NVIDIA CMP 30HX vs NVIDIA GB10 Comparison

Where Each One Wins

The NVIDIA GB10 is the clear winner in every recorded benchmark category. It takes both of the two head-to-head tests, with an 84.3% lead in Geekbench OpenCL and an 83.5% lead in Geekbench Vulkan. That is not a narrow margin; it is a decisive generational gap. The GB10 also holds a 95th percentile rank among all GPUs in the database, while the CMP 30HX sits at the 89th percentile.

The use-case split is straightforward: the GB10 is a server-class Blackwell part designed for compute-heavy workloads, and it dominates in every measured metric. The CMP 30HX, a Turing-based mining GPU from 2021, is end-of-life and offers no display outputs, making it unsuitable for any interactive or rendering workload. The GB10, by contrast, has a single HDMI output and integrates into a system as an IGP (integrated graphics processor) with a PCIe 5.0 x16 interface.

For users looking at raw compute throughput, the GB10 delivers 29.71 TFLOPS of FP32 performance, nearly six times the CMP 30HX's 5.027 TFLOPS. In FP16, the GB10 again leads with 29.71 TFLOPS (1:1 ratio), while the CMP 30HX offers 10.05 TFLOPS (2:1 ratio). The GB10 also has far more shading units (6144 vs. 1408), more TMUs (384 vs. 88), and the same 48 ROPs. The CMP 30HX has no ray tracing cores and no tensor cores, whereas the GB10 includes 48 RT cores and 384 tensor cores.

Memory is another area where the GB10 is unmatched: 128 GB of LPDDR5X on a 256-bit bus, delivering 273.2 GB/s of bandwidth. The CMP 30HX has 6 GB of GDDR6 on a 192-bit bus, but a higher bandwidth of 336.0 GB/s. That is the one metric where the CMP 30HX wins: memory bandwidth per second. But it comes with a fraction of the capacity, and the GB10's bandwidth is still ample for its compute class. The GB10 also runs at a higher boost clock (2418 MHz vs. 1785 MHz) and a higher base clock (1665 MHz vs. 1530 MHz).

In short, the GB10 wins on compute, memory capacity, feature set, and API support breadth (though the GB10's APIs are listed as N/A, the CMP 30HX supports DirectX 12, OpenGL 4.6, and Vulkan 1.4). The CMP 30HX wins only on memory bandwidth and has a lower power draw (125 W vs. 140 W), but that advantage is irrelevant given its massive performance deficit.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The NVIDIA GB10 has an average benchmark score of 117393, while the NVIDIA CMP 30HX has an average score of 63842. The GB10 is roughly 84% ahead.

Q: Does the CMP 30HX have any ray tracing or tensor cores?

A: No. The CMP 30HX lists no RT cores and no tensor cores. The GB10, in contrast, has 48 RT cores and 384 tensor cores.

Q: What is the memory capacity difference between the two?

A: The GB10 has 128 GB of LPDDR5X memory, while the CMP 30HX has 6 GB of GDDR6 memory. The GB10 uses a 256-bit bus, the CMP 30HX uses a 192-bit bus.

Q: Which GPU has higher memory bandwidth?

A: The CMP 30HX has higher memory bandwidth at 336.0 GB/s, compared to the GB10's 273.2 GB/s, despite the GB10 having far more memory capacity.

Q: Is the CMP 30HX still in production?

A: No, the CMP 30HX is end-of-life. The GB10 is listed as active production.

Q: What is the power consumption difference?

A: The GB10 has a TDP of 140 W, while the CMP 30HX has a TDP of 125 W. Both suggest a 300 W power supply.

Head-to-Head Benchmarks

The recorded data shows two Geekbench tests, and the GB10 wins both by very large margins. In Geekbench OpenCL, the GB10 scores 120137 against the CMP 30HX's 65199, a delta of 84.3%. In Geekbench Vulkan, the GB10 scores 114648 against 62484, a delta of 83.5%. These are not close contests; the GB10 nearly doubles the CMP 30HX in both workloads.

To put these numbers in context, the GB10's nearest rival in the database is the NVIDIA RTX 4000 SFF Ada Generation, which averages 117088, just 0.3% behind the GB10's average of 117393. The GB10 also leads the AMD Radeon PRO W7700 (118976 average, 1.3% behind) and the NVIDIA Tesla V100 SXM2 16 GB (114395, 2.6% behind). The CMP 30HX, by contrast, sits very close to its nearest rivals: the AMD Radeon RX 9060 XT LP (63830, 0% delta), the AMD Radeon RX 7600M (63775, 0.1% ahead), and the AMD Radeon Pro Vega 56 (63693, 0.2% ahead). The CMP 30HX is effectively in a dead heat with those cards, while the GB10 is in a different performance class entirely.

The FP32 compute numbers reinforce this. The GB10 posts 29.71 TFLOPS, while the CMP 30HX posts 5.027 TFLOPS. That is roughly a 5.9x difference. Pixel rate: 116.1 GPixel/s vs. 85.68 GPixel/s. Texture rate: 928.5 GTexel/s vs. 157.1 GTexel/s. Every compute metric except memory bandwidth favors the GB10, and often by an order of magnitude.

Specification Differences

The two cards differ in nearly every specification. The GB10 uses a GB20B chip on a 5 nm process, while the CMP 30HX uses a TU116 chip on a 12 nm process. The GB10 has 6144 shading units, 384 TMUs, and 48 ROPs; the CMP 30HX has 1408 shading units, 88 TMUs, and 48 ROPs. The GB10 includes 48 RT cores and 384 tensor cores; the CMP 30HX has none.

Clock speeds differ: the GB10 runs at a base of 1665 MHz and boosts to 2418 MHz. The CMP 30HX runs at 1530 MHz base and 1785 MHz boost. Memory clocks: the GB10's memory runs at 1067 MHz (8.5 Gbps effective), while the CMP 30HX's memory runs at 1750 MHz (14 Gbps effective). The GB10 has 128 GB of LPDDR5X on a 256-bit bus; the CMP 30HX has 6 GB of GDDR6 on a 192-bit bus. Bandwidth favors the CMP 30HX: 336.0 GB/s vs. 273.2 GB/s.

Power: the GB10 is rated at 140 W, the CMP 30HX at 125 W. The GB10 is an IGP with no power connectors and a slot width of IGP; the CMP 30HX is a dual-slot card with one 8-pin connector. The GB10 uses a PCIe 5.0 x16 interface; the CMP 30HX uses PCIe 1.0 x4. Display outputs: the GB10 has one HDMI, the CMP 30HX has none. Dimensions: the GB10 is 150 mm long, 51 mm high, and 150 mm wide; the CMP 30HX is 229 mm long, 111 mm high, and 35 mm wide.

Die size differs as well: the GB10 is 382 mm², the CMP 30HX is 284 mm². The CMP 30HX has 6,600 million transistors, while the GB10's transistor count is listed as unknown. The CMP 30HX has a transistor density of 23.2M per mm². The GB10's density is not listed. The GB10's release date is 2025-10-14, the CMP 30HX's is 2021-02-24. The GB10 is active, the CMP 30HX is end-of-life.

Architecture Differences

The GB10 is built on Blackwell 2.0, part of the Server Blackwell (Bxx) generation, fabricated on a 5 nm process by TSMC. The CMP 30HX is built on Turing, part of the Mining GPUs generation, also fabricated by TSMC but on a 12 nm process. The GB10 is the successor generation to Server Hopper and the predecessor to Server Rubin; the CMP 30HX has no listed predecessor or successor.

The GB10's architecture includes 48 RT cores and 384 tensor cores, enabling ray tracing and AI acceleration. The CMP 30HX has neither, which is consistent with its purpose as a mining-focused product. The GB10's FP16 performance is at a 1:1 ratio with FP32 (both 29.71 TFLOPS), meaning it does not boost FP16 throughput. The CMP 30HX has a 2:1 ratio, delivering 10.05 TFLOPS of FP16 from 5.027 TFLOPS of FP32.

API support: the GB10 lists N/A for DirectX, OpenGL, and Vulkan, which is typical for a compute/IGP server part that does not target traditional graphics APIs. The CMP 30HX supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.4, though it has no display outputs to use them with.

The GB10's memory subsystem is a unified LPDDR5X pool of 128 GB, which is unusually large for a GPU and suggests it is designed for large models or data sets. The CMP 30HX's 6 GB GDDR6 is small by comparison. The GB10's bandwidth is lower (273.2 GB/s) than the CMP 30HX's (336.0 GB/s), but the capacity difference is 21x in favor of the GB10.

In terms of physical integration, the GB10 is designed as an IGP, meaning it sits on a motherboard without its own slot or power connectors. The CMP 30HX is a traditional add-in board, dual-slot, with a single 8-pin power connector. The GB10's PCIe 5.0 x16 interface offers far more bandwidth to the host system than the CMP 30HX's PCIe 1.0 x4, which is a legacy low-bandwidth interface. The GB10 also has a much smaller footprint (150 mm length vs. 229 mm), making it suitable for dense server environments.

The production statuses reinforce the generational gap: the GB10 is active and current, while the CMP 30HX is end-of-life. The launch MSRP for the GB10 is 3,999 USD; for the CMP 30HX it is 799 USD. The GB10 is positioned as a high-end server compute part, while the CMP 30HX was a mid-range mining card that is no longer sold. The benchmark data shows that the GB10's performance is not just a step up; it is a leap across nearly every measurable dimension.