AMD Radeon RX 6700S vs NVIDIA GeForce RTX 2070 Comparison

The Radeon RX 6700S and GeForce RTX 2070 make for an unusual comparison: a mobile-focused RDNA 2 part squaring off against a desktop Turing card. On paper the 2070 looks dominant, and the recorded data largely confirms it, winning nine of ten head-to-head benchmarks. Yet the margins tell a more interesting story than a clean sweep, and the efficiency story underlying those numbers is where the RX 6700S becomes genuinely intriguing.

FAQ

Q: Which GPU wins more benchmarks?

A: The GeForce RTX 2070 wins 9 of the 10 recorded head-to-head tests. The Radeon RX 6700S takes exactly one: PassMark DirectX 11, where it scores 134 against 129, a 3.9% advantage.

Q: How close is the overall performance gap?

A: The RTX 2070's average benchmark score is 18789, which sits below the RX 6700S at 22154. That difference is driven partly by the RX 6700S's much higher Geekbench Metal score of 72864, a test the 2070 does not have a recorded entry for. In direct head-to-head comparisons, the 2070's typical winning margin lands between roughly 3% and 11%.

Q: Where does the RTX 2070 win by the largest margin?

A: PassMark DirectX 10, where the 2070 scores 111 versus 87, a 21.6% gap. That is by far the biggest delta in the dataset.

Q: How much power does each GPU use?

A: The RX 6700S is rated at 80 W TDP, while the RTX 2070 is rated at 175 W, more than double. The 2070 requires a 1x 8-pin power connector and a suggested 450 W PSU; the RX 6700S needs no power connector at all.

Q: Where does each GPU rank against all GPUs in the database?

A: The RX 6700S sits at the 67th percentile versus all GPUs; the RTX 2070 sits at the 63rd percentile. Despite losing most head-to-head tests, the RX 6700S ranks slightly higher on that aggregate measure.

Q: Do both support the same graphics APIs?

A: Yes. Both report DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4 support.

Architecture Differences

These two GPUs come from opposite design philosophies. The RX 6700S is built on the Navi 23 chip using AMD's RDNA 2.0 architecture on TSMC's 7 nm process, while the RTX 2070 uses the TU106 chip on Turing, manufactured on a 12 nm process at TSMC.

The process gap shows up vividly in the physical data. The RX 6700S packs 11,060 million transistors into a 237 mm² die, giving a density of 46.7M transistors per mm². The RTX 2070 carries 10,800 million transistors across a much larger 445 mm² die, at 24.3M per mm². Nearly the same transistor count in roughly half the area, and that is before considering the power story. This is the heart of the comparison: the RX 6700S achieves results within single-digit percentages of the 2070 in many tests while drawing 80 W against 175 W.

The architectural feature sets also diverge. The RTX 2070 includes 288 tensor cores alongside 36 RT cores; the RX 6700S has 28 RT cores and no listed tensor cores. NVIDIA's part carries dedicated machine learning hardware that AMD's design omits, which may explain part of the compute benchmark gap where the 2070 consistently leads.

Memory subsystems differ sharply too. Both use 8 GB of GDDR6 running at an effective 14 Gbps, but the 2070's 256-bit bus delivers 448.0 GB/s of bandwidth against the RX 6700S's 128-bit bus and 224.0 GB/s. Exactly half the bandwidth is a significant handicap the mobile AMD chip has to overcome, and mostly does.

Head-to-Head Benchmarks

Walking through the recorded results, the RTX 2070's dominance is broad but not deep. In the modern API tests, the 2070 scores 1569 in 3DMark Steel Nomad DX12 versus 1451, a 7.5% win. Geekbench Vulkan goes to the 2070 82521 to 76150 (7.7%), and Geekbench OpenCL by a wider 11.3% margin, 79966 to 70965.

The PassMark suite follows the same pattern. G3D score: 16094 to 15094, wait, 16094 to 15066, a 6.4% edge. GPU compute: 6411 to 5990, 6.6% in the 2070's favor. DirectX 9: 211 to 195 (7.6%). DirectX 12 is nearly a tie, 60 to 58, just 3.3% apart. G2D: 819 to 739 (9.8%).

Then there is the outlier. PassMark DirectX 11 is the RX 6700S's sole win, 134 to 129, a 3.9% margin. What makes this result worth questioning is its context: the 2070 wins DirectX 9 by 7.6%, DirectX 10 by 21.6%, and DirectX 12 by 3.3%, yet drops the middle API. Is this a driver quirk in older API paths, or sample noise? The database alone cannot say, but it does suggest the RX 6700S is not uniformly behind in rasterization workloads.

Also notable: the RX 6700S posts a Geekbench Metal score of 72864, and the 2070 has no recorded Metal result. That Apple-platform data point contributes to the RX 6700S's higher average benchmark score of 22154 versus 18789, and to its 67th percentile ranking versus the 2070's 63rd. Direct head-to-head results and aggregate database positioning point in different directions here, which is exactly why reading both matters.

Specification Differences

The fields where these GPUs differ:

• Process: 7 nm (RX 6700S) vs 12 nm (RTX 2070), both TSMC
• Die size and density: 237 mm² at 46.7M/mm² vs 445 mm² at 24.3M/mm²
• Shading units: 1792 vs 2304
• TMUs: 112 vs 144
• RT cores: 28 vs 36
• Tensor cores: none listed vs 288
• Clocks: the RX 6700S runs a 1700 MHz base, 1890 MHz game, 2000 MHz boost; the 2070 runs 1410 MHz base and 1620 MHz boost
• Memory bus and bandwidth: 128-bit, 224.0 GB/s vs 256-bit, 448.0 GB/s (same 8 GB GDDR6 at 14 Gbps effective)
• Bus interface: PCIe 4.0 x8 vs PCIe 3.0 x16
• TDP: 80 W vs 175 W
• Form factor: IGP-style mobile integration with no power connectors vs dual-slot desktop card with 1x 8-pin, 450 W suggested PSU, and dimensions of 229 mm length, 113 mm height, 35 mm width
• Outputs: portable-device-dependent vs 1x DVI, 1x HDMI 2.0, 2x DisplayPort 1.4a, 1x USB Type-C
• Release dates: 2022-01-03 vs 2018-10-16
• Pixel rate: 128.0 GPixel/s for the RX 6700S vs 103.7 GPixel/s for the 2070, a rare spec win for AMD driven by higher clocks
• Texture rate: 224.0 GTexel/s vs 233.3 GTexel/s
• FP32: 7.168 TFLOPS vs 7.465 TFLOPS; FP16: 14.34 vs 14.93 TFLOPS

Both are end-of-life products. The RTX 2070's launch MSRP was 499 USD.

The Verdict

Strictly from the recorded data, the RTX 2070 is the faster GPU in nine of ten head-to-head tests, typically by single-digit percentages, with one double-digit blowout in PassMark DirectX 10. If raw benchmark performance across DirectX 9, 12, Vulkan, OpenCL, and compute workloads is the deciding factor, the 2070 is the pick.

But the performance-per-watt picture inverts that conclusion. The RX 6700S delivers results within roughly 3 to 11% of the 2070 while operating at 80 W versus 175 W, with no power connector, on a die less than half the size. For anyone constrained by power delivery or platform integration, that trade is the entire argument. The RX 6700S also holds a higher database percentile (67 vs 63) and a higher average benchmark score (22154 vs 18789), buoyed by its Metal result.

The 2070 adds 288 tensor cores and double the memory bandwidth, plus a full desktop display output array. The RX 6700S counters with higher clock speeds, a newer 7 nm process, a PCIe 4.0 link, and a 2022 release against the 2070's 2018 origin.

Where Each One Wins

RTX 2070 territory: compute and OpenCL work (6411 vs 5990 in PassMark GPU compute, 79966 vs 70965 in Geekbench OpenCL), Vulkan rendering (82521 vs 76150), modern DirectX 12 gaming (Steel Nomad 1569 vs 1451; PassMark DX12 60 vs 58), legacy DirectX 9 and 10 titles (211 vs 195, and the 21.6% gulf of 111 vs 87), 2D workloads (819 vs 739), and general 3D throughput (PassMark G3D 16094 vs 15066). Its 448 GB/s of bandwidth and tensor core hardware back up these leads on paper.

RX 6700S territory: DirectX 11 rendering (134 vs 129), Apple Metal environments (72864, where the 2070 has no recorded result), and any scenario where the 80 W TDP, connector-free design, and half-size die matter more than the last few percentage points of benchmark performance. Its higher pixel fill rate of 128.0 GPixel/s versus 103.7 also stands out as a spec-level advantage.

The data frames this as a contest between a desktop card that wins almost everything and a mobile chip that gets remarkably close on a fraction of the power. Which one "wins" depends almost entirely on which of those constraints applies.