RELEASEJanuary 30, 2025
FOUNDRYTSMC
PROCESS5 nm
ARCHITECTUREBlackwell 2.0
ROPs176
TMUs680
SPs21,760CUDA
VRAM24GBGDDR7
TDP600W
DIE FLOORPLANSCHEMATIC · NOT TO SCALE
GB202-240-K1-A1 · 5 nm
area 750 mm²
- GPC / SM
- 12 / 170 · 22 fused off
- CUDA
- 21 760
- L2 cache
- 128 MB
- Memory controller
- 16 × 32-bit
- Media engine
- NVENC · NVDEC
- PCIe
- 5.0
SECTIONSclick a block or section
General
11Identity
ManufacturerNVIDIA
ArchitectureBlackwell 2.0
NameGB202-240-K1-A1
Package
PackageFCBGA-5800
Package size63 mm × 56 mm
Silicon
FoundryTSMC
Process node5 nm
Process type4N FinFET
Transistors92,200M
Density122.9M/mm²
Die size750mm²
Configuration
19Compute blocks
Compute units (SM/CU)170
GPCs / Shader engines12
TPCs / WGPs96
Execution units (FP32)21,760CUDA
TMUs680
ROPs176
Cache
Texture L1 cache per SM64KB
L1 cache / Shared Memory per SM128KB
L2 cache131,072KB
Power
Max TDP600W
Specialised units
FP16 units24,576
FP64 units384
INT32 units24,576
SFUs3,072
Tensor cores768
Tensor cores generation5
RT cores192
RT cores generation4
Bus
Bus interfacePCI Express 5.0
Memory
03Max VRAM24GB
Max memory bus512bit
VRAM typeGDDR7
Clock speeds
03Base clock2,017MHz
Boost clock2,407MHz
Memory clock1,750MHz
APIs
11CUDA
12.0
Full
Vulkan
1.4
Full
OpenCL
3.0
Full
OpenGL
4.6
Full
DirectXFeature Level 12_2 (DirectX 12 Ultimate)
12.2
Full
Shader Model
6.9
Full
WDDMDESCRIPTIONWindows Display Driver Model — the Windows display driver model version
3.2
Full
DirectMLDESCRIPTIONMicrosoft machine-learning API on top of DirectX 12; runs on any DirectX 12 GPU
—
Full
VDPAUDESCRIPTIONVideo Decode and Presentation API for Unix — hardware decode feature set
Feature Set K
Full
DirectComputeDESCRIPTIONDirect3D compute shaders; the version is the compute shader model
5.0
Full
NVIDIA OptiXDESCRIPTIONNVIDIA ray-tracing API; the version is the newest OptiX SDK the chip supports
9.1
Full
SUPPORT LEVEL✓ Full◐ Partial✕ None
Instructions
14MMA / HMMADESCRIPTIONWarp-level matrix multiply-accumulate on the tensor cores (PTX mma.sync / wmma). From sm_70.
Full
IDP4ADESCRIPTION8-bit Integer Dot Product and Accumulate into 32 bits (PTX dp4a). From sm_61.
Full
IDP2ADESCRIPTION16-bit by 8-bit Integer Dot Product and Accumulate (PTX dp2a). From sm_61.
Full
TMA / CP.ASYNC.BULKDESCRIPTIONTensor Memory Accelerator — asynchronous bulk tensor copies between global and shared memory (SASS UTMALDG/UBLKCP). From sm_90.
Full
MBARRIERDESCRIPTIONShared-memory hardware barrier that tracks the transaction (byte) count of asynchronous copies (SASS SYNCS). Basic mbarrier from sm_80, transaction counting from sm_90.
Full
ELECTDESCRIPTIONElects a single leader thread in the warp (PTX elect.sync). From sm_90.
Full
FFMADESCRIPTIONSingle-Precision Fused Multiply-Add. From sm_20; sm_1x only had an unfused MAD.
Full
HFMA2DESCRIPTIONPacked fused multiply-add of two FP16 (or BF16) values per instruction (PTX fma.f16x2 / fma.bf16x2). FP16 from sm_53, BF16 from sm_80.
Full
DFMADESCRIPTIONDouble-Precision Fused Multiply-Add. From sm_13; throughput depends on the number of FP64 units.
Full
LOP3.LUTDESCRIPTIONArbitrary three-input bitwise logic function from a truth table (Three-Input Bitwise Logic Operation). From sm_50.
Full
SHFLDESCRIPTIONExchanges registers between warp threads without shared memory (Warp Shuffle). From sm_30.
Full
LDGSTSDESCRIPTIONAsynchronous global-to-shared memory copy that bypasses registers (PTX cp.async). From sm_80.
Full
ATOMG / ATOMSDESCRIPTIONHardware atomics on global (ATOMG) and shared (ATOMS) memory (PTX atom). Global from sm_11, shared from sm_12.
Full
MUFUDESCRIPTIONSpecial functions on the SFUs: sin, cos, ex2, lg2, rcp, rsqrt (Multi-Function Unit). On every CUDA architecture.
Full
SUPPORT LEVEL✓ Full◐ Partial✕ None
AI / Compute
21CUDA
Full
Vulkan
Full
OpenCL
Full
Floating-point formats
FP64Stream processors only
FP32Stream processors only
TF32Tensor/Matrix only
FP16Stream + Tensor/Matrix
BF16Stream + Tensor/Matrix
FP8Tensor/Matrix only
MXFP8Tensor/Matrix only
MXFP6Tensor/Matrix only
FP4Tensor/Matrix only
MXFP4Tensor/Matrix only
NVFP4Tensor/Matrix only
Integer formats
INT32Stream processors only
INT16Stream processors only
INT8Stream + Tensor/Matrix
INT4Tensor/Matrix only
INT2None
INT1None
Other
Async DMA / TMA✓ Yes
SUPPORT LEVEL✓ Full◐ Partial✕ None
Theoretical performance
13FP16 (half)1:1118.3TFLOPS
FP32 (single)118.3TFLOPS
FP64 (double)1:641.85TFLOPS
INT8 (shaders)473.2TOPS
INT4 (shaders)473.2TOPS
Tensor formats
TF32 (tensor)236.6TFLOPS
BF16 (tensor)473.2TFLOPS
FP8 (tensor)946.4TFLOPS
FP6 (tensor)946.4TFLOPS
FP4 (tensor)1.893PFLOPS
INT8 (tensor)946.4TOPS
INT4 (tensor)1.893POPS
With sparsity✓ Yes
Measured performance
01FP16 (half)0TFLOPS
Media
04NVENCDESCRIPTIONHardware video encoder
9th Gen
NVDECDESCRIPTIONHardware video decoder
6th Gen
PureVideo HDDESCRIPTIONNVIDIA hardware video decode feature set
VP11
AV1✓ Yes
// DIE FLOORPLANReference data. No buy prices, stock, or ads.UMTALE.LAB/DB · 100