AMD Radeon HD 8970M vs AMD Radeon R9 M390X Comparison

The AMD Radeon HD 8970M and AMD Radeon R9 M390X are two mobile graphics solutions from different generations of AMD's product stack, separated by roughly two years of architectural evolution. The data shows both cards occupy the same performance percentile tier at 66, yet they achieve this through markedly different design philosophies and benchmark profiles. The HD 8970M, built on GCN 1.0, delivers a single OpenCL score of 21,237, while the R9 M390X, using GCN 3.0, posts an average benchmark score of 20,662 across Metal and Vulkan tests, with individual scores of 24,373 and 16,951 respectively. This variance in API-specific results tells a nuanced story about where each card excels and where it falls short.

Head-to-Head Benchmarks

The most direct comparison available comes from the average benchmark scores and their placement among nearest rivals. The HD 8970M averages 21,237, which places it 0.4% ahead of the AMD Radeon RX Vega M GL (21,153) and 1% ahead of the NVIDIA GeForce RTX 5050 (21,035). It trails the NVIDIA RTX A4000 Mobile (21,379) by 0.7% and the NVIDIA Quadro RTX 5000 (21,629) by 1.8%. This clustering indicates the HD 8970M sits in a tightly contested performance band where even small percentage differences separate competitors.

The R9 M390X, by contrast, averages 20,662, which is 0.2% behind the AMD Radeon RX 5600 XT (20,713) but 0.4% ahead of the Intel Arc A750 (20,582). It also edges out the Intel Arc B570 (20,556) by 0.5% and the NVIDIA GeForce RTX 3070 Mobile (20,534) by 0.6%. The R9 M390X's average is therefore 575 points lower than the HD 8970M's average, a gap of roughly 2.8%. However, this aggregate figure masks significant variance in the R9 M390X's individual results.

The R9 M390X's Metal score of 24,373 is its strongest showing, substantially higher than its Vulkan score of 16,951 — a difference of 7,422 points or approximately 43.8%. This wide spread suggests the card's performance is highly dependent on the API in use. The HD 8970M's lone OpenCL score of 21,237 falls between these two extremes, closer to the R9 M390X's Vulkan result than its Metal result. In raw compute terms, the R9 M390X's FP32 throughput is 2.961 TFLOPS versus the HD 8970M's 2.304 TFLOPS, a 28.5% advantage. Yet the benchmark averages tell a different story, with the HD 8970M holding a 2.8% lead in aggregated scores.

Where Each One Wins

The HD 8970M wins on consistency. Its single benchmark result of 21,237 in OpenCL represents a stable, predictable performance profile that aligns closely with its nearest rivals, all within a 3% band. This makes it a reliable choice for workloads that leverage OpenCL, where the data shows it performing on par with much newer discrete GPUs like the RTX 5050 and RTX A4000 Mobile. The card's 28.80 GPixel/s pixel rate and 72.00 GTexel/s texture rate indicate solid rasterization capabilities for its era, though these figures are not directly benchmarked.

The R9 M390X wins decisively in Metal-based applications. Its Metal score of 24,373 is 14.8% higher than the HD 8970M's OpenCL score, and 43.8% higher than its own Vulkan score. This suggests the card is optimized for Apple's Metal API, making it the stronger choice for macOS or Metal-centric rendering workloads. The R9 M390X also holds advantages in memory bandwidth (160.0 GB/s vs. 153.6 GB/s) and texture rate (92.54 GTexel/s vs. 72.00 GTexel/s), which could translate to better performance in texture-heavy scenes. However, its Vulkan result of 16,951 is 20.2% below the HD 8970M's OpenCL score, indicating a weakness in Vulkan-optimized applications.

For users prioritizing a single, dependable compute score, the HD 8970M's 21,237 is the higher number. For those targeting Metal-specific workloads, the R9 M390X's 24,373 is clearly superior. The R9 M390X's 75 W TDP also gives it a power efficiency edge over the HD 8970M's 100 W, though this is not reflected in benchmark scores.

Architecture Differences

The two cards represent a generational leap in AMD's graphics core architecture. The HD 8970M uses GCN 1.0, built on the Neptune chip, while the R9 M390X uses GCN 3.0, built on the Amethyst chip. This architectural shift brings significant changes in compute capabilities and feature support. The R9 M390X doubles the transistor count from 2,800 million to 5,000 million, and its die size grows from 212 mm² to 366 mm², resulting in a slightly higher transistor density of 13.7M per mm² versus 13.2M per mm².

The R9 M390X's GCN 3.0 architecture supports DirectX 12 (12_0), while the HD 8970M's GCN 1.0 only reaches DirectX 12 (11_1). Both cards support OpenGL 4.6 and Vulkan 1.2.170, indicating modern API compatibility despite their age. The R9 M390X also gains FP16 support with a 1:1 ratio (2.961 TFLOPS), whereas the HD 8970M has no listed FP16 capability. This makes the R9 M390X more suitable for workloads that leverage half-precision arithmetic, which can be critical in machine learning inference or certain graphics effects.

The R9 M390X's shader configuration is substantially larger: 2,048 shading units versus 1,280, and 128 texture mapping units versus 80. Both cards share 32 ROPs. The R9 M390X's FP32 throughput of 2.961 TFLOPS represents a 28.5% increase over the HD 8970M's 2.304 TFLOPS, yet its pixel rate is lower at 23.14 GPixel/s versus 28.80 GPixel/s. This suggests the HD 8970M has a more efficient ROP pipeline despite fewer compute units, possibly due to higher clock speeds.

Specification Differences

The specification sheets reveal several key differences beyond architecture. The HD 8970M has defined clock speeds of 850 MHz base and 900 MHz boost, while the R9 M390X lists no base or boost clocks in the data. Memory clocks differ slightly: the HD 8970M runs at 1200 MHz (4.8 Gbps effective), while the R9 M390X runs at 1250 MHz (5 Gbps effective). This translates to a bandwidth advantage for the R9 M390X at 160.0 GB/s versus 153.6 GB/s.

Both cards feature 4 GB of GDDR5 memory on a 256-bit bus, and both use MXM Module slot width with PCIe 3.0 x16 interfaces. Display outputs are "Portable Device Dependent" for both, meaning they rely on laptop implementations. The R9 M390X has no power connectors listed, while the HD 8970M's power connectors are also absent from the data. The R9 M390X's TDP is 75 W, notably lower than the HD 8970M's 100 W, indicating better power efficiency despite the larger die and higher transistor count.

The HD 8970M's generation is "Solar System (HD 8900M)" with predecessor "London" and successor "Gem System." The R9 M390X's generation is "Gem System (R9 M300)" with predecessor "Solar System" and successor "Polaris Mobile." This lineage confirms the direct generational relationship between the two cards. The HD 8970M was released on 2013-05-13, while the R9 M390X followed on 2015-05-04, a gap of 23 months. Both are marked as end-of-life production status. Neither card has a launch MSRP in the data.

FAQ

Q: Which card has the higher average benchmark score?

A: The AMD Radeon HD 8970M has an average benchmark score of 21,237, while the AMD Radeon R9 M390X averages 20,662. This gives the HD 8970M a 2.8% lead in aggregated performance.

Q: How do the two cards compare in terms of power consumption?

A: The R9 M390X has a TDP of 75 W, while the HD 8970M has a TDP of 100 W. The R9 M390X therefore consumes 25% less power, making it the more power-efficient option despite having a larger die.

Q: What is the best benchmark result for each card?

A: The HD 8970M's only listed result is 21,237 in Geekbench OpenCL. The R9 M390X's best result is 24,373 in Geekbench Metal, which is 14.8% higher than the HD 8970M's OpenCL score. Its Vulkan score is 16,951.

Q: Do both cards support the same DirectX version?

A: No. The R9 M390X supports DirectX 12 (12_0), while the HD 8970M supports DirectX 12 (11_1). Both cards support OpenGL 4.6 and Vulkan 1.2.170.

Q: Which card has more shading units?

A: The R9 M390X has 2,048 shading units, compared to the HD 8970M's 1,280. The R9 M390X also has 128 TMUs versus 80, and both have 32 ROPs.

Q: How does memory bandwidth differ between the two cards?

A: The R9 M390X has a bandwidth of 160.0 GB/s from its 1250 MHz memory clock (5 Gbps effective), while the HD 8970M has 153.6 GB/s from its 1200 MHz clock (4.8 Gbps effective). Both use 4 GB GDDR5 on a 256-bit bus.