What is the hashrate of an AMD EPYC 7742 on RandomX?

A single AMD EPYC 7742 typically delivers roughly 35,000–50,000 hashes per second (35–50 kH/s) on Monero’s RandomX algorithm with a well-configured native miner. Around 40 kH/s per processor is a useful planning estimate—not a guarantee—and you should not expect a browser miner to match that figure.

What that estimate covers

Hashrate measures how many mining calculations your processor completes each second. Here, 1 kH/s means 1,000 calculations per second. RandomX is Monero’s mining algorithm: the set of calculations your computer performs to help secure the network.

The estimate above assumes:

• One EPYC 7742, rather than a server containing two processors.
• A native miner, such as XMRig, running directly on your operating system.
• Suitable memory, adequate cooling, and mining-focused settings.
• Little competition from other demanding software.

Be careful when comparing published results. A result labeled “EPYC 7742” may come from a dual-socket server, meaning it has two physical processors. Its combined hashrate is not the performance of one chip.

Why its 64 cores do not tell the whole story

The EPYC 7742 has 64 physical cores, 128 processing threads, and 256 MB of L3 cache. Threads let a processor handle multiple streams of work. L3 cache is fast storage inside the processor that keeps frequently needed data close by.

RandomX benefits from about 2 MB of L3 cache per mining thread. On paper, the 7742’s cache capacity supports 128 threads, but that does not mean using every thread always produces the best result.

Its cache is divided among groups of cores, and memory access matters. You should test several thread counts rather than assume the highest setting is fastest.

RandomX’s fast mode also uses a 2,080 MB dataset, a working collection of mining data. A native miner may create additional copies to keep data close to different processor regions.

What moves your result up or down

The biggest setup details are:

• Memory configuration: The 7742 supports eight memory channels. Properly populated channels help avoid starving its many cores of data.
• Memory placement: NUMA, short for non-uniform memory access, describes how some memory is closer to particular cores. Correct placement reduces delays.
• Huge pages: These are larger blocks of operating-system memory. Enabling them can improve RandomX performance.
• Power and cooling: Server power limits or overheating can reduce sustained clock speeds.
• Background work: Virtual machines and other server jobs can take resources away from mining.

Measure performance after the system settles into sustained operation. A brief peak is less useful than a stable average.

What to expect in your browser

For browser-based mining, do not use the native 35–50 kH/s range as your expected output. Browsers restrict access to processor and memory controls that native miners use for speed.

Your result depends on the browser, the mining implementation, and how many worker threads it can use effectively. Test your actual setup, and watch temperatures and power consumption. Hashrate alone does not tell you whether mining covers electricity costs.

The bottom line

For one AMD EPYC 7742, use about 40 kH/s as a rough native RandomX planning figure, with 35–50 kH/s as a reasonable configuration-dependent range. Confirm whether any benchmark uses one or two processors, and measure browser performance separately.