NVIDIA GeForce GTX 1630 vs NVIDIA GeForce RTX 2070 SUPER Comparison

The NVIDIA GeForce GTX 1630 and the NVIDIA GeForce RTX 2070 SUPER make for one of the stranger head-to-head pairings in the database, because on paper they belong to entirely different weight classes, yet the recorded benchmark data contains a quirk worth investigating. Both cards are end-of-life products built on the same Turing architecture and the same 12 nm TSMC process, and that is where most of the similarity ends. The question this comparison raises is not really which card wins, since the RTX 2070 SUPER sweeps every shared benchmark, but how a card with roughly a quarter of the compute throughput in the shared tests still sits at the 70th percentile of all GPUs in the database while its dominant rival sits at the 65th. That discrepancy deserves a closer look at what the scores actually measure.

Head-to-Head Benchmarks

Only two tests appear in both cards' result sets, and the RTX 2070 SUPER wins both by enormous margins. In Geekbench OpenCL, the GTX 1630 posts a score of 24858 against 83358 for the RTX 2070 SUPER, a delta of 70.2 percent in the SUPER's favor. Geekbench Vulkan is even more lopsided: 23695 versus 90637, a 73.9 percent gap. In practical terms, the RTX 2070 SUPER delivers close to four times the GTX 1630's throughput in these compute-oriented workloads.

The scale of the gap tracks with the raw specifications. The RTX 2070 SUPER manages 9.062 TFLOPS of FP32 compute against 1.828 TFLOPS for the GTX 1630, and its texture rate of 283.2 GTexel/s is roughly five times the GTX 1630's 57.12 GTexel/s. Pixel throughput tells a similar story: 113.3 GPixel/s versus 28.56 GPixel/s. Given those ratios, benchmark deltas in the 70 percent range are exactly what the hardware predicts.

What complicates the picture is the percentile data. The GTX 1630, with zero head-to-head wins and an average benchmark score of 24277, is recorded at the 70th percentile versus all GPUs in the database. The RTX 2070 SUPER, with an average score of 20282 and two wins, sits at the 65th percentile. The likely explanation is that the percentile calculation reflects the composition of each card's benchmark portfolio rather than a direct strength comparison: the GTX 1630's average is drawn from only two Geekbench tests, while the RTX 2070 SUPER's average spans ten tests including several Passmark DirectX subtests with small raw values, such as 67 in Passmark DirectX 12 and 132 in Passmark DirectX 10, plus a 1651 in 3DMark Steel Nomad DX12. Averages across such different test mixes are not directly comparable, and the percentile inversion is a symptom of that.

The rivals lists reinforce this. The GTX 1630's nearest rivals by average score include the RTX 2080 SUPER (avgScore 24170, a delta of just 0.4 percent) and the AMD Radeon RX 6600 XT (24442, minus 0.7 percent), cards that would comfortably outperform it in gaming. Meanwhile the RTX 2070 SUPER's nearest rivals, the Quadro M4000M, RTX 3070 Mobile, Intel Arc B570, and Arc A750, all cluster within 1.5 percent of its average. The GTX 1630's rival clustering appears to be an artifact of its thin, two-test record, not evidence of real-world parity with an RTX 2080 SUPER.

Architecture Differences

Both GPUs are Turing-family parts fabricated by TSMC on a 12 nm process, but they occupy opposite ends of that architecture's range. The GTX 1630 uses the TU117 chip with 4,700 million transistors packed into a 200 mm² die, giving a density of 23.5M transistors per mm². The RTX 2070 SUPER uses the much larger TU104, with 13,600 million transistors across 545 mm² at 25.0M per mm². Nearly three times the transistor budget translates directly into execution resources: 2560 shading units, 160 TMUs, and 64 ROPs on the TU104, versus 512 shading units, 32 TMUs, and 16 ROPs on the TU117.

The feature gap is equally stark. The RTX 2070 SUPER carries 40 RT cores and 320 tensor cores, hardware the GTX 1630 lacks entirely, with its RT core and tensor core fields recorded as null. This is why the SUPER supports DirectX 12 Ultimate (12_2) while the GTX 1630 stops at DirectX 12 (12_1), the difference covering ray-tracing and related feature levels. Both support OpenGL 4.6 and Vulkan 1.4.

Memory subsystems diverge sharply. Both use GDDR6, but the GTX 1630 pairs 4 GB with a 64 bit bus running at 1500 MHz (12 Gbps effective) for 96.00 GB/s of bandwidth. The RTX 2070 SUPER pairs 8 GB with a 256 bit bus at 1750 MHz (14 Gbps effective) for 448.0 GB/s, more than 4.5 times the bandwidth plus double the capacity. Clocks are surprisingly close: the GTX 1630 actually boosts higher on paper at 1785 MHz versus 1770 MHz, with base clocks of 1740 MHz against 1605 MHz. Core count, not frequency, is what separates them.

Power and physical profiles differ just as much. The GTX 1630 is a 75 W, single-slot card needing no auxiliary power connectors, 145 mm long, 69 mm tall, and 18 mm wide, with a suggested PSU of 250 W. The RTX 2070 SUPER is a 215 W dual-slot card requiring one 6-pin and one 8-pin connector, measuring 267 mm by 116 mm by 35 mm, with a 550 W suggested PSU. Display output counts favor the SUPER: one HDMI 2.0, three DisplayPort 1.4a, and one USB Type-C, against the GTX 1630's DVI, single HDMI 2.0, and single DisplayPort 1.4a. Both use PCIe 3.0 x16.

FAQ

Q: Which GPU wins the shared benchmarks?

A: The RTX 2070 SUPER wins both, taking Geekbench OpenCL 83358 to 24858 (a 70.2 percent gap) and Geekbench Vulkan 90637 to 23695 (73.9 percent). The head-to-head record is two wins to zero.

Q: Why does the GTX 1630 show a higher percentile than the RTX 2070 SUPER despite losing every shared test?

A: The GTX 1630 sits at the 70th percentile versus all GPUs and the SUPER at the 65th, but their averages come from different test mixes. The GTX 1630's 24277 average is built from just two Geekbench results, while the SUPER's 20282 average spans ten tests including low-valued Passmark DirectX subtests. The inversion reflects portfolio composition, not superior performance.

Q: Can either card run DirectX 12 Ultimate titles with full feature support?

A: Only the RTX 2070 SUPER. It reports DirectX 12 Ultimate (12_2) with 40 RT cores and 320 tensor cores. The GTX 1630 reports DirectX 12 (12_1) and has no RT or tensor hardware.

Q: Do these cards need different power supplies?

A: Yes. The GTX 1630 has a 75 W TDP, no power connectors, and a 250 W suggested PSU. The RTX 2070 SUPER has a 215 W TDP, one 6-pin plus one 8-pin connector, and a 550 W suggested PSU.

Q: How much memory bandwidth does each card have?

A: The GTX 1630 provides 96.00 GB/s across a 64 bit bus with 4 GB of GDDR6. The RTX 2070 SUPER provides 448.0 GB/s across a 256 bit bus with 8 GB of GDDR6.

Q: Are both cards still in production?

A: No. Both are listed as end-of-life. The RTX 2070 SUPER launched on July 8, 2019 with a launch MSRP of 499 USD, and the GTX 1630 followed on June 27, 2022.

The Verdict

The data leaves no ambiguity about performance. Any workload captured in the shared benchmarks favors the RTX 2070 SUPER by roughly 70 percent or more, and its specification sheet, with five times the shading units, double the memory, more than four times the bandwidth, and dedicated RT and tensor hardware, explains why. Users choosing on capability should pick the RTX 2070 SUPER in every scenario the database measures.

The GTX 1630's case rests entirely on constraints the benchmarks do not score. It draws 75 W with no auxiliary connectors, fits a single slot, and measures only 145 mm in length, making it viable in compact or power-limited systems where a 267 mm, dual-slot, 215 W card with two power connectors simply cannot go. For buyers whose limiting factor is the chassis or the PSU rather than the desire for frame rates, the GTX 1630 is the only one of the two that fits the requirement. Everyone else should take the SUPER.

Specification Differences

• Chip: TU117 versus TU104
• Generation: GeForce 16 versus GeForce 20 (series: GeForce 10-series listing versus GeForce 20-series)
• Transistors: 4,700 million versus 13,600 million
• Die size: 200 mm² versus 545 mm²
• Transistor density: 23.5M/mm² versus 25.0M/mm²
• Base clock: 1740 MHz versus 1605 MHz
• Boost clock: 1785 MHz versus 1770 MHz
• Memory size: 4 GB versus 8 GB
• Bus width: 64 bit versus 256 bit
• Memory speed: 1500 MHz, 12 Gbps effective versus 1750 MHz, 14 Gbps effective
• Bandwidth: 96.00 GB/s versus 448.0 GB/s
• Shading units: 512 versus 2560
• TMUs: 32 versus 160
• ROPs: 16 versus 64
• RT cores: none versus 40
• Tensor cores: none versus 320
• Pixel rate: 28.56 GPixel/s versus 113.3 GPixel/s
• Texture rate: 57.12 GTexel/s versus 283.2 GTexel/s
• FP32: 1.828 TFLOPS versus 9.062 TFLOPS
• FP16: 3.656 TFLOPS versus 18.12 TFLOPS
• TDP: 75 W versus 215 W
• Slot width: Single-slot versus Dual-slot
• Power connectors: None versus 1x 6-pin + 1x 8-pin
• Suggested PSU: 250 W versus 550 W
• Display outputs: 1x DVI, 1x HDMI 2.0, 1x DisplayPort 1.4a versus 1x HDMI 2.0, 3x DisplayPort 1.4a, 1x USB Type-C
• DirectX support: 12 (12_1) versus 12 Ultimate (12_2)
• Dimensions: 145 x 69 x 18 mm versus 267 x 116 x 35 mm
• Release date: June 27, 2022 versus July 8, 2019
• Successor: GeForce 20 versus GeForce 30

Both share the Turing architecture, TSMC fabrication, a 12 nm node, GDDR6 memory, PCIe 3.0 x16, OpenGL 4.6, Vulkan 1.4, and end-of-life status.

Where Each One Wins

The RTX 2070 SUPER wins everywhere the database measures performance: Geekbench OpenCL by 70.2 percent, Geekbench Vulkan by 73.9 percent, and by extension every workload that scales with its 9.062 TFLOPS of FP32 throughput, its 448.0 GB/s of bandwidth, its 8 GB frame buffer, or its RT and tensor cores for DirectX 12 Ultimate features. Compute, modern gaming, higher-resolution texture work, and any task where memory capacity matters all fall to the SUPER decisively.

The GTX 1630 wins on nothing benchmarked, and that result should be stated plainly. Its victories are physical and electrical: a 75 W envelope, zero power connectors, a single-slot 18 mm width, and a 145 mm length. In small-form-factor builds or machines with a 250 W-class PSU and no spare cables, it is the only option of the two that can be installed at all. That is the entire case for it, and given the benchmark record, it is a case that should only be made when those constraints are binding.