Best hardware for Blocknet mining

This guide shows which machines mine Blocknet fastest with seine, how many threads your RAM allows, and what hashrate to expect, so you can pick hardware or size up the computer you already have.

What is the best hardware for Blocknet mining?

Blocknet’s proof of work is Argon2id with a 2 GiB memory block per hash. That makes it memory-hard: hashrate depends as much on memory bandwidth and latency as on core count, and on how many of those 2 GiB blocks fit in RAM at once. A strong memory system is what the top machines in the table below have in common.

Best measured hashrate by generation

These are the best results measured with seine for each hardware generation. The last column is the RAM the mining threads use for their memory blocks: threads × 2 GiB.

The server rows ran on AWS cloud machines with 2 MiB huge pages, and the M4 Max row on a 48 GB MacBook Pro. Each figure is completed hashes divided by elapsed time, averaged over repeated runs, measured between July and October 2026. The Zen 5 result is the mean of four runs with seine v0.2.22 on an AWS m8a.8xlarge, a 32-core EPYC 9R45 machine with 128 GiB of RAM.

Results for every processor tested, with the evidence for each row, are on the performance page.

Why memory bandwidth sets your hashrate

Each Argon2id hash fills its entire 2 GiB block once and, while doing so, reads earlier parts of it back. A processor’s L3 cache is measured in megabytes: on the Zen 5 server, each group of eight cores shares 32 MiB, so a single hash works through 64 times more memory than that cache holds. Most of the work is memory traffic.

The numbers show it. At 49.4 H/s, the Zen 5 server writes about 99 GiB of fresh memory blocks every second (49.4 × 2 GiB). Hashrate climbs with threads until the memory system is busy: on that machine, 16 threads mined 37.8 H/s and 32 threads mined 49.4 H/s. Seine measures that curve on your machine and settles on the best thread count.

Three things follow when you choose hardware:

• Memory bandwidth counts as much as cores. Weigh the memory system as heavily as the core count when you compare machines.
• RAM sets the thread ceiling. Every thread needs its own 2 GiB block, so RAM often limits the thread count before cores do.
• The kernel matters on the same chip. Seine carries several Argon2id kernels per processor family. On the EPYC 9R45, the streaming kernel mines 30% faster than AVX2 (34.47 to 44.74 H/s at 24 threads). On AWS Graviton4, the SHA3 kernel with prefetching mines 56% faster than plain NEON (13.80 to 21.50 H/s). Seine picks the faster one while it mines.

How much RAM do you need for each mining thread?

Here is the arithmetic for common sizes. “On paper” is RAM ÷ 2 GiB. The last column subtracts seine’s 5% reserve and about 2-3 GB for the operating system, so treat it as a rough guide. For laptops and older PCs with 4-12 GB, see the guide to mining on an old PC or laptop.

Seine also caps threads at the processor’s logical core count, so the real limit is whichever runs out first. Take a 16-core desktop processor with 32 hardware threads. With 32 GB of RAM it runs about 14 mining threads, and with 64 GB about 29. To run 32 threads you need 64 GiB for the memory blocks alone plus headroom, which is why the 32-thread Zen 5 result came from a machine with 128 GiB.

If you’re buying RAM for mining, match it to the threads you want: about 2 GB per thread, plus a few gigabytes for the system.

Can you mine Blocknet with your CPU and NVIDIA GPU together?

Yes. On Windows and Linux, seine mines on every CPU core and every NVIDIA graphics card it detects at the same time, and the hashrates add up. You don’t need a flag for this; it is the default.

• Windows: the download includes the CUDA runtime libraries, so you only need an up-to-date NVIDIA driver.
• Linux: install the NVIDIA driver and the NVRTC libraries from your distribution or from NVIDIA.
• First launch: seine tunes its kernel settings for each card and caches the result, so later starts take seconds.

On Apple Silicon Macs, seine mines on the CPU by default, the fastest setup measured on the M4 Max. The NVIDIA guide covers multi-GPU rigs and requirements.

What does seine tune automatically?

You don’t need to configure anything to reach the results above. Run seine, paste your address or @name, and accept the default pool, bntpool.com. The mining guide walks through those steps. While it mines, seine:

• Sizes the thread count to your RAM and cores, then compares thread counts on real mining work and keeps the best one.
• Picks the CPU kernel, such as AVX2, AVX-512, the streaming kernel on Zen 5 EPYC or SHA3 on Arm, switching only after two confirmed gains.
• Places threads on big servers. On Linux, it spreads threads across NUMA nodes and L3 cache groups, which added 16.7% on a rented EPYC 8434P at 24 threads. The guide to mining on servers has the details.
• Protects your memory. It keeps a 5% reserve, parks threads if RAM gets tight and brings them back when it frees up.
• Remembers what worked, so the next start begins from the tuned settings.

None of this reconnects to the pool, so tuning costs you no mining time. The automatic tuning guide explains each step.

One optional extra is worth doing on Linux and WSL: reserving huge pages added about 4-6% with 2 MiB pages and a further 2% with 1 GiB pages on the validation machine. The huge pages guide has the commands, and the large servers guide covers EPYC and Xeon machines.

How do you check your own hashrate?

The quickest way is to download seine and start mining: the dashboard shows your hashrate within moments, and you can look up your address on bntpool.com to see what the pool records. Give the tuner a little time to settle before you read the number.

To measure the processor without connecting to a pool, run seine’s offline benchmark:

./seine --bench --bench-kind backend --backend cpu --bench-secs 20 --bench-rounds 3

It reports completed hashes per second, the same measure the table above uses, so you can compare your machine with the published results.