On the tested four-core server, one question took 580 ms at p50 with Laya’s English checkpoint, 193 ms with its multilingual checkpoint, and 584 ms with typed-decisions. Those are in-process results for a specific machine and software version—not a general speed guarantee. The 2026 Laya v0.3.20 benchmark used an AWS m7a.xlarge with an AMD EPYC 9R14, four physical cores, no SMT, 16 GiB RAM, fp32, and OMP_NUM_THREADS=4. Laya’s benchmark document and its project repository provide the source context.
What the four-core benchmark measured
The server test ran in-process, so its timings do not include HTTP handling, network transfer, or request queuing. Each call alternated a three-option choice question and a noul question. The guide reports p50 latency and says p95 was within 2% of p50 for every row. The Laya project’s BENCHMARKS.md states: “Every checkpoint answered byte-identical questions in each run (fixed seed).”
As an Amazon Associate I earn from qualifying purchases.
| Checkpoint | 1 question | 5 questions | 10 questions | 50 questions | Cold load |
|---|---|---|---|---|---|
laya (English) |
580 ms | 3,072 ms | 6,244 ms | 35,969 ms | 4.4 s |
laya-multilingual |
193 ms | 912 ms | 1,842 ms | 11,157 ms | 2.5 s |
laya-typed-decisions |
584 ms | 2,819 ms | 6,031 ms | 35,653 ms | 0.5 s |
All table timings are from the benchmark guide’s transcription of the project benchmark results. Cold-load times are approximate and depend on the operating system’s file cache. The 0.5-second typed-decisions cold load is a separate startup measurement; it does not imply faster question processing.
How to interpret the speed differences
Checkpoint choice changes latency
For one question, multilingual took about one-third the time of English on this machine. The guide notes that the checkpoints use different encoder sizes: mmBERT-base has 322 million parameters for multilingual, while ModernBERT-large has 421 million for English. The benchmark does not break down how much of the measured gap comes from model size or other factors. Treat the result as specific to this setup, and choose a checkpoint for the task and language coverage it supports rather than latency alone.
#1 Best Overall
- The world’s fastest gaming processor, built on AMD ‘Zen5’ technology and Next Gen 3D V-Cache.
- 8 cores and 16 threads, delivering +~16% IPC uplift and great power efficiency
- 96MB L3 cache with better thermal performance vs. previous gen and allowing higher clock speeds, up to 5.2GHz
- Drop-in ready for proven Socket AM5 infrastructure
- Cooler not included
More questions cost close to proportionally more
From one to ten questions, elapsed time grew roughly with the question count: ten-question calls took 6.244 seconds on English, 6.031 seconds on typed-decisions, and 1.842 seconds on multilingual. In this report, a 50-question call cost about 15–20% more per question than the smaller batches. Batching therefore delivered little per-question savings on this CPU.
Cold loading is distinct from serving a loaded model
A cold load happens before the checkpoint can answer the first request. Preloading checkpoints intended for user traffic avoids putting that startup delay on the first request, at the cost of keeping their memory resident.
Rank #2
- Can deliver fast 100 plus FPS performance in the world's most popular games, discrete graphics card required
- 6 Cores and 12 processing threads, bundled with the AMD Wraith Stealth cooler
- 4.2 GHz Max Boost, unlocked for overclocking, 19 MB cache, DDR4-3200 support
- For the advanced Socket AM4 platform
CPU results are not GPU results
The guide separately reports one-question p50 timings of 39.5 ms for English Laya and 32.8 ms for multilingual on a Tesla T4. These are GPU measurements from different hardware and are not part of the four-core CPU test. They provide context for considering GPU evaluation when throughput or low latency is important, but do not predict performance on a different GPU or deployment.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Memory and thread settings matter
Memory figures have different meanings
The benchmark script peaked at 9.3 GiB while as many as five checkpoints were loaded. That is a whole-run peak, not a measurement of memory required by one checkpoint. The project’s requirements guide estimates about 3 GB of free RAM for one checkpoint, recommends 8 GB for the default Router and 16 GB for a server preloading all three, and lists download sizes of about 808 MB for English or typed-decisions and 647 MB for multilingual. Those estimates and recommendations are distinct from the benchmark’s measured peak.
Rank #3
- AMD Ryzen 9 9950X3D Gaming and Content Creation Processor
- Max. Boost Clock : Up to 5.7 GHz; Base Clock: 4.3 GHz
- Form Factor: Desktops , Boxed Processor
- Architecture: Zen 5; Former Codename: Granite Ridge AM5
Check thread configuration before judging CPU performance
A separate laptop test illustrates how thread settings can change results; it is not the four-core server benchmark. On a Ryzen 9 6900HX under WSL2, a three-question HTTP call took 9,396 ms at p50 with the stated default thread settings and 783 ms after setting PyTorch intra-op threads to 8 and inter-op threads to 1—about twelve times faster.
On that same laptop, one in-process question took 910 ms with one thread, 374 ms with four, 329 ms with eight, and 388 ms when using every vCPU. The best result in this small set was eight threads, not all available vCPUs. The practical lesson is to test thread settings on the target machine instead of assuming that using every core will be fastest.
Rank #4
- Pure gaming performance with smooth 100+ FPS in the world's most popular games
- 6 Cores and 12 processing threads, based on AMD "Zen 5" architecture
- 5.4 GHz Max Boost, unlocked for overclocking, 38 MB cache, DDR5-5600 support
- For the state-of-the-art Socket AM5 platform, can support PCIe 5.0 on select motherboards
- Cooler not included
What these results mean for a deployment
The measurements suggest a CPU may suit background classification and some interactive multilingual routing. English calls containing several questions may feel slow, and high-throughput, low-latency serving is a reason to evaluate GPU options. These are interpretations of the reported tests, not guarantees about a particular production workload.
For a realistic estimate, run the intended checkpoint and request mix on the target machine. Include the deployment’s actual concurrency and HTTP path: the four-core figures are in-process measurements, while the separate laptop HTTP result shows that serving configuration can materially affect elapsed time. The documented numbers apply to Laya v0.3.20; they should not be treated as performance results for later versions without a new run. The cited four-core test has not been independently reproduced.
Quick Recap
Best Value
- Processor provides dependable and fast execution of tasks with maximum efficiency.Graphics Frequency : 2200 MHZ.Number of CPU Cores : 8. Maximum Operating Temperature (Tjmax) : 89°C.
- Ryzen 7 product line processor for better usability and increased efficiency
- 5 nm process technology for reliable performance with maximum productivity
- Octa-core (8 Core) processor core allows multitasking with great reliability and fast processing speed
- 8 MB L2 plus 96 MB L3 cache memory provides excellent hit rate in short access time enabling improved system performance
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
