AMD EPYC 9475F vs AMD Ryzen Threadripper 9970X Comparison

Where Each One Wins

The benchmark record splits cleanly along workload type. The AMD EPYC 9475F takes 9 of 11 head-to-head tests, and the margins are often decisive. The AMD Ryzen Threadripper 9970X wins exactly two tests, both single-thread measurements, and both with identical scores of 4530 against the EPYC's 3779. That 16.6% gap in single-thread performance is the Threadripper's sole territory, and it is a meaningful one for lightly threaded applications.

Everything else belongs to the EPYC. The largest EPYC advantage appears in prime number finding, where it scores 1507 versus 615, a 145% delta. Physics simulation shows a similar pattern: 16443 versus 6835, a 140.6% lead. These are not marginal wins; they represent fundamentally different throughput capabilities. Data encryption favors the EPYC by 34.4% (116648 versus 86765), and random string sorting favors it by 32.6% (253936 versus 191445). Floating point math runs 31.3% ahead (406524 versus 309719), and integer math runs 30.2% ahead (605696 versus 465378). Even the multithread score, the closest of the EPYC's wins, shows a 14% advantage (122476 versus 107399).

The use-case split is therefore straightforward. The Threadripper 9970X is the choice when per-core responsiveness dominates, such as interactive workloads, compilation steps that are largely serial, or applications that scale poorly across many cores. The EPYC 9475F is the choice when raw parallel throughput matters more than clock speed, which covers most server, database, scientific, and content-rendering scenarios. The average benchmark score reinforces this: the EPYC sits at 350933, the Threadripper at 279778, a difference of roughly 25% in aggregate performance.

Architecture Differences

Both processors share the Zen 5 architecture and the 4 nm TSMC process node, but the similarities end there. The EPYC 9475F is built on the Turin codename, part of the EPYC 9005 series, and uses the AMD Socket SP5. The Threadripper 9970X uses the Shimada Peak codename, belongs to the 9000 series, and fits the AMD Socket sTR5. The EPYC packs 48 cores and 96 threads, while the Threadripper offers 32 cores and 64 threads. That 16-core gap is the primary source of the EPYC's multithreaded dominance.

Cache organization differs substantially. The EPYC provides 80 KB of L1 per core, 1 MB of L2 per core, and 256 MB of shared L3 cache. The Threadripper provides 64 KB of L1 per core, the same 1 MB of L2 per core, but only 128 MB of L3. The EPYC's L3 capacity is double that of the Threadripper, which helps explain its advantage in data compression and encryption workloads that benefit from larger working sets residing on-chip.

Memory architecture also diverges sharply. The EPYC uses a twelve-channel DDR5 memory bus with a recorded bandwidth of 576.0 GB/s. The Threadripper uses a quad-channel DDR5 bus with 204.8 GB/s bandwidth. That is a 2.8x difference in theoretical memory bandwidth, and it directly impacts the EPYC's leads in random string sorting and floating point math, both of which are memory-latency and bandwidth sensitive. Both support ECC memory, a point in favor of reliability for either platform.

The transistor and die counts reflect the scale difference. The EPYC uses 66,520 million transistors across 8 dies of 70.6 mm² each. The Threadripper uses 33,260 million transistors across 4 dies of the same 70.6 mm² size. The EPYC effectively doubles the silicon area and transistor budget. PCIe lane counts differ as well: the EPYC provides Gen 5 with 128 lanes (CPU only), while the Threadripper provides Gen 5 with 80 lanes (CPU only). Neither includes integrated graphics, and both are actively in production.

Clock speeds tell the other side of the story. The Threadripper has a base clock of 4.00 GHz and a boost clock of 5.40 GHz. The EPYC has a base clock of 3.65 GHz and a boost clock of 4.80 GHz. The Threadripper's 0.35 GHz base advantage and 0.60 GHz boost advantage explain its single-thread win. Higher clocks come at the cost of core count, and the TDP reflects that trade: the Threadripper is rated at 350, the EPYC at 400. The EPYC is also multiplier-locked, while the Threadripper has an unlocked multiplier for overclocking. Release dates differ by roughly nine months: the EPYC launched in October 2024, the Threadripper in July 2025.

The Verdict

The data points to a simple conclusion: choose the AMD EPYC 9475F for parallel throughput, choose the AMD Ryzen Threadripper 9970X for single-thread performance. The EPYC wins 9 of 11 benchmarks and holds an average score advantage of 350933 versus 279778, roughly 25% higher. Its wins include every multithreaded and memory-intensive test, with margins ranging from 14% to 145%. The Threadripper wins only the single-thread tests, but it wins those by 16.6%, which is a substantial margin for that metric.

The EPYC's 48 cores, 256 MB L3, twelve-channel memory, and 128 PCIe lanes make it the clear server and workstation processor for database, virtualization, scientific computing, and any workload that can use more than 32 threads. The Threadripper's 32 cores, 128 MB L3, quad-channel memory, and higher clocks make it the better fit for desktop workstation use where some applications remain single-threaded and where the unlocked multiplier allows tuning. The Threadripper also carries a lower TDP of 350 versus 400, which may matter in compact workstation builds.

The launch MSRP for the EPYC 9475F is $7592, and for the Threadripper 9970X it is $2499. The benchmark data does not establish a direct price-to-performance relationship, but the average score difference of 25% is far smaller than the launch price gap. Users who need the EPYC's specific capabilities will find them in abundance. Users who do not need 48 cores will likely find the Threadripper's single-thread advantage more relevant to their daily workloads.

FAQ

Q: Which processor wins more benchmarks overall?

A: The AMD EPYC 9475F wins 9 of 11 head-to-head tests. The AMD Ryzen Threadripper 9970X wins the remaining 2, both single-thread measurements.

Q: How large is the single-thread performance gap?

A: The Threadripper 9970X scores 4530 in single-thread tests, which is 16.6% higher than the EPYC 9475F's 3779.

Q: What is the biggest benchmark margin between the two?

A: The EPYC 9475F leads in prime number finding by 145%, scoring 1507 versus 615. Physics is close behind at 140.6% (16443 versus 6835).

Q: Do both processors support ECC memory?

A: Yes, both list ECC memory as supported. They also both support DDR5, though with different channel configurations: twelve-channel for the EPYC, quad-channel for the Threadripper.

Q: Which processor has more cache and memory bandwidth?

A: The EPYC 9475F has 256 MB of shared L3 cache and 576.0 GB/s memory bandwidth. The Threadripper 9970X has 128 MB of L3 and 204.8 GB/s bandwidth.

Q: Are both processors on the same manufacturing process?

A: Yes, both use the 4 nm process from TSMC and the Zen 5 architecture. They differ in codename, socket, core count, and clock speeds.

Head-to-Head Benchmarks

The single-thread test is the only place where the Threadripper 9970X asserts itself. Its score of 4530 against the EPYC's 3779 represents a 16.6% lead, and the identical result appears in both single-thread benchmark entries. This is the expected outcome given the Threadripper's 5.40 GHz boost clock versus 4.80 GHz, but the magnitude is worth noting. A 16.6% single-thread gap is not trivial; it will be visible in everyday desktop responsiveness and in serial code paths.

The EPYC's wins are led by the two extreme outliers. Prime number finding shows a 145% delta, which is the largest of any test. This workload is highly integer-intensive and scales almost perfectly with core count, so the EPYC's 48 cores versus 32 cores produce a massive advantage. Physics simulation shows a 140.6% delta, another workload that parallelizes well across many cores. These two tests alone establish the EPYC as the throughput king.

The mid-range wins are equally informative. Data encryption shows a 34.4% lead (116648 versus 86765), which reflects both the core count and the larger L3 cache. Random string sorting shows a 32.6% lead (253936 versus 191445), a test that is sensitive to memory bandwidth and cache capacity. Floating point math runs 31.3% ahead (406524 versus 309719), and integer math runs 30.2% ahead (605696 versus 465378). Data compression shows a 22.7% lead (2156305 versus 1757998), and extended instructions run 21.7% ahead (173169 versus 142342). The multithread score, at 14% (122476 versus 107399), is the closest EPYC win, but it is still a clear victory.

The average benchmark scores put the EPYC at 350933, which places it in the 100th percentile of all CPUs in the database. The Threadripper sits at 279778, in the 99th percentile. The nearest rivals for the EPYC include the AMD EPYC 9754 at 364371 (3.7% higher), the Intel Xeon 6960P at 365194 (3.9% higher), the AMD EPYC 9655 at 373479 (6% higher), and the AMD EPYC 9535 at 379408 (7.5% higher). The Threadripper's nearest rivals include the Intel Xeon 6780E at 280438 (0.2% higher), the AMD EPYC 9565 at 285471 (2% higher), the Intel Xeon 696X at 286102 (2.2% higher), and the AMD EPYC 9555P at 287066 (2.5% higher). These rival positions show that both processors sit near the top of their respective performance tiers, with the EPYC holding a higher absolute standing.

Specification Differences

The core counts differ by 16 cores: the EPYC 9475F has 48 cores and 96 threads, while the Threadripper 9970X has 32 cores and 64 threads. Base clocks differ by 0.35 GHz (3.65 versus 4.00), and boost clocks differ by 0.60 GHz (4.80 versus 5.40). TDP differs by 50 (400 versus 350). The sockets are different: SP5 for the EPYC, sTR5 for the Threadripper. Codename differs: Turin versus Shimada Peak. The EPYC uses 8 dies of 70.6 mm² each with 66,520 million transistors; the Threadripper uses 4 dies of the same size with 33,260 million transistors.

Cache differs in L1 size per core (80 KB versus 64 KB) and total L3 (256 MB versus 128 MB). L2 is identical at 1 MB per core. Memory channels differ: twelve for the EPYC, quad for the Threadripper, leading to 576.0 GB/s versus 204.8 GB/s bandwidth. PCIe lanes differ: 128 versus 80, both Gen 5. The EPYC is multiplier-locked, the Threadripper is unlocked. Release dates differ: October 2024 versus July 2025. The EPYC's part number is 100-000001143, the Threadripper's is 100-000001594. Both are actively in production, both lack integrated graphics, and both support ECC memory. The EPYC is in the Server/Workstation market segment, the Threadripper in the Desktop segment.