AMD Radeon 680M vs AMD Radeon 860M Comparison

Head-to-Head Benchmarks

The recorded data includes two direct benchmark comparisons between the AMD Radeon 680M and the AMD Radeon 860M. The first test, Geekbench OpenCL, shows the older Radeon 680M ahead with a score of 23468 against 22759 for the 860M, a margin of 3.1%. This is a narrow win, indicating that in this particular compute workload the 680M’s raw shader throughput holds a slight edge. The 680M uses 768 shading units at a base clock of 2000 MHz and a boost of 2200 MHz, while the 860M operates with 512 shading units at a much lower base of 600 MHz but a substantially higher boost of 3000 MHz. In OpenCL, the 680M’s higher base clock and larger shader count appear to compensate for the 860M’s boost advantage.

The second test, Geekbench Vulkan, reverses the outcome decisively. The 860M scores 30043 against 21965 for the 680M, a delta of 26.9% in favor of the newer part. This is a large gap, and it suggests that the 860M’s architecture, RDNA 3.5, extracts significantly more performance from the Vulkan API than the RDNA 2.0 implementation in the 680M. The 860M also benefits from a 3000 MHz boost clock, which is 800 MHz higher than the 680M’s 2200 MHz boost. Even with fewer shading units (512 vs 768), TMUs (32 vs 48), and ROPs (16 vs 32), the 860M delivers a much stronger Vulkan result.

Looking at average benchmark scores across all recorded tests, the 860M sits at 26401, while the 680M averages 15270. That is a 72.9% higher average for the 860M, though the comparison is complicated by the fact that the 680M has an additional test in its benchmark set, 3DMark Steel Nomad DX12, where it scores 378. The 860M has no recorded score for that test. The average score field aggregates only the tests available for each GPU, so the 680M’s average includes a low DX12 result that drags its mean down. Even so, the Vulkan result alone shows a substantial performance lead for the 860M.

The percentile data places the 680M at the 57th percentile of all GPUs, while the 860M reaches the 72nd percentile. This means the 860M ranks higher relative to the broader GPU landscape, despite having fewer execution resources. The nearest rivals for the 680M include the NVIDIA GeForce GTX 580 (average score 15283, delta -0.1%), the NVIDIA GeForce RTX 2060 (15290, -0.1%), the NVIDIA GeForce RTX 3050 OEM (15199, +0.5%), and the AMD Radeon RX 7600 (15171, +0.7%). The 680M is effectively clustered with these desktop-oriented GPUs, all within 0.7% of each other. For the 860M, the nearest rivals are the NVIDIA GeForce MX550 (26421, -0.1%), the NVIDIA GeForce RTX 5060 (26331, +0.3%), the AMD Radeon RX 5700 XT 50th Anniversary (26553, -0.6%), and the NVIDIA RTX A4000 (26683, -1.1%). The 860M’s average score places it in a higher performance tier altogether, competing with dedicated discrete GPUs rather than older integrated graphics.

The head-to-head win count is even at one apiece. The 680M takes the OpenCL test, the 860M takes the Vulkan test. The magnitude of the Vulkan win, however, is much larger than the OpenCL win. A 26.9% advantage is not a marginal difference; it indicates a fundamental improvement in how the newer architecture handles the Vulkan workload. The 3.1% OpenCL lead for the 680M is minor and falls within the range of normal test variance for GPUs with different driver optimizations.

The Verdict

The benchmark data shows two different strengths. The AMD Radeon 680M, released in January 2023, is built on RDNA 2.0 using a 6 nm TSMC process. It has 768 shading units, 48 TMUs, 32 ROPs, and 12 ray tracing cores. The AMD Radeon 860M, released in February 2025, uses RDNA 3.5 on a 4 nm TSMC process, with 512 shading units, 32 TMUs, 16 ROPs, and 8 ray tracing cores. The 860M has fewer of every execution unit, yet it wins the Vulkan benchmark by a wide margin and holds the higher overall percentile ranking.

For users prioritizing Vulkan-based workloads or modern graphics APIs, the 860M is the clear choice based on the recorded data. Its 30043 Vulkan score versus 21965 for the 680M is a decisive gap. The 860M also carries a 72nd percentile ranking versus the 680M’s 57th, and its average benchmark score of 26401 is far above the 680M’s 15270. The 860M’s nearest rival list includes the NVIDIA RTX A4000, a professional discrete GPU, and the AMD Radeon RX 5700 XT 50th Anniversary, both desktop-class parts, which shows the 860M is punching well above its integrated graphics classification.

The 680M still holds relevance for OpenCL compute tasks, where its 3.1% advantage over the 860M is measurable. It also has a higher texture rate (105.6 GTexel/s vs 96.00 GTexel/s) and a higher pixel rate (70.40 GPixel/s vs 48.00 GPixel/s), along with a higher FP32 throughput at 3.379 TFLOPS versus 3.072 TFLOPS. These raw throughput numbers explain the OpenCL result. The 680M also has a lower TDP of 50 W compared to the 860M’s 15 W, meaning the 860M delivers its higher Vulkan performance while drawing less power. The 860M’s FP16 throughput is identical to its FP32 at 3.072 TFLOPS with a 1:1 ratio, whereas the 680M achieves 6.758 TFLOPS FP16 with a 2:1 ratio. This suggests the 680M has a dedicated FP16 path that the 860M does not, which may matter for specific compute workloads.

The data does not support a blanket winner. The 860M is the better overall GPU for modern graphics workloads, particularly those using Vulkan, and it ranks higher in the global percentile distribution. The 680M remains competitive in OpenCL and offers higher raw throughput metrics. The choice depends on the workload.

Architecture Differences

The 680M is based on RDNA 2.0, the same architecture used in the earlier Navi II IGP generation. Its chip is Rembrandt+, manufactured on a 6 nm process at TSMC. The 860M moves to RDNA 3.5, a newer architecture generation, and uses the Krackan Point chip on a 4 nm TSMC process. The process node shrink from 6 nm to 4 nm is a significant architectural change, allowing higher clock speeds despite lower power consumption.

The 680M has 13,100 million transistors on a 208 mm² die, giving a transistor density of 63.0 million per square millimeter. The 860M’s transistor count and die size are listed as unknown, so no direct density comparison is possible. The 680M’s larger die and higher transistor count reflect its older, less dense process. The 860M’s 4 nm node likely contributes to its ability to reach a 3000 MHz boost clock while maintaining a 15 W TDP.

Clock behavior differs sharply. The 680M has a base clock of 2000 MHz and a boost of 2200 MHz. The 860M has a base of 600 MHz and a boost of 3000 MHz. This is a striking configuration: the 860M starts much lower but boosts to a far higher frequency. The 680M’s narrower clock range suggests a design tuned for sustained moderate frequencies, while the 860M relies on aggressive boosting under load.

The execution unit counts differ across all categories. The 680M has 768 shading units, 48 TMUs, 32 ROPs, and 12 ray tracing cores. The 860M has 512 shading units, 32 TMUs, 16 ROPs, and 8 ray tracing cores. The 680M leads in every count, yet the 860M wins the Vulkan benchmark. This indicates that RDNA 3.5 delivers higher per-unit efficiency, likely through architectural improvements in instruction scheduling, cache behavior, or shader compiler utilization.

Both GPUs support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. Both use a PCIe 4.0 x8 bus interface and have no power connectors, consistent with their IGP classification. Display outputs are listed as portable device dependent for both, meaning the actual connectors depend on the host laptop. Memory is system shared for both, with bus width and bandwidth listed as system dependent. Neither GPU has dedicated VRAM; both rely on the host system’s memory.

The 860M’s FP16 throughput is exactly equal to its FP32 at 3.072 TFLOPS with a 1:1 ratio. The 680M’s FP16 is 6.758 TFLOPS with a 2:1 ratio, meaning it can process two FP16 operations per FP32 operation. This architectural difference suggests the 680M has a separate fast FP16 path, while the 860M treats FP16 and FP32 with equal throughput. For workloads that leverage FP16, the 680M may have an advantage despite its lower overall FP32.

Specification Differences

The two GPUs differ in several recorded specifications. The 680M uses the Rembrandt+ chip, while the 860M uses the Krackan Point chip. The architecture differs: RDNA 2.0 for the 680M, RDNA 3.5 for the 860M. The process node is 6 nm for the 680M and 4 nm for the 860M, both at TSMC.

Transistor count is 13,100 million for the 680M and unknown for the 860M. Die size is 208 mm² for the 680M and unknown for the 860M. Transistor density is 63.0 million per square millimeter for the 680M, with no figure for the 860M.

Base clock is 2000 MHz for the 680M and 600 MHz for the 860M. Boost clock is 2200 MHz for the 680M and 3000 MHz for the 860M. Shading units are 768 versus 512. TMUs are 48 versus 32. ROPs are 32 versus 16. Ray tracing cores are 12 versus 8.

Pixel rate is 70.40 GPixel/s for the 680M and 48.00 GPixel/s for the 860M. Texture rate is 105.6 GTexel/s for the 680M and 96.00 GTexel/s for the 860M. FP32 throughput is 3.379 TFLOPS for the 680M and 3.072 TFLOPS for the 860M. FP16 throughput is 6.758 TFLOPS (2:1) for the 680M and 3.072 TFLOPS (1:1) for the 860M.

TDP is 50 W for the 680M and 15 W for the 860M. Release dates differ: January 2023 for the 680M, February 2025 for the 860M. The 680M’s predecessor is listed as Vega II IGP and its successor as Navi III IGP. The 860M’s predecessor is Navi II IGP, with no successor listed. Neither GPU has a recorded launch MSRP, so no pricing data is available.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The AMD Radeon 860M has an average benchmark score of 26401, while the AMD Radeon 680M averages 15270.

Q: How large is the Vulkan performance difference?

A: The 860M scores 30043 in Geekbench Vulkan, while the 680M scores 21965. The 860M leads by 26.9%.

Q: Does the 680M win any benchmark?

A: Yes, the 680M wins the Geekbench OpenCL test with a score of 23468 versus 22759 for the 860M, a 3.1% advantage.

Q: What are the TDP values for each GPU?

A: The 680M has a TDP of 50 W, and the 860M has a TDP of 15 W.

Q: Which GPU has more shading units?

A: The 680M has 768 shading units, while the 860M has 512 shading units.

Q: What is the percentile ranking for each GPU?

A: The 680M is at the 57th percentile of all GPUs, and the 860M is at the 72nd percentile.

Where Each One Wins

The AMD Radeon 680M wins in OpenCL compute workloads based on the recorded Geekbench OpenCL score. Its higher FP32 throughput (3.379 TFLOPS), higher texture rate (105.6 GTexel/s), and higher pixel rate (70.40 GPixel/s) support this result. The 680M also offers a 2:1 FP16 ratio, reaching 6.758 TFLOPS, which could benefit applications that use half-precision arithmetic. Its larger execution unit counts, 768 shading units, 48 TMUs, and 32 ROPs, provide more parallel resources for compute-heavy tasks. The 680M’s 50 W TDP is higher, but for systems that can supply that power, it translates into sustained throughput in OpenCL.

The AMD Radeon 860M wins decisively in Vulkan workloads, with a 26.9% lead over the 680M. Its 30043 Geekbench Vulkan score places it in a higher performance tier, as reflected in its 72nd percentile ranking. The 860M’s 3000 MHz boost clock is the highest clock figure in either GPU’s specification, and the RDNA 3.5 architecture appears to extract more performance per shading unit than RDNA 2.0. The 860M also delivers this performance at a 15 W TDP, which is one-third of the 680M’s power draw. For thin-and-light laptops or systems with limited cooling, the 860M’s efficiency advantage is substantial.

The 860M’s nearest rival list includes the NVIDIA GeForce MX550 (delta -0.1%), the NVIDIA GeForce RTX 5060 (delta +0.3%), the AMD Radeon RX 5700 XT 50th Anniversary (delta -0.6%), and the NVIDIA RTX A4000 (delta -1.1%). These are discrete GPUs, several of them desktop-class, which indicates the 860M competes beyond the integrated graphics segment. The 680M’s nearest rivals, including the GTX 580, RTX 2060, RTX 3050 OEM, and RX 7600, are also discrete GPUs, but they cluster within a much narrower score band around 15200 to 15290. The 680M sits at the upper end of that cluster with a 0.7% delta over the RX 7600.

For 3DMark Steel Nomad DX12, only the 680M has a recorded score of 378. The 860M has no entry for that test, so no direct comparison is possible. The 680M’s inclusion of this test in its benchmark set lowers its average score, since 378 is far below its OpenCL and Vulkan results. The 860M’s average is computed from only two tests, both of which are higher than the 680M’s corresponding scores except for OpenCL.

In terms of raw throughput specifications, the 680M leads in pixel rate, texture rate, FP32, and FP16. The 860M leads in boost clock and efficiency, as measured by TDP. The 860M’s 1:1 FP16 ratio means it does not accelerate half-precision beyond full-precision rates, which is a notable difference from the 680M’s 2:1 ratio. Applications that rely on FP16 throughput would favor the 680M, while applications that scale with Vulkan efficiency and high boost clocks would favor the 860M.

The data indicates the 860M is the stronger all-around GPU for modern graphics workloads, with a higher average score, higher percentile, and a large Vulkan advantage. The 680M remains relevant for OpenCL-specific tasks and for workloads that benefit from its higher raw throughput figures and FP16 acceleration. Both GPUs are active in production, and both are classified as IGPs with no power connectors and system-shared memory. The 860M’s newer architecture and lower power draw make it the more advanced part, but the 680M’s specific wins in OpenCL and FP16 capability prevent a complete sweep.