NVIDIA CMP 90HX vs NVIDIA GB10 Comparison

The Verdict

The NVIDIA GB10 is the clear performance leader in this comparison. Its Geekbench OpenCL score of 120,137 is 74.1% higher than the NVIDIA CMP 90HX's 69,000, a decisive margin that places the GB10 in the 95th percentile of all GPUs in the database. The CMP 90HX, while still respectable at the 90th percentile, trails by a significant gap. The GB10 is the choice for anyone running OpenCL workloads or Vulkan applications, as it also records a Geekbench Vulkan score of 114,648, a result the CMP 90HX cannot match since it has no display outputs and is not designed for general graphics tasks. The CMP 90HX is a legacy mining part, now end-of-life, while the GB10 is an active, integrated server processor.

Architecture Differences

The two chips come from entirely different NVIDIA architectural generations and manufacturing processes. The GB10 uses the GB20B chip built on the Blackwell 2.0 architecture, fabricated on a 5 nm process at TSMC. Its die measures 382 mm². The CMP 90HX, by contrast, uses the GA102 chip built on the Ampere architecture, fabricated on an 8 nm process at Samsung. Its die is substantially larger at 628 mm², with 28,300 million transistors and a transistor density of 45.1M per mm².

The GB10 is a server-oriented part, listed under the Server Blackwell (Bxx) generation, with its predecessor being Server Hopper and successor Server Rubin. The CMP 90HX belongs to the Mining GPUs generation and has no listed predecessor or successor. The GB10 is an integrated graphics processor (IGP), requiring no power connectors and consuming a 140 W TDP, with a suggested PSU of 300 W. The CMP 90HX is a dual-slot card with two 8-pin power connectors, a 320 W TDP, and a suggested PSU of 700 W.

Memory configurations differ sharply. The GB10 offers 128 GB of LPDDR5X memory on a 256-bit bus, yielding 273.2 GB/s of bandwidth. The CMP 90HX offers only 10 GB of GDDR6X memory on a 320-bit bus, but its bandwidth is far higher at 760.3 GB/s due to the faster memory type and wider bus. The GB10's memory clock is listed at 1067 MHz with 8.5 Gbps effective, while the CMP 90HX runs at 1188 MHz with 19 Gbps effective. The CMP 90HX also has a different bus interface: PCIe 1.0 x4, whereas the GB10 uses PCIe 5.0 x16. The CMP 90HX has no display outputs, while the GB10 includes a single HDMI output.

The GB10 features 6,144 shading units, 384 TMUs, 48 ROPs, 48 RT cores, and 384 tensor cores. The CMP 90HX has 6,400 shading units, 200 TMUs, 80 ROPs, 50 RT cores, and 200 tensor cores. The GB10's core clock is higher: 1665 MHz base and 2418 MHz boost, versus the CMP 90HX's 1500 MHz base and 1710 MHz boost. The GB10 also has a higher texture rate of 928.5 GTexel/s versus the CMP 90HX's 342.0 GTexel/s, though the CMP 90HX has a higher pixel rate at 136.8 GPixel/s versus the GB10's 116.1 GPixel/s. In raw FP32 and FP16 compute, the GB10 delivers 29.71 TFLOPS in both, while the CMP 90HX delivers 21.89 TFLOPS in both.

The API support also differs. The GB10 lists no DirectX, OpenGL, or Vulkan support in the database. The CMP 90HX supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Head-to-Head Benchmarks

The database records only one shared benchmark between the two: Geekbench OpenCL. The GB10 scores 120,137, and the CMP 90HX scores 69,000. The delta is 74.1% in favor of the GB10. This is a substantial performance gap that dwarfs the differences seen when comparing each card to its nearest rivals.

Looking at the GB10's nearest rivals in the database, its average benchmark score is 117,393. It sits 0.3% ahead of the NVIDIA RTX 4000 SFF Ada Generation (117,088), 1.3% behind the AMD Radeon PRO W7700 (118,976), 2.6% ahead of the NVIDIA Tesla V100 SXM2 16 GB (114,395), and 3.0% ahead of the NVIDIA RTX A5500 Mobile (113,944). The GB10's OpenCL score of 120,137 is therefore above its own average, indicating strong performance in that specific test.

The CMP 90HX's average benchmark score is 69,000, which equals its only recorded OpenCL score. Its nearest rivals cluster closely around it: the Intel Arc A770 is 0.3% ahead (68,809), the AMD Radeon Instinct MI25 is 0.6% ahead (68,562), the AMD Radeon Pro WX 8200 is 1.2% behind (69,870), and the NVIDIA Quadro P6000 is 1.4% behind (69,986). The CMP 90HX is essentially performance-equivalent to these older or mid-range parts. The 74.1% lead of the GB10 over the CMP 90HX is far larger than any of these intra-tier gaps, making the performance hierarchy unambiguous.

Specification Differences

The following fields differ between the NVIDIA GB10 and NVIDIA CMP 90HX:

• Chip: GB20B vs GA102
• Architecture: Blackwell 2.0 vs Ampere
• Generation: Server Blackwell (Bxx) vs Mining GPUs
• Process node: 5 nm vs 8 nm
• Foundry: TSMC vs Samsung
• Die size: 382 mm² vs 628 mm²
• Transistors: unknown vs 28,300 million
• Transistor density: null vs 45.1M / mm²
• Base clock: 1665 MHz vs 1500 MHz
• Boost clock: 2418 MHz vs 1710 MHz
• Memory clock: 1067 MHz 8.5 Gbps effective vs 1188 MHz 19 Gbps effective
• Memory size: 128 GB vs 10 GB
• Memory type: LPDDR5X vs GDDR6X
• Memory bus width: 256 bit vs 320 bit
• Memory bandwidth: 273.2 GB/s vs 760.3 GB/s
• Shading units: 6144 vs 6400
• TMUs: 384 vs 200
• ROPs: 48 vs 80
• RT cores: 48 vs 50
• Tensor cores: 384 vs 200
• Pixel rate: 116.1 GPixel/s vs 136.8 GPixel/s
• Texture rate: 928.5 GTexel/s vs 342.0 GTexel/s
• FP32: 29.71 TFLOPS vs 21.89 TFLOPS
• FP16: 29.71 TFLOPS (1:1) vs 21.89 TFLOPS (1:1)
• TDP: 140 W vs 320 W
• Slot width: IGP vs Dual-slot
• Power connectors: None vs 2x 8-pin
• Suggested PSU: 300 W vs 700 W
• Bus interface: PCIe 5.0 x16 vs PCIe 1.0 x4
• Display outputs: 1x HDMI vs No outputs
• DirectX: N/A vs 12 Ultimate (12_2)
• OpenGL: N/A vs 4.6
• Vulkan: N/A vs 1.4
• Length: 150 mm 5.9 inches vs 285 mm 11.2 inches
• Height: 51 mm 2 inches vs 112 mm 4.4 inches
• Width: 150 mm 5.9 inches vs null
• Production status: Active vs End-of-life
• Release date: 2025-10-14 vs 2021-07-27
• Predecessor: Server Hopper vs null
• Successor: Server Rubin vs null
• Launch MSRP: 3,999 USD vs null

FAQ

Q: Which GPU has the higher average benchmark score?

A: The NVIDIA GB10 has an average benchmark score of 117,393, placing it in the 95th percentile of all GPUs. The NVIDIA CMP 90HX has an average score of 69,000, placing it in the 90th percentile.

Q: How much faster is the GB10 in OpenCL?

A: In the Geekbench OpenCL test, the GB10 scores 120,137 versus the CMP 90HX's 69,000, a 74.1% advantage for the GB10.

Q: Does the CMP 90HX support display outputs?

A: No, the CMP 90HX has no display outputs. The GB10, by contrast, includes one HDMI output.

Q: What are the power requirements for each card?

A: The GB10 has a TDP of 140 W and a suggested PSU of 300 W, with no power connectors. The CMP 90HX has a TDP of 320 W, a suggested PSU of 700 W, and requires two 8-pin power connectors.

Q: Which card has more memory bandwidth?

A: The CMP 90HX has significantly higher memory bandwidth at 760.3 GB/s, despite having only 10 GB of memory. The GB10 has 128 GB of memory but a lower bandwidth of 273.2 GB/s.

Q: Is the CMP 90HX still in production?

A: No, the CMP 90HX is listed as end-of-life. The GB10 is listed as active production. The CMP 90HX was released in 2021, while the GB10 was released in 2025.

Q: Does the GB10 support modern graphics APIs?

A: The database lists no DirectX, OpenGL, or Vulkan support for the GB10. The CMP 90HX supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.