AMD Radeon HD 8790M vs NVIDIA Quadro K620 Comparison

Head-to-Head Benchmarks

The database records two synthetic benchmark comparisons between the NVIDIA Quadro K620 and the AMD Radeon HD 8790M, and the results split decisively by workload type. In the Geekbench OpenCL test, the NVIDIA Quadro K620 posts a score of 6693 against the AMD Radeon HD 8790M's 5017, a commanding 33.4% advantage. This is the larger of the two deltas and represents the clearest single-workload victory for either card. The OpenCL result aligns with the Quadro K620's positioning as a professional compute-oriented product; its compute throughput is rated at 863.2 GFLOPS FP32, which is notably higher than the AMD part's 691.2 GFLOPS FP32. The gap in raw shading power, combined with the NVIDIA card's higher clock speeds (1058 MHz base versus 850 MHz base), helps explain why the OpenCL score leans so heavily in one direction.

The reverse is true in the Geekbench Vulkan test, where the AMD Radeon HD 8790M takes the win with a score of 6365 against the Quadro K620's 5870. The margin here is 7.8% in AMD's favor, a smaller but still meaningful edge. Vulkan is a lower-level API that can expose architectural efficiencies differently than OpenCL, and the GCN 1.0 design of the AMD chip appears to handle this particular workload more effectively. Interestingly, the AMD card's memory subsystem is substantially faster on paper: it uses GDDR5 at 64.00 GB/s bandwidth, while the Quadro K620 uses DDR3 at 28.80 GB/s. That doubling of memory bandwidth may be a contributing factor in the Vulkan result, as that API can be sensitive to memory latency and throughput in certain scenes.

Looking at the broader context, both cards sit in the lower-middle range of the database's performance distribution. The Quadro K620 holds a 36th percentile ranking across all GPUs, while the AMD Radeon HD 8790M sits at the 33rd percentile. Their average benchmark scores are 6282 and 5691 respectively, meaning the NVIDIA card is roughly 10.4% ahead on the aggregate metric. Yet the head-to-head data shows a tie in terms of individual wins: each card claims one benchmark, and the deltas are asymmetric, with NVIDIA's OpenCL win being far more pronounced than AMD's Vulkan win.

The nearest rivals for each card further contextualize these results. The Quadro K620's average score of 6282 places it within 0.2% of the NVIDIA GeForce RTX 5070 Ti SUPER and the NVIDIA GeForce RTX 4070 Ti SUPER AD102, both of which average 6270. It also trails the AMD Radeon R7 M350 (6327) by 0.7% and the AMD Radeon Pro WX 4100 (6330) by 0.8%. These are tight margins, suggesting the Quadro K620 is competitive with a cluster of very different GPUs in the database's aggregate metric. For the AMD Radeon HD 8790M, an average score of 5691 puts it ahead of the NVIDIA Quadro M500M (5604) by 1.6%, while it sits just behind the Intel Iris Pro Graphics P6300 (5712) by 0.4%, the NVIDIA GeForce GTX 670MX (5721) by 0.5%, and the NVIDIA GeForce GTX 550 Ti (5731) by 0.7%. The AMD card's rival cluster is consistently lower-scoring than the Quadro K620's, reinforcing the overall performance hierarchy even as the Vulkan result complicates the picture.

The Verdict

The data presents a clear split recommendation. For users whose workloads emphasize OpenCL compute, the NVIDIA Quadro K620 is the unambiguous choice. Its 33.4% lead in that benchmark is substantial and consistent with its higher FP32 throughput, higher base and boost clocks, and larger transistor budget on the same 28 nm process. The Quadro K620 also holds the higher aggregate score (6282 versus 5691), the better percentile ranking (36th versus 33rd), and the more favorable position among its nearest rivals, all of which cluster at or above the 6270 mark. The NVIDIA card's professional lineage, with its Quadro branding and workstation-oriented feature set, further supports its suitability for compute-heavy tasks.

Conversely, the AMD Radeon HD 8790M is the better option for Vulkan-based workloads, where its 7.8% advantage is recorded. This is a narrower win, but it is a real one, and it suggests that the GCN 1.0 architecture has particular strengths in modern API environments. The AMD card's GDDR5 memory, with 64.00 GB/s of bandwidth versus the Quadro K620's 28.80 GB/s, is a significant architectural difference that likely contributes to this result. However, the overall benchmark picture favors the NVIDIA card, and the Vulkan margin alone does not offset the much larger OpenCL deficit.

From a pure performance standpoint, the Quadro K620 is the stronger all-around card. The data shows one dominant win for NVIDIA, one modest win for AMD, and a higher average score for NVIDIA. The card also offers a more complete feature package: it supports Vulkan 1.4 (versus 1.2.170 for AMD), has a higher pixel rate (17.98 GPixel/s versus 7.200 GPixel/s), and a higher texture rate (26.98 GTexel/s versus 21.60 GTexel/s). The AMD part does match the NVIDIA card in shading units (384 each) and TMUs (24 each), but its lower ROP count (8 versus 16) and lower clocks undercut its peak throughput in most measured scenarios.

The practical verdict is workload-dependent. Users running OpenCL-heavy applications, professional visualization tools, or any compute that leverages the Quadro's driver optimizations should select the NVIDIA Quadro K620. Users specifically targeting Vulkan performance, particularly in scenarios where memory bandwidth is a bottleneck, may find the AMD Radeon HD 8790M adequate, but the data does not support choosing it as a general-purpose performer. The aggregate benchmark score, the percentile ranking, and the head-to-head OpenCL margin all point to NVIDIA as the safer, faster default.

FAQ

Q: Which GPU wins in OpenCL performance?

A: The NVIDIA Quadro K620 wins decisively, scoring 6693 against the AMD Radeon HD 8790M's 5017, a 33.4% advantage.

Q: Which GPU wins in Vulkan performance?

A: The AMD Radeon HD 8790M wins, scoring 6365 against the NVIDIA Quadro K620's 5870, a 7.8% margin.

Q: What is the average benchmark score for each card?

A: The NVIDIA Quadro K620 averages 6282, while the AMD Radeon HD 8790M averages 5691, putting the NVIDIA card approximately 10.4% ahead.

Q: How do these cards rank against all GPUs in the database?

A: The NVIDIA Quadro K620 sits at the 36th percentile, while the AMD Radeon HD 8790M sits at the 33rd percentile.

Q: What are the closest rivals for each card by average score?

A: For the NVIDIA Quadro K620, the nearest rivals include the NVIDIA GeForce RTX 5070 Ti SUPER and NVIDIA GeForce RTX 4070 Ti SUPER AD102, both at 6270 (0.2% behind). For the AMD Radeon HD 8790M, the closest rival is the Intel Iris Pro Graphics P6300 at 5712 (0.4% ahead).

Q: Which card has higher memory bandwidth?

A: The AMD Radeon HD 8790M has higher memory bandwidth at 64.00 GB/s using GDDR5, versus the NVIDIA Quadro K620's 28.80 GB/s using DDR3.

Specification Differences

The two cards differ across most core specification fields. The NVIDIA Quadro K620 uses a GM107 chip based on the Maxwell architecture, while the AMD Radeon HD 8790M uses a Mars chip based on GCN 1.0. Both are fabricated on a 28 nm process at TSMC, but the transistor counts diverge sharply: the NVIDIA card packs 1,870 million transistors on a 148 mm² die, while the AMD card has 950 million transistors on a 77 mm² die. The resulting transistor densities are similar, at 12.6M / mm² for NVIDIA and 12.3M / mm² for AMD, but the absolute scale difference is significant.

Clock speeds also differ. The Quadro K620 runs at 1058 MHz base and 1124 MHz boost, while the HD 8790M runs at 850 MHz base and 900 MHz boost. Memory clocks are listed as 900 MHz with 1800 Mbps effective for the NVIDIA card, and 1000 MHz with 4 Gbps effective for the AMD card. Memory types differ as well: the NVIDIA card uses 2 GB of DDR3, while the AMD card uses 2 GB of GDDR5. Both have a 128 bit bus width, but bandwidth is 28.80 GB/s for NVIDIA and 64.00 GB/s for AMD.

The shading unit count is identical at 384, and TMUs match at 24, but ROPs differ: 16 for NVIDIA versus 8 for AMD. Pixel rate is 17.98 GPixel/s for NVIDIA versus 7.200 GPixel/s for AMD, and texture rate is 26.98 GTexel/s versus 21.60 GTexel/s. FP32 compute is 863.2 GFLOPS for NVIDIA versus 691.2 GFLOPS for AMD. The NVIDIA card has a TDP of 45 W and a suggested PSU of 200 W, while the AMD card's TDP and suggested PSU are not recorded.

Form factor and interfaces differ completely. The Quadro K620 is a single-slot card with a 160 mm (6.3 inches) length and 69 mm (2.7 inches) height, using a PCIe 2.0 x16 interface. The HD 8790M is an MXM Module using an MXM-A (3.0) interface with no recorded dimensions. Display outputs are 1x DVI and 1x DisplayPort 1.2 for NVIDIA, while the AMD card's outputs are listed as portable device dependent. API support also varies: NVIDIA lists DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.4; AMD lists DirectX 12 (11_1), OpenGL 4.6, and Vulkan 1.2.170. Both cards are end-of-life, with the NVIDIA card released later (2014-07-21 versus 2013-03-31).

Architecture Differences

The architectural divide is fundamental. NVIDIA's Quadro K620 is built on the Maxwell architecture, which succeeded the Kepler generation and is noted for its efficiency improvements. The chip is designated GM107, and the card's generation is listed as Quadro Kepler (Kx200), though the architecture itself is Maxwell. The AMD Radeon HD 8790M uses the GCN 1.0 architecture, with the Mars chip, and belongs to the Solar System (HD 8700M) generation. These are entirely different design philosophies: Maxwell focuses on per-clock efficiency and higher clocks, while GCN 1.0 emphasizes throughput and memory bandwidth.

The transistor budget reflects this. The NVIDIA chip uses 1,870 million transistors, nearly double the AMD chip's 950 million, on a die that is almost twice as large (148 mm² versus 77 mm²). This allows the Quadro K620 to maintain a higher shading unit count per clock and a deeper pipeline, which is visible in its higher pixel and texture rates. The AMD chip's smaller footprint and lower transistor count are offset by its GDDR5 memory, which provides more than double the bandwidth of the NVIDIA card's DDR3.

Cache and memory architecture differences are implied by the specifications. The AMD card's 64.00 GB/s bandwidth with GDDR5 is a significant advantage in memory-bound scenarios, and its 128 bit bus is the same width as NVIDIA's, but the memory type difference changes how efficiently that bus is utilized. The NVIDIA card's higher ROP count (16 versus 8) gives it a clear edge in pixel fill rate, which is reflected in the 17.98 GPixel/s versus 7.200 GPixel/s figures. Texture rate also favors NVIDIA, 26.98 GTexel/s versus 21.60 GTexel/s, despite the matching 24 TMUs, due to the clock advantage.

Feature support differs at the API level. The NVIDIA card supports Vulkan 1.4, while the AMD card supports Vulkan 1.2.170. DirectX support is nominally similar, with NVIDIA at 12 (11_0) and AMD at 12 (11_1), though the feature level difference is minor. Both support OpenGL 4.6. Neither card has ray tracing cores or tensor cores, and neither lists FP16 support. The NVIDIA card's predecessor is Quadro Fermi and its successor is Quadro Maxwell, while the AMD card's predecessor is London and its successor is Gem System, reflecting their respective product lineages.