NVIDIA PCIe Cards

Select All

Product Name	Memory	GPU Die	SM Count	CUDA Cores	TMUs	ROPs	Tensor Cores	Tensor Gen	RT Cores	RT Gen	Base Clock	Boost Clock	Mem Clock	Mem Bus	Mem BW	Mem Type	L1 per SM	L2 Cache	FP16 TFLOPS (non-Tensor)	FP4 Dense TFLOPS	FP4 Sparse TFLOPS (2:4)	FP8 Dense TFLOPS (FP16 acc)	FP8 Sparse TFLOPS (FP16 acc, 2:4)	FP16 Tensor Dense TFLOPS (FP16 acc)	FP16 Tensor Sparse TFLOPS (FP16 acc, 2:4)	Official AI TOPS	RT TFLOPS	Power	PCIe	Notes
NVIDIA GeForce RTX 5090	32 GB GDDR7	GB202	170	21760	680	176	680	5th Gen	170	4th Gen	2017 MHz	2407 MHz	1750 MHz (28 Gbps)	512-bit	1792 GB/s	GDDR7	128 KB	96 MB	104.8	1676	3352	838	1676	419	838	3352 AI TOPS (sparse)	317.5	575 W	PCIe 5.0 x16	GeForce flagship; GB202 with 170 SMs enabled; Official AI TOPS is FP4 Sparse value
NVIDIA GeForce RTX 5090 D	32 GB GDDR7	GB202	170	21760	680	176	680	5th Gen	170	4th Gen	2017 MHz	2407 MHz	1750 MHz (28 Gbps)	512-bit	1792 GB/s	GDDR7	128 KB	96 MB	104.8	1188	2375	594	1188	297	594	2375 AI TOPS (sparse, export control restricted)	317.5	575 W	PCIe 5.0 x16	China market exclusive; AI compute limited by export controls (FP4 Sparse 2375 TOPS)
NVIDIA GeForce RTX 5090 D V2	24 GB GDDR7	GB202	170	21760	680	176	680	5th Gen	170	4th Gen	2017 MHz	2407 MHz	1750 MHz (28 Gbps)	384-bit	1344 GB/s	GDDR7	128 KB	96 MB	104.8	1188	2375	594	1188	297	594	2375 AI TOPS (sparse, export control restricted)	317.5	575 W	PCIe 5.0 x16	China market exclusive V2; memory reduced to 24 GB / 384-bit; same AI compute as D version
NVIDIA RTX 6000D	84 GB GDDR7	GB202	156	19968	624	192	624	5th Gen	156	4th Gen	1590 MHz	2430 MHz	1750 MHz (28 Gbps)	448-bit	1568 GB/s	GDDR7 ECC	128 KB	128 MB	97	1556	3112	778	1556	389	778	~3112 AI TOPS (sparse, estimated)	302	600 W	PCIe 5.0 x16	China market exclusive; SM count and memory reduced for export compliance; compute is estimated
NVIDIA RTX PRO 5000 72 GB Blackwell	72 GB GDDR7	GB202	110	14080	440	160	440	5th Gen	110	4th Gen	1740 MHz	2377 MHz	1750 MHz (28 Gbps)	384-bit	1344 GB/s	GDDR7 ECC	128 KB	96 MB	66.94	1071	2142	536	1071	268	536	2142 AI TOPS (sparse)	209	300 W	PCIe 5.0 x16	Professional workstation GPU; 72 GB large memory version; same core config as 48 GB version
NVIDIA RTX PRO 5000 Blackwell	48 GB GDDR7	GB202	110	14080	440	160	440	5th Gen	110	4th Gen	1740 MHz	2377 MHz	1750 MHz (28 Gbps)	384-bit	1344 GB/s	GDDR7 ECC	128 KB	96 MB	66.94	1071	2142	536	1071	268	536	2142 AI TOPS (sparse)	209	300 W	PCIe 5.0 x16	Professional workstation GPU; standard 48 GB version
NVIDIA RTX PRO 6000 Blackwell	96 GB GDDR7	GB202	188	24064	752	192	752	5th Gen	188	4th Gen	1590 MHz	2617 MHz	1750 MHz (28 Gbps)	512-bit	1792 GB/s	GDDR7 ECC	128 KB	128 MB	126	2015	4030	1008	2015	504	1008	4000 AI TOPS (sparse, ≈4030)	381.8	600 W	PCIe 5.0 x16	Professional workstation flagship; full GB202; L2 Cache 128 MB; Official AI TOPS is FP4 Sparse value
NVIDIA RTX PRO 6000 Blackwell Max-Q	96 GB GDDR7	GB202	188	24064	752	192	752	5th Gen	188	4th Gen	1035 MHz	2280 MHz	1750 MHz (28 Gbps)	512-bit	1792 GB/s	GDDR7 ECC	128 KB	128 MB	109.7	1756	3511	878	1756	439	878	3511 AI TOPS (sparse)	332.6	300 W	PCIe 5.0 x16	Low-power workstation version; 300 W; suitable for dense deployment; Official AI TOPS is FP4 Sparse value
NVIDIA RTX PRO 6000 Blackwell Server	96 GB GDDR7	GB202	188	24064	752	192	752	5th Gen	188	4th Gen	1590 MHz	2617 MHz	1750 MHz (28 Gbps)	512-bit	1597 GB/s	GDDR7 ECC	128 KB	128 MB	120	2015	4030	1008	2015	504	1008	4000 AI TOPS (sparse, ≈4030)	355	600 W	PCIe 5.0 x16	Server edition; passive cooling; 600 W; suitable for dense data center deployment
NVIDIA RTX PRO 6000D Blackwell Max-Q	96 GB GDDR7	GB202	188	24064	752	192	752	5th Gen	188	4th Gen	1590 MHz	2288 MHz	1750 MHz (28 Gbps)	512-bit	1792 GB/s	GDDR7 ECC	128 KB	128 MB	109.7	1756	3511	878	1756	439	878	3511 AI TOPS (sparse, estimated)	332.6	300 W	PCIe 5.0 x16	China market exclusive Max-Q version; specs close to Max-Q; Official AI TOPS is FP4 Sparse value
																														All China exclusive version information comes from leaked sources and may be inaccurate