AMD EPYC 7663 vs AMD EPYC 9335 Comparison

Head-to-Head Benchmarks

The head-to-head comparison between the AMD EPYC 9335 and the AMD EPYC 7663 reveals a decisive pattern: the older EPYC 7663 wins eight of the eleven recorded benchmarks, while the newer EPYC 9335 takes three. However, the margins tell a more nuanced story than the raw win count.

The EPYC 7663 delivers its most dominant performance in the physics test, scoring 8020 against the EPYC 9335's 1905, a 76.2% advantage. This is the largest delta in the entire comparison. The encryption workload also heavily favors the older chip, with the EPYC 7663 scoring 111725 versus 63159, a 43.5% gap. Data compression shows a 16.9% lead for the EPYC 7663 (1447020 vs 1203096), while random string sorting goes to the EPYC 7663 by 36.8% (184367 vs 116608). Prime number finding, integer math, multithread, and floating-point workloads all favor the EPYC 7663, with deltas of 49.7%, 26%, 19.8%, and 9.3% respectively.

The EPYC 9335, despite losing most tests, wins the extended instructions benchmark decisively. Its score of 105706 beats the EPYC 7663's 73719 by 43.4%. Single-thread performance also goes to the newer chip, with the EPYC 9335 scoring 2732 versus 2606, a 4.8% edge. These two wins highlight the architectural generational leap in per-core efficiency.

The multithread score is particularly telling for server workloads: the EPYC 7663's 82087 beats the EPYC 9335's 65811. The physics result, with its 76.2% gap, suggests the EPYC 7663 handles simulation or physics-based calculations with far greater efficiency, likely due to its higher core count. The encryption result reinforces this pattern, showing the 56-core chip maintaining a substantial lead in security-heavy tasks.

The single-thread win for the EPYC 9335, while modest at 4.8%, indicates that Zen 5's IPC improvements translate to real-world single-core responsiveness. Yet this advantage does not scale across the broader workload suite, as the EPYC 7663's sheer core count overwhelms the newer architecture in most parallel tasks.

FAQ

Q: Which processor wins the majority of head-to-head benchmarks?

A: The AMD EPYC 7663 wins 8 of the 11 recorded benchmarks, including data compression, encryption, prime numbers, floating-point math, integer math, multithread, physics, and random string sorting. The AMD EPYC 9335 wins only 3: extended instructions, single-thread, and single-thread (duplicate test).

Q: What is the largest performance gap between the two CPUs?

A: The physics benchmark shows the largest delta, with the EPYC 7663 scoring 8020 versus the EPYC 9335's 1905, a 76.2% advantage for the older chip. The second-largest gap is in encryption, where the EPYC 7663 leads by 43.5%.

Q: How does single-thread performance compare?

A: The EPYC 9335 wins single-thread performance with a score of 2732 versus the EPYC 7663's 2606, a 4.8% edge. This is the only benchmark where the newer chip's per-core advantage is visible.

Q: Which CPU shows a stronger result in extended instruction workloads?

A: The EPYC 9335 dominates extended instructions, scoring 105706 against the EPYC 7663's 73719, a 43.4% lead. This likely reflects Zen 5's improved instruction set handling.

Q: What does the multithread score indicate for server workloads?

A: The EPYC 7663's multithread score of 82087 exceeds the EPYC 9335's 65811 by 19.8%. This suggests the 56-core chip handles thread-heavy parallel workloads more effectively despite its older architecture.

Q: How much slower is the EPYC 9335 in data compression?

A: The EPYC 7663 scores 1447020 in data compression versus the EPYC 9335's 1203096, a 16.9% deficit for the newer chip. The 56-core processor also leads in random string sorting by 36.8% (184367 vs 116608).

The Verdict

The data presents a clear choice based on workload profile. The AMD EPYC 7663 is the superior processor for most parallel, compute-intensive tasks. It wins multithread, integer math, floating-point math, prime number finding, encryption, data compression, physics, and random string sorting. For database workloads, scientific simulations, encryption-heavy services, or any application leveraging many cores, the EPYC 7663's 56 cores and 112 threads deliver measurably higher throughput.

The AMD EPYC 9335 is the better pick for workloads that prioritize single-thread responsiveness and extended instruction set execution. Its 4.8% single-thread advantage and 43.4% extended instructions lead are meaningful for applications that rely on per-core performance, such as certain database query processing or scientific code that is not fully parallelized. The newer architecture also offers superior memory bandwidth (576.0 GB/s versus 204.8 GB/s) and PCIe Gen 5 support, which may matter for I/O-bound scenarios that the benchmark suite does not capture.

However, the benchmark results are unambiguous: in the recorded tests, the EPYC 7663 wins the majority. The physics test, with its 76.2% gap, suggests the older chip is disproportionately strong in simulation workloads. The encryption result (43.5% lead) makes the EPYC 7663 the safer choice for security-focused deployments. The EPYC 9335 should be selected only when per-core performance or the latest memory and I/O standards are the deciding factors, not when raw parallel compute is the priority.

Specification Differences

The two processors differ fundamentally in several recorded specifications. The EPYC 9335 has 32 cores and 64 threads, while the EPYC 7663 has 56 cores and 112 threads. Base clocks are 3.00 GHz for the EPYC 9335 versus 2000.00 MHz for the EPYC 7663; boost clocks are 4.40 GHz versus 3.50 GHz respectively.

Thermal design power differs, with the EPYC 9335 rated at 210 and the EPYC 7663 at 240. The socket changes from AMD Socket SP5 (EPYC 9335) to AMD Socket SP3 (EPYC 7663). Memory support moves from DDR5 (EPYC 9335) to DDR4 (EPYC 7663), with the memory bus changing from Twelve-channel to Eight-channel. Memory bandwidth drops from 576.0 GB/s to 204.8 GB/s.

PCIe support advances from Gen 4, 128 Lanes (CPU only) on the EPYC 7663 to Gen 5, 128 Lanes (CPU only) on the EPYC 9335. Release dates are 2024-10-09 for the EPYC 9335 and 2021-03-14 for the EPYC 7663. The launch MSRP is $3178 for the EPYC 9335 and $6366 for the EPYC 7663.

Architecture Differences

The architectural gap spans two generations. The EPYC 9335 uses Zen 5 architecture under the codename Turin, built on a 4 nm process at TSMC. The EPYC 7663 uses Zen 3 architecture under the codename Milan, built on a 7 nm process at TSMC. Transistor counts are nearly identical (33,260 million versus 33,200 million), but die sizes differ: the EPYC 9335 uses 4x 70.6 mm² dies, while the EPYC 7663 uses 8x 81 mm² dies.

Cache configurations diverge significantly. The EPYC 9335 has 80 KB of L1 cache per core, 1 MB of L2 per core, and 128 MB of shared L3. The EPYC 7663 has 64 KB of L1 per core, 512 KB of L2 per core, and 256 MB of shared L3. The older chip offers double the L3 cache, which likely contributes to its strong performance in data-heavy workloads like compression and sorting.

The process node shrink from 7 nm to 4 nm explains the EPYC 9335's higher clocks and lower TDP despite fewer cores. The Zen 5 architecture brings improved per-core instructions, evidenced by the 43.4% extended instructions lead. Memory architecture changes from eight-channel DDR4 to twelve-channel DDR5, tripling theoretical bandwidth (576.0 GB/s versus 204.8 GB/s). PCIe generation advances from Gen 4 to Gen 5, doubling per-lane bandwidth for I/O.

Where Each One Wins

The AMD EPYC 7663 is the clear winner in eight benchmark categories: data compression (16.9% lead), data encryption (43.5% lead), find prime numbers (49.7% lead), floating-point math (9.3% lead), integer math (26% lead), multithread (19.8% lead), physics (76.2% lead), and random string sorting (36.8% lead). These results point to workloads involving large datasets, parallel math operations, cryptographic processing, and simulation tasks. The 56-core configuration and 256 MB L3 cache give it a decisive edge in throughput-oriented applications.

The AMD EPYC 9335 wins in extended instructions (43.4% lead) and single-thread performance (4.8% lead). These wins suggest suitability for code that leverages advanced instruction sets, such as AVX-512 style operations, and for applications where per-core latency matters more than total throughput. The higher boost clock (4.40 GHz versus 3.50 GHz) and newer Zen 5 core explain this advantage.

For practical deployment, the choice depends on the workload mix. If the server runs a mix of parallel batch jobs, database queries with heavy sorting, or encryption services, the EPYC 7663's benchmark dominance makes it the data-driven choice. If the workload is dominated by single-threaded processes or relies heavily on extended instruction execution, the EPYC 9335 offers a measurable edge. The physics result (76.2% gap) is the strongest signal that the EPYC 7663 handles simulation-style workloads far better, while the extended instructions result (43.4% gap) shows the EPYC 9335 excels at modern instruction-heavy code.