MohamedElashri/nvidia_gpu_information.md

Created September 19, 2024 10:47

Star () You must be signed in to star a gist
Fork () You must be signed in to fork a gist

Learn more about clone URLs
Clone this repository at <script src="https://gist.github.com/MohamedElashri/09453956091ea4ca506bf88bd8eaa1d6.js"></script>
Save MohamedElashri/09453956091ea4ca506bf88bd8eaa1d6 to your computer and use it in GitHub Desktop.

Download ZIP

Information about some NVIDIA GPUs

Raw

nvidia_gpu_information.md

Memory Type	NVIDIA GeForce RTX 3090	NVIDIA RTX A5000	NVIDIA GeForce RTX 2080 Ti
Global Memory	24 GB GDDR6X	24 GB GDDR6	11 GB GDDR6
Memory Interface	384-bit	384-bit	352-bit
Memory Bandwidth	936 GB/s	768 GB/s	616 GB/s
L2 Cache	6 MB	Not specified	5.5 MB
Shared Memory	164 KB per SM	164 KB per SM	64 KB per SM
Constant Memory	64 KB	64 KB	64 KB
Texture Memory	Part of global memory	Part of global memory	Part of global memory
Local Memory	Part of global memory	Part of global memory	Part of global memory
Registers	Not specified	Not specified	64K 32-bit registers per SM
L1 Cache	128 KB per SM	128 KB per SM	64 KB per SM

Sources:

Memory Type	NVIDIA GeForce RTX 3090	NVIDIA RTX A5000	NVIDIA GeForce RTX 2080 Ti	Relative Speed	Scope
Global Memory	24 GB GDDR6X[1]	24 GB GDDR6[2]	11 GB GDDR6[3]	Slow	All threads
Memory Interface	384-bit[1]	384-bit[2]	352-bit[3]	N/A	N/A
Memory Bandwidth	936 GB/s[1]	768 GB/s[2]	616 GB/s[3]	N/A	N/A
L1 Cache	128 KB/SM	128 KB/SM	64 KB/SM	Fastest	Thread block
L2 Cache	6 MB[1]	6 MB[2]	5.5 MB[3]	Very Fast	All SMs
Shared Memory	48 KB/SM	48 KB/SM	48 KB/SM	Very Fast	Thread block
Registers	65,536/SM	65,536/SM	65,536/SM	Fastest	Thread
Texture Memory	Uses L1/L2	Uses L1/L2	Uses L1/L2	Fast	Thread block
Local Memory	Uses Global	Uses Global	Uses Global	Slow	Thread
Constant Memory	64 KB	64 KB	64 KB	Fast	All threads (read-only)

NVIDIA GeForce RTX 3090

Memory Type	Typical Size	Relative Speed	Scope
Global Memory	24 GB GDDR6X[1]	Slow	All threads
Memory Interface	384-bit[1]	N/A	N/A
Memory Bandwidth	936 GB/s[1]	N/A	N/A
L1 Cache	128 KB/SM	Fastest	Thread block
L2 Cache	6 MB[1]	Very Fast	All SMs
Shared Memory	48 KB/SM	Very Fast	Thread block
Registers	65,536/SM	Fastest	Thread
Texture Memory	Uses L1/L2	Fast	Thread block
Local Memory	Uses Global	Slow	Thread
Constant Memory	64 KB	Fast	All threads (read-only)

Specification	Details
CUDA Cores	10496
Base Clock	1395 MHz
Boost Clock	1695 MHz (Gaming Mode), 1725 MHz (OC Mode)
Memory Size	24 GB GDDR6X
Memory Interface	384-bit
Memory Bandwidth	936 GB/s
Memory Speed	19.5 Gbps
RT Cores	82 (2nd Generation)
Tensor Cores	328 (3rd Generation)
Architecture	Ampere
Power Consumption	350W
Recommended PSU	750W
Power Connectors	2x 8-pin
Maximum Digital Resolution	7680x4320
Standard Display Connectors	HDMI 2.1, 3x DisplayPort 1.4a
Multi Monitor Support	4
Card Dimensions	313 mm x 138 mm (L x W)
Slot Width	3-slot

NVIDIA RTX A5000

Memory Type	Typical Size	Relative Speed	Scope
Global Memory	24 GB GDDR6[2]	Slow	All threads
Memory Interface	384-bit[2]	N/A	N/A
Memory Bandwidth	768 GB/s[2]	N/A	N/A
L1 Cache	128 KB/SM	Fastest	Thread block
L2 Cache	6 MB[2]	Very Fast	All SMs
Shared Memory	48 KB/SM	Very Fast	Thread block
Registers	65,536/SM	Fastest	Thread
Texture Memory	Uses L1/L2	Fast	Thread block
Local Memory	Uses Global	Slow	Thread
Constant Memory	64 KB	Fast	All threads (read-only)

Specification	Details
CUDA Cores	8192
Base Clock	Not specified
Boost Clock	Not specified
Memory Size	24 GB GDDR6
Memory Interface	384-bit
Memory Bandwidth	768 GB/s
Memory Speed	Not specified
RT Cores	64 (2nd Generation)
Tensor Cores	256 (3rd Generation)
Architecture	Ampere
Power Consumption	230W
Recommended PSU	Not specified
Power Connectors	1x 8-pin PCIe
Maximum Digital Resolution	4x 4096x2160 @ 120 Hz
Standard Display Connectors	4x DisplayPort 1.4a
Multi Monitor Support	4
Card Dimensions	4.4" H x 10.5" L
Slot Width	Dual slot

NVIDIA GeForce RTX 2080 Ti

Memory Type	Typical Size	Relative Speed	Scope
Global Memory	11 GB GDDR6[3]	Slow	All threads
Memory Interface	352-bit[3]	N/A	N/A
Memory Bandwidth	616 GB/s[3]	N/A	N/A
L1 Cache	64 KB/SM	Fastest	Thread block
L2 Cache	5.5 MB[3]	Very Fast	All SMs
Shared Memory	48 KB/SM	Very Fast	Thread block
Registers	65,536/SM	Fastest	Thread
Texture Memory	Uses L1/L2	Fast	Thread block
Local Memory	Uses Global	Slow	Thread
Constant Memory	64 KB	Fast	All threads (read-only)

Specification	Details
CUDA Cores	4352
Base Clock	1350 MHz
Boost Clock	1545 MHz (1635 MHz for some models)
Memory Size	11 GB GDDR6
Memory Interface	352-bit
Memory Bandwidth	616 GB/s
Memory Speed	14 Gbps
RT Cores	68 (1st Generation)
Tensor Cores	544 (2nd Generation)
Architecture	Turing
Power Consumption	250W (up to 300W for some models)
Recommended PSU	650W
Power Connectors	2x 8-pin
Maximum Digital Resolution	7680x4320
Standard Display Connectors	DisplayPort 1.4a, HDMI 2.0b, USB Type-C
Multi Monitor Support	4
Card Dimensions	Varies by model (e.g., 327 x 140 x 55.6 mm for some)
Slot Width	2-slot (may vary by model)

Memory Type	NVIDIA GeForce RTX 3090	NVIDIA RTX A5000	NVIDIA GeForce RTX 2080 Ti
Global Memory	24 GB GDDR6X	24 GB GDDR6	11 GB GDDR6
Memory Interface	384-bit	384-bit	352-bit
Memory Bandwidth	936 GB/s	768 GB/s	616 GB/s
Memory Speed	19.5 Gbps	Not specified	14 Gbps
L2 Cache	6 MB	Not specified	5.5 MB
Shared Memory	164 KB per SM	164 KB per SM	64 KB per SM
Constant Memory	64 KB	64 KB	64 KB
Texture Memory	Part of global memory	Part of global memory	Part of global memory
Local Memory	Part of global memory	Part of global memory	Part of global memory
Registers	Not specified	Not specified	64K 32-bit registers per SM
L1 Cache	128 KB per SM	128 KB per SM	64 KB per SM
ECC Support	Not specified	Yes	Not specified

[1] Source: NVIDIA GeForce RTX 3090 specifications

[2] Source: NVIDIA RTX A5000 specifications

[3] Source: NVIDIA GeForce RTX 2080 Ti specifications

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment