7B-v0.1 (bnb)

8.01 tokens/s

LLM Perf, Mistral-7B-v0.1 (gptq)

25.7 tokens/s

LLM Perf, Mistral-7B-v0.1 (unquantized)

15.5 tokens/s

Additional

Slots

Dual-slot Single-slot

Release Date

July 16, 2021 Sept. 13, 2018

Display Outputs

No outputs
No outputs

Renting is cheaper than buying