NVIDIA CMP 90HX vs NVIDIA RTX 4000 SFF Ada Generation Comparison
NVIDIA CMP 90HX
RTX 4000 SFF Ada Generation
PERFORMANCE BENCHMARKS
Analysis: NVIDIA CMP 90HX vs NVIDIA RTX 4000 SFF Ada Generation
Where Each One Wins
The recorded data shows a clear split between these two NVIDIA cards, but it is not a split based on raw compute capability alone. The NVIDIA RTX 4000 SFF Ada Generation wins the only shared benchmark, Geekbench OpenCL, with a score of 124812 against the CMP 90HX's 69000. That is an 80.9% advantage, a decisive margin that places the RTX 4000 SFF firmly ahead in general-purpose compute workloads as measured by that test.
The RTX 4000 SFF also has a second benchmark result in the database, Geekbench Vulkan at 109364, which the CMP 90HX has no recorded result for. This means the Ada card can demonstrate graphics API performance, while the CMP 90HX cannot be evaluated on that metric at all. The CMP 90HX, by contrast, has only one recorded benchmark, and it loses that comparison by a wide margin.
Where the CMP 90HX might claim a theoretical edge is in raw peak throughput. Its FP32 rating of 21.89 TFLOPS exceeds the RTX 4000 SFF's 19.17 TFLOPS, a difference of roughly 14%. Similarly, its texture rate of 342.0 GTexel/s is higher than the 299.5 GTexel/s of the RTX 4000 SFF, and its pixel rate of 136.8 GPixel/s beats 99.84 GPixel/s. These are peak specification numbers, not measured benchmark results, so the database's actual test data favors the RTX 4000 SFF. The CMP 90HX also carries 10 GB of GDDR6X memory on a 320-bit bus, yielding 760.3 GB/s of bandwidth, versus 20 GB of GDDR6 on a 160-bit bus at 280.0 GB/s for the Ada card. Bandwidth is another peak spec, not a measured workload result.
The percentile data reinforces the overall positioning. The RTX 4000 SFF sits at the 95th percentile among all GPUs tracked in the database, with an average benchmark score of 117088. The CMP 90HX sits at the 90th percentile with an average of 69000. The Ada card's nearest rivals include the NVIDIA GB10 at 117393 (0.3% higher), the AMD Radeon PRO W7700 at 118976 (1.6% higher), the Tesla V100 SXM2 16 GB at 114395 (2.4% lower), and the RTX A5500 Mobile at 113944 (2.8% lower). The CMP 90HX's nearest rivals are the Intel Arc A770 at 68809 (0.3% higher), the AMD Radeon Instinct MI25 at 68562 (0.6% higher), the AMD Radeon Pro WX 8200 at 69870 (1.2% lower), and the NVIDIA Quadro P6000 at 69986 (1.4% lower). The RTX 4000 SFF competes in a higher performance tier, while the CMP 90HX clusters with older workstation and midrange cards.
Architecture Differences
The two cards come from different NVIDIA architectures built at different foundries. The RTX 4000 SFF Ada Generation uses the AD104 chip on the Ada Lovelace architecture, fabricated by TSMC on a 5 nm process. It packs 35,800 million transistors into a 294 mm² die, yielding a transistor density of 121.8M per mm². The CMP 90HX uses the GA102 chip on the Ampere architecture, fabricated by Samsung on an 8 nm process. It holds 28,300 million transistors across a much larger 628 mm² die, for a density of just 45.1M per mm². The process node difference is stark: 5 nm versus 8 nm.
The generation labels reinforce the positioning. The RTX 4000 SFF is listed under "Workstation Ada" generation, while the CMP 90HX belongs to "Mining GPUs". The RTX 4000 SFF has a predecessor in Workstation Ampere and a successor in Blackwell PRO W, so it sits inside a clear product lineage. The CMP 90HX has neither a predecessor nor a successor recorded in the database.
Core counts differ across every unit. The RTX 4000 SFF has 6144 shading units, 192 TMUs, 64 ROPs, 48 RT cores, and 192 tensor cores. The CMP 90HX has 6400 shading units, 200 TMUs, 80 ROPs, 50 RT cores, and 200 tensor cores. The CMP 90HX leads on every count, but the Ada card compensates with a smaller die and much higher transistor density.
The memory subsystems are fundamentally different. The RTX 4000 SFF uses 20 GB of GDDR6 on a 160-bit bus at 280.0 GB/s. The CMP 90HX uses 10 GB of GDDR6X on a 320-bit bus at 760.3 GB/s. The CMP 90HX has more than 2.7 times the bandwidth of the RTX 4000 SFF, but half the capacity. Clock behavior also differs. The RTX 4000 SFF runs a 720 MHz base and 1560 MHz boost, with memory at 1750 MHz (14 Gbps effective). The CMP 90HX runs a 1500 MHz base and 1710 MHz boost, with memory at 1188 MHz (19 Gbps effective). The CMP 90HX has a much higher base clock, more than double, and a higher boost clock.
Power delivery and physical design diverge sharply. The RTX 4000 SFF has a 70 W TDP, no power connectors, and a suggested PSU of 250 W. The CMP 90HX has a 320 W TDP, requires two 8-pin connectors, and asks for a 700 W suggested PSU. Both are dual-slot cards, but the RTX 4000 SFF measures 168 mm (6.6 inches) in length and 69 mm (2.7 inches) in height, while the CMP 90HX measures 285 mm (11.2 inches) in length and 112 mm (4.4 inches) in height. The CMP 90HX is substantially larger.
Display output is another major divergence. The RTX 4000 SFF provides four mini-DisplayPort 1.4a outputs, so it can drive professional workstation displays. The CMP 90HX has no display outputs at all. The bus interface also differs: the RTX 4000 SFF uses PCIe 4.0 x16, while the CMP 90HX uses PCIe 1.0 x4. The CMP 90HX's connectivity is severely limited. API support is identical, with both cards listing DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.
FAQ
Q: Which card is faster in the only benchmark they share?
A: The NVIDIA RTX 4000 SFF Ada Generation scores 124812 in Geekbench OpenCL, while the NVIDIA CMP 90HX scores 69000. The RTX 4000 SFF wins by 80.9%.
Q: Does the CMP 90HX have any performance advantage at all?
A: In peak specifications, yes. The CMP 90HX rates at 21.89 TFLOPS FP32 versus 19.17 TFLOPS for the RTX 4000 SFF, and it has higher texture (342.0 GTexel/s versus 299.5 GTexel/s) and pixel rates (136.8 GPixel/s versus 99.84 GPixel/s). The measured Geekbench OpenCL result, however, favors the RTX 4000 SFF by a wide margin.
Q: Why can the RTX 4000 SFF be used in a workstation environment but the CMP 90HX cannot?
A: The RTX 4000 SFF has four mini-DisplayPort 1.4a outputs. The CMP 90HX has no display outputs, making it unsuitable for any visual workstation task that requires a monitor connection.
Q: How do the memory configurations compare?
A: The RTX 4000 SFF has 20 GB of GDDR6 on a 160-bit bus with 280.0 GB/s bandwidth. The CMP 90HX has 10 GB of GDDR6X on a 320-bit bus with 760.3 GB/s bandwidth. The CMP 90HX offers more bandwidth, the RTX 4000 SFF offers double the capacity.
Q: Which card draws less power?
A: The RTX 4000 SFF has a 70 W TDP and requires no power connectors, with a 250 W suggested PSU. The CMP 90HX has a 320 W TDP, requires two 8-pin connectors, and asks for a 700 W suggested PSU.
Q: What is the production status of each card?
A: The RTX 4000 SFF Ada Generation is listed as Active, released on 2023-03-20. The CMP 90HX is listed as End-of-life, released on 2021-07-27.
Specification Differences
The two cards differ across nearly every recorded specification. The RTX 4000 SFF Ada Generation uses the AD104 chip on a 5 nm TSMC process with 35,800 million transistors on a 294 mm² die. The CMP 90HX uses the GA102 chip on an 8 nm Samsung process with 28,300 million transistors on a 628 mm² die. Transistor density is 121.8M per mm² for the Ada card versus 45.1M per mm² for the Ampere card.
Clock speeds differ substantially. The RTX 4000 SFF runs at 720 MHz base and 1560 MHz boost. The CMP 90HX runs at 1500 MHz base and 1710 MHz boost. Memory clocks are 1750 MHz (14 Gbps effective) for the Ada card and 1188 MHz (19 Gbps effective) for the Ampere card.
Memory capacity, type, bus width, and bandwidth all differ. The RTX 4000 SFF has 20 GB GDDR6 on a 160-bit bus at 280.0 GB/s. The CMP 90HX has 10 GB GDDR6X on a 320-bit bus at 760.3 GB/s.
Compute unit counts differ across the board. The RTX 4000 SFF has 6144 shading units, 192 TMUs, 64 ROPs, 48 RT cores, and 192 tensor cores. The CMP 90HX has 6400 shading units, 200 TMUs, 80 ROPs, 50 RT cores, and 200 tensor cores. Pixel rate is 99.84 GPixel/s for the Ada card and 136.8 GPixel/s for the Ampere card. Texture rate is 299.5 GTexel/s versus 342.0 GTexel/s. FP32 is 19.17 TFLOPS versus 21.89 TFLOPS, and FP16 matches at 19.17 TFLOPS (1:1) versus 21.89 TFLOPS (1:1).
Power and physical specs differ widely. The RTX 4000 SFF has a 70 W TDP, no power connectors, a 250 W suggested PSU, and dimensions of 168 mm by 69 mm. The CMP 90HX has a 320 W TDP, two 8-pin connectors, a 700 W suggested PSU, and dimensions of 285 mm by 112 mm. Both are dual-slot cards.
Connectivity differs completely. The RTX 4000 SFF uses PCIe 4.0 x16 and has four mini-DisplayPort 1.4a outputs. The CMP 90HX uses PCIe 1.0 x4 and has no display outputs. The API list is identical for both: DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.
Release and status information also differs. The RTX 4000 SFF was released on 2023-03-20 and is Active, with a predecessor in Workstation Ampere and a successor in Blackwell PRO W. The CMP 90HX was released on 2021-07-27 and is End-of-life, with no predecessor or successor recorded.
Head-to-Head Benchmarks
The database contains exactly one head-to-head benchmark result for these two cards, and it is decisive. In Geekbench OpenCL, the NVIDIA RTX 4000 SFF Ada Generation scores 124812 against the NVIDIA CMP 90HX's 69000. The delta is 80.9% in favor of the Ada card. That is not a narrow victory; it is a doubling of the CMP 90HX's measured performance in this workload.
The RTX 4000 SFF also has a Geekbench Vulkan result of 109364, which is close to its OpenCL score. The CMP 90HX has no Vulkan measurement in the database, so no comparison can be made on that API. The absence of a second benchmark for the CMP 90HX means the available record is one-sided.
Context from the nearest rivals helps interpret these scores. The RTX 4000 SFF's average benchmark score of 117088 places it just below the NVIDIA GB10 (117393, 0.3% higher) and the AMD Radeon PRO W7700 (118976, 1.6% higher), and above the Tesla V100 SXM2 16 GB (114395, 2.4% lower) and the RTX A5500 Mobile (113944, 2.8% lower). The CMP 90HX's average of 69000 sits just above the Intel Arc A770 (68809, 0.3% higher) and the AMD Radeon Instinct MI25 (68562, 0.6% higher), and just below the AMD Radeon Pro WX 8200 (69870, 1.2% lower) and the NVIDIA Quadro P6000 (69986, 1.4% lower). In other words, the RTX 4000 SFF operates in a performance class roughly 70% higher than the CMP 90HX based on average scores.
The peak specification gap tells a different story. The CMP 90HX leads in FP32 by about 14%, in texture rate by about 14%, in pixel rate by about 37%, and in memory bandwidth by about 171%. Yet the measured OpenCL result runs in the opposite direction. The explanation lies in the broader system context. The CMP 90HX is a mining-focused card with no display outputs, a PCIe 1.0 x4 bus interface, and an end-of-life status. The RTX 4000 SFF is an active workstation card with PCIe 4.0 x16, four display outputs, and a much lower power envelope. Benchmark workloads that depend on driver maturity, memory capacity, and system integration will favor the Ada architecture card, even when raw peak numbers favor the Ampere part.
The percentile rankings reinforce the measured hierarchy. The RTX 4000 SFF sits at the 95th percentile among all GPUs, while the CMP 90HX sits at the 90th. The Ada card's average benchmark score of 117088 is 48,088 points higher than the CMP 90HX's 69000, which is a 69.7% advantage. The single head-to-head result of 80.9% is even larger than the average-score gap, indicating that the OpenCL workload specifically favors the RTX 4000 SFF more than the average of all tracked benchmarks would suggest. The CMP 90HX's nearest rival cluster, with the Quadro P6000 and Radeon Pro WX 8200, reflects an older performance tier, while the RTX 4000 SFF's rivals, including the GB10 and Radeon PRO W7700, reflect a modern workstation tier. The data shows two cards from different eras and different purposes, and the measured results place the RTX 4000 SFF clearly ahead.