AMD Radeon RX 5600M vs NVIDIA CMP 90HX Comparison

The NVIDIA CMP 90HX and the AMD Radeon RX 5600M occupy two very different corners of the GPU landscape: one is a desktop mining card built from the GA102 silicon behind NVIDIA's high-end Ampere parts, the other a mobile RDNA 1.0 chip designed for laptops. The database places them in roughly adjacent performance territory by percentile rank, with the CMP 90HX sitting in the 90th percentile against all GPUs and the RX 5600M in the 85th, yet the recorded measurements suggest the gap between them is wider than those percentiles imply. What follows is a closer look at where each card stands according to the data.

Head-to-Head Benchmarks

Only one test appears in the recorded head-to-head set for this pairing, and it is decisive. In Geekbench OpenCL, the CMP 90HX scores 69000 against 59589 for the RX 5600M, a delta of 15.8 percent in NVIDIA's favor. That is the sole direct comparison in the database, and the win tally reflects it: one win for the CMP 90HX, zero for the RX 5600M.

Context matters here. The CMP 90HX's 69000 OpenCL score sits within a razor-thin cluster of professional and enthusiast parts: the Intel Arc A770 averages 68809 (a delta of just 0.3 percent), the AMD Radeon Instinct MI25 averages 68562 (0.6 percent behind), the AMD Radeon Pro WX 8200 averages 69870 (1.2 percent ahead), and the NVIDIA Quadro P6000 averages 69986 (1.4 percent ahead). In other words, the mining card lands in a statistical dead heat with datacenter and workstation hardware in this compute test.

The RX 5600M's 59589 OpenCL result tells a different story about its competitive set. Its nearest rivals include the Intel Arc A530M at 46614 (delta 0), the AMD Radeon RX 6550M at 46702 (0.2 percent behind the AMD mobile part), the NVIDIA RTX A2000 at 46043 (1.2 percent behind), and, curiously, the NVIDIA RTX 5880 Ada Generation at 45972 (1.4 percent behind). That last entry is worth pausing on: an average benchmark score of 46601 for a mobile Navi part sitting alongside an Ada-generation professional GPU in the averages. It suggests the RX 5600M's average is pulled across heterogeneous test suites, since its individual results span 1320 in 3DMark Steel Nomad DX12, 76653 in Geekbench Metal, 59589 in Geekbench OpenCL, and 48843 in Geekbench Vulkan. The Metal score of 76653 is actually the highest single number recorded for either card in this comparison, though it comes from a test the CMP 90HX does not have a recorded entry in.

FAQ

Q: Which GPU wins the only shared benchmark?

A: The NVIDIA CMP 90HX. It scores 69000 in Geekbench OpenCL versus 59589 for the RX 5600M, a 15.8 percent advantage.

Q: What percentile does each card hold against all GPUs in the database?

A: The CMP 90HX sits in the 90th percentile; the RX 5600M sits in the 85th percentile.

Q: Are both cards still in production?

A: No. Both are listed as end-of-life. The RX 5600M was released on July 6, 2020, and the CMP 90HX on July 27, 2021.

Q: Which card has more memory bandwidth?

A: The CMP 90HX, with 760.3 GB/s across a 320-bit GDDR6X bus. The RX 5600M delivers 288.0 GB/s across a 192-bit GDDR6 bus.

Q: Do either of these cards support ray tracing hardware?

A: The CMP 90HX lists 50 RT cores and 200 tensor cores. The RX 5600M has no RT cores or tensor cores listed, consistent with its RDNA 1.0 architecture.

Q: Can the CMP 90HX drive a display?

A: No. The database lists its display outputs as none, which fits its origin as a dedicated mining card. The RX 5600M's outputs are listed as portable device dependent, as befits a mobile part integrated into laptop designs.

Architecture Differences

These two chips could hardly be more different in origin. The CMP 90HX uses the GA102 die on Samsung's 8 nm process, packing 28,300 million transistors into a 628 mm² die for a density of 45.1M per mm². The RX 5600M uses the Navi 10 die on TSMC's 7 nm process, with 10,300 million transistors on a 251 mm² die at 41.0M per mm². The AMD die is less than half the area and roughly a third of the transistor count, which is a natural consequence of one being a big-desktop-class silicon and the other a mobile design.

The compute resource gap follows the same pattern. The CMP 90HX carries 6400 shading units, 200 TMUs, and 80 ROPs, producing 21.89 TFLOPS of FP32 and 21.89 TFLOPS of FP16 at a 1:1 ratio. The RX 5600M has 2304 shading units, 144 TMUs, and 64 ROPs, reaching 5.829 TFLOPS FP32 but 11.66 TFLOPS FP16 thanks to a 2:1 rate. Interestingly, that 2:1 FP16 ratio means the mobile AMD part more than doubles its FP32 throughput in half-precision work, while the NVIDIA card gains nothing by dropping precision. Whether that matters depends entirely on workload, but it is one of the few spec dimensions where the RX 5600M narrows the raw gap.

Clocks tell the opposite story. The CMP 90HX runs a 1500 MHz base and 1710 MHz boost, well above the RX 5600M's 1035 MHz base, 1265 MHz boost, and 1190 MHz game clock. The NVIDIA card's higher clocks on vastly larger silicon explain its 342.0 GTexel/s texture rate and 136.8 GPixel/s pixel rate, versus 182.2 GTexel/s and 80.96 GPixel/s for the AMD part.

Feature support also diverges. The CMP 90HX lists DirectX 12 Ultimate (feature level 12_2), while the RX 5600M lists DirectX 12 (12_1). Both support OpenGL 4.6 and Vulkan 1.4. The bus interfaces are a study in extremes: the mining card uses PCIe 1.0 x4, a minimal link reflecting its mining purpose, while the RX 5600M uses a full PCIe 4.0 x16 interface.

The Verdict

The data points to a clear split. For raw compute throughput, the CMP 90HX dominates on paper: nearly four times the FP32 TFLOPS, more than 2.6 times the memory bandwidth, double the memory capacity at 10 GB versus 6 GB, and a 15.8 percent lead in the one shared benchmark. Anyone whose workload resembles the OpenCL tests recorded here, and who can supply the card's 320 W TDP through two 8-pin connectors with a suggested 700 W PSU, gets the stronger number.

But the RX 5600M is the only one of the two that functions as a general-purpose graphics product. It is an integrated mobile part with a 150 W TDP, no power connectors, display outputs that depend on the host device, and a modern PCIe 4.0 x16 link. The CMP 90HX has no display outputs at all, a PCIe 1.0 x4 bus, and end-of-life status. For anything involving a screen, the data effectively disqualifies it despite the stronger benchmark number.

Notably, neither card has a recorded launch MSRP in the database, so no comparison can be made on that axis.

Specification Differences

The fields where these two differ, per the recorded data:

• Chip: GA102 (CMP 90HX) versus Navi 10 (RX 5600M)
• Architecture: Ampere versus RDNA 1.0
• Generation: Mining GPUs versus Navi Mobile (RX 5000M)
• Process node and foundry: 8 nm at Samsung versus 7 nm at TSMC
• Transistors, die size, density: 28,300 million on 628 mm² at 45.1M/mm² versus 10,300 million on 251 mm² at 41.0M/mm²
• Clocks: 1500/1710 MHz base/boost versus 1035/1265 MHz base/boost with a 1190 MHz game clock
• Memory: 10 GB GDDR6X on a 320-bit bus at 760.3 GB/s versus 6 GB GDDR6 on a 192-bit bus at 288.0 GB/s; 19 Gbps effective versus 12 Gbps effective memory speed
• Shader resources: 6400 shading units, 200 TMUs, 80 ROPs, 50 RT cores, 200 tensor cores versus 2304 shading units, 144 TMUs, 64 ROPs, and no RT or tensor cores listed
• Throughput: 136.8 GPixel/s and 342.0 GTexel/s versus 80.96 GPixel/s and 182.2 GTexel/s
• FP32/FP16: 21.89/21.89 TFLOPS (1:1) versus 5.829/11.66 TFLOPS (2:1)
• Power and form factor: 320 W TDP, dual-slot, 2x 8-pin connectors, 700 W suggested PSU, 285 mm long by 112 mm tall versus 150 W TDP, IGP slot width, no power connectors, no suggested PSU listed
• Bus: PCIe 1.0 x4 versus PCIe 4.0 x16
• Display: none versus portable device dependent
• DirectX: 12 Ultimate (12_2) versus 12 (12_1)
• Release dates: July 27, 2021 versus July 6, 2020; both end-of-life

Where Each One Wins

The CMP 90HX wins wherever brute compute and memory throughput decide the outcome. Its Geekbench OpenCL score of 69000 is 15.8 percent clear of the RX 5600M, its bandwidth advantage of 760.3 GB/s to 288.0 GB/s is enormous, and its place among rivals like the Arc A770, Radeon Instinct MI25, Radeon Pro WX 8200, and Quadro P6000 shows it performing in professional-compute company. Workloads that lean on FP32 throughput, wide memory buses, or large memory footprints favor it decisively.

The RX 5600M wins on suitability, and on one recorded number: its Geekbench Metal score of 76653 is the highest single benchmark entry between the two cards, albeit in a test without a CMP 90HX comparison. It also wins on efficiency-oriented design choices evident in the data: a 150 W TDP against 320 W, no external power connectors, a modern PCIe 4.0 x16 interface, and display capability. Its rival cluster, containing the Arc A530M, RX 6550M, RTX A2000, and RTX 5880 Ada Generation, places it as a mid-range mobile performer in the 85th percentile.

The open question the data raises is whether that 15.8 percent OpenCL gap would hold across other test types. With only one shared benchmark recorded, the honest read is that the CMP 90HX leads where measured, and the RX 5600M remains the only practical choice of the two for anything resembling a conventional graphics use case.