NVIDIA GeForce RTX 3090 vs NVIDIA Tesla M60 Comparison

Head-to-Head Benchmarks

The benchmark data records only two shared tests between the NVIDIA Tesla M60 and the NVIDIA GeForce RTX 3090, and in both cases the RTX 3090 takes a commanding lead. The most decisive margin appears in Geekbench OpenCL, where the RTX 3090 scores 172,758 against the Tesla M60's 29,506. That is a delta of -82.9% from the RTX 3090's perspective, meaning the Tesla M60 trails by roughly 5.9x in raw compute throughput as measured by this workload. The gap is not subtle: the RTX 3090 delivers over 143,000 additional points in this single test, a difference larger than the Tesla M60's entire score.

The second shared benchmark, Geekbench Vulkan, narrows the margin but still favors the RTX 3090 decisively. Here the RTX 3090 posts 53,927 points versus the Tesla M60's 31,473, a delta of -41.6%. In percentage terms, the RTX 3090 is about 71% faster in this API workload. The Vulkan result is interesting because it shows the Tesla M60 closing some of the gap relative to the OpenCL test, suggesting that its Maxwell architecture handles the Vulkan driver path relatively better than OpenCL. Still, the absolute difference of 22,454 points remains substantial.

Looking at the broader database context, the Tesla M60's average benchmark score of 30,490 places it at the 75th percentile of all GPUs. Its nearest rivals are tightly clustered: the NVIDIA CMP 70HX scores 30,476 (a delta of 0%), the AMD Radeon RX 6700 scores 30,433 (0.2% behind), the AMD Radeon RX 6800 scores 30,095 (1.3% behind), and the NVIDIA GeForce RTX 3070 Ti scores 29,945 (1.8% behind). This means the Tesla M60, despite being an older enterprise card, sits at the top of its immediate competitive tier by a hair, though the margins are so small that they fall within measurement noise. Its 4.825 TFLOPS FP32 throughput and 160.4 GB/s memory bandwidth are evidently sufficient to keep it competitive with much newer consumer cards in synthetic averages.

The RTX 3090, by contrast, holds a 73rd percentile position with an average score of 27,565, which is lower than the Tesla M60's average despite the RTX 3090 winning every head-to-head test. This apparent contradiction resolves when examining its nearest rivals: the AMD Radeon RX 7800M scores 27,883 (1.1% ahead), the AMD Radeon Pro Vega 20 scores 27,839 (1% ahead), the NVIDIA GeForce RTX 4070 Mobile scores 27,435 (0.5% behind), and the AMD Radeon RX 6700 XT scores 27,425 (0.5% behind). The RTX 3090's average is dragged down by its inclusion of legacy DirectX 9, 10, and 11 tests where it scores much lower (268, 182, and 220 respectively) relative to its modern API scores. The Tesla M60's average only includes two OpenCL and Vulkan results, which skews its average upward.

Architecture Differences

The two GPUs come from entirely different architectural eras. The Tesla M60 uses the GM204 chip built on Maxwell 2.0 architecture, fabricated on a 28 nm process at TSMC. It packs 5,200 million transistors into a 398 mm² die, yielding a transistor density of 13.1 million per square millimeter. The RTX 3090 uses the GA102 chip on Ampere architecture, built on Samsung's 8 nm process with 28,300 million transistors in a 628 mm² die, achieving 45.1 million transistors per square millimeter. The density difference is stark: the Ampere chip crams over three times more transistors per area, reflecting the five-year gap in manufacturing technology between the two releases.

Compute resources diverge dramatically. The Tesla M60 has 2,048 shading units, 128 texture mapping units, and 64 raster output pipelines. The RTX 3090 has 10,496 shading units, 328 TMUs, and 112 ROPs. The RTX 3090 also adds dedicated hardware that the Maxwell card completely lacks: 82 RT cores for ray tracing and 328 tensor cores for AI acceleration. These features appear nowhere in the Tesla M60's specification, representing a fundamental capability difference rather than just a raw throughput gap.

Clock behavior also differs. The Tesla M60 runs at a 557 MHz base clock with a 1178 MHz boost, while the RTX 3090 operates at 1395 MHz base and 1695 MHz boost. The RTX 3090's boost clock is 44% higher than the Tesla M60's boost, and when combined with its 5.1x shading unit advantage, the theoretical peak FP32 throughput of 35.58 TFLOPS versus 4.825 TFLOPS becomes clear. The RTX 3090 also supports FP16 at the same 35.58 TFLOPS rate (1:1 ratio), while the Tesla M60 lists no FP16 capability at all.

Memory architecture is another major divide. The Tesla M60 uses 8 GB of GDDR5 on a 256-bit bus with 160.4 GB/s bandwidth. The RTX 3090 has 24 GB of GDDR6X on a 384-bit bus with 936.2 GB/s bandwidth, a 5.8x increase in memory throughput. The RTX 3090's memory clock runs at 1219 MHz base with 19.5 Gbps effective data rate, substantially faster than the Tesla M60's 1253 MHz memory clock at 5 Gbps effective. These differences matter for workloads that stream large datasets through the GPU.

The Verdict

The benchmark data positions these cards for completely different purposes. The RTX 3090 wins both recorded head-to-head tests and does so by margins that cannot be dismissed. In OpenCL, it outscores the Tesla M60 by a factor of nearly six, and in Vulkan it leads by roughly 71%. If a workload can leverage the RTX 3090's modern feature set, its 82 RT cores, 328 tensor cores, and 24 GB of GDDR6X memory, there is no contest.

However, the Tesla M60's 75th percentile standing and its average score of 30,490, which actually exceeds the RTX 3090's 27,565 average, indicate that it remains relevant in specific enterprise contexts. The Tesla M60's nearest rivals include the RTX 3070 Ti, which it edges by 1.8%, and the RX 6800, which it leads by 1.3%. These are modern consumer cards, yet the Tesla M60 holds its own in the database's aggregate metrics. This suggests that for compute workloads that do not require ray tracing or tensor operations, the older Maxwell card still delivers competitive throughput.

The production status for both cards is end-of-life, so neither represents a future-proof purchase. The RTX 3090's launch MSRP was 1,499 USD, a fact worth noting but not weighing into performance analysis. The Tesla M60 has no recorded launch MSRP in the database.

Specification Differences

The table below lists only the fields where the two differ. Identical values, such as OpenGL 4.6 and Vulkan 1.4 API support, are omitted.

| Specification | Tesla M60 | RTX 3090 |

|---|---|---|

| Architecture | Maxwell 2.0 | Ampere |

| Process Node | 28 nm | 8 nm |

| Foundry | TSMC | Samsung |

| Transistors | 5,200 million | 28,300 million |

| Die Size | 398 mm² | 628 mm² |

| Transistor Density | 13.1M / mm² | 45.1M / mm² |

| Base Clock | 557 MHz | 1395 MHz |

| Boost Clock | 1178 MHz | 1695 MHz |

| Memory Clock | 1253 MHz, 5 Gbps effective | 1219 MHz, 19.5 Gbps effective |

| Memory Size | 8 GB | 24 GB |

| Memory Type | GDDR5 | GDDR6X |

| Memory Bus Width | 256 bit | 384 bit |

| Memory Bandwidth | 160.4 GB/s | 936.2 GB/s |

| Shading Units | 2048 | 10496 |

| TMUs | 128 | 328 |

| ROPs | 64 | 112 |

| RT Cores | None | 82 |

| Tensor Cores | None | 328 |

| Pixel Rate | 75.39 GPixel/s | 189.8 GPixel/s |

| Texture Rate | 150.8 GTexel/s | 556.0 GTexel/s |

| FP32 | 4.825 TFLOPS | 35.58 TFLOPS |

| FP16 | Not listed | 35.58 TFLOPS (1:1) |

| TDP | 300 W | 350 W |

| Slot Width | Dual-slot | Triple-slot |

| Power Connectors | 1x 8-pin | 1x 12-pin |

| Suggested PSU | 700 W | 750 W |

| Bus Interface | PCIe 3.0 x16 | PCIe 4.0 x16 |

| Display Outputs | No outputs | 1x HDMI 2.1, 3x DisplayPort 1.4a |

| DirectX Version | 12 (12_1) | 12 Ultimate (12_2) |

| Dimensions | 267 mm / 10.5 inches length | 336 mm / 13.2 inches length, 140 mm / 5.5 inches height, 61 mm / 2.4 inches width |

| Release Date | 2015-08-29 | 2020-08-31 |

| Predecessor | Tesla Kepler | GeForce 20 |

| Successor | Tesla Pascal | GeForce 40 |

FAQ

Q: Which GPU wins in Geekbench OpenCL?

A: The RTX 3090 wins with a score of 172,758 versus the Tesla M60's 29,506, a delta of -82.9% from the RTX 3090's perspective.

Q: How does the Tesla M60 compare to its nearest rival, the NVIDIA CMP 70HX?

A: The Tesla M60 averages 30,490 points, while the CMP 70HX averages 30,476, a delta of 0%. The two are effectively tied.

Q: Does the RTX 3090 support ray tracing hardware?

A: Yes, it has 82 RT cores. The Tesla M60 has no RT cores listed in the database.

Q: What is the memory bandwidth difference between the two cards?

A: The Tesla M60 provides 160.4 GB/s over a 256-bit GDDR5 bus, while the RTX 3090 provides 936.2 GB/s over a 384-bit GDDR6X bus.

Q: Which card has a higher average benchmark score in the database?

A: The Tesla M60 has an average score of 30,490, which is higher than the RTX 3090's 27,565, despite the RTX 3090 winning all head-to-head tests.

Q: Do both cards support the same DirectX version?

A: No. The Tesla M60 supports DirectX 12 (12_1), while the RTX 3090 supports DirectX 12 Ultimate (12_2).

Where Each One Wins

The RTX 3090 wins every recorded head-to-head benchmark, so its strengths are straightforward. In Geekbench OpenCL, it is 5.9x faster, a margin that comes from its 35.58 TFLOPS FP32 throughput, 10,496 shading units, and 936.2 GB/s memory bandwidth. In Geekbench Vulkan, its 71% lead reflects the same architectural advantages, though the smaller gap hints that the Tesla M60's Maxwell design retains some efficiency in that API. The RTX 3090 also has exclusive capabilities: 82 RT cores for ray-traced workloads, 328 tensor cores for AI inference and DLSS-style acceleration, and 24 GB of GDDR6X memory for large datasets. Its PCIe 4.0 interface doubles the bus bandwidth available to the Tesla M60's PCIe 3.0 connection, which matters for data transfer-bound tasks.

The Tesla M60's advantages are narrower but real. Its 75th percentile ranking versus the RTX 3090's 73rd percentile shows that in the aggregate database metrics, it actually sits slightly higher. Its average score of 30,490 beats the RTX 3090's 27,565, driven by the fact that its benchmark set excludes legacy DirectX 9, 10, and 11 tests where the RTX 3090 scores poorly (268, 182, and 220 respectively). The Tesla M60 also draws less power at 300 W versus 350 W, uses a standard 8-pin connector instead of a 12-pin, occupies a dual-slot design rather than triple-slot, and is shorter at 267 mm versus 336 mm. For a server environment with no display output requirement, the Tesla M60's lack of display connectors is not a disadvantage. Its 2048 shading units and 4.825 TFLOPS are sufficient for compute tasks that do not need tensor or RT acceleration, and its 8 GB of GDDR5 memory, while smaller, is still usable for many inference and rendering workloads.

The data indicates a clear split: the RTX 3090 is the choice for any workload touching modern features, ray tracing, or massive memory footprints, while the Tesla M60 remains a capable, lower-power option for legacy compute stacks that only use OpenCL or Vulkan and do not require the RTX 3090's specialized hardware. The RTX 3090's wins in both shared tests are decisive, but the Tesla M60's competitive standing among its own peers, including a 1.8% edge over the RTX 3070 Ti, shows that age alone does not disqualify it from productive service.