Skip to main content

Genetic Optimization in MT5 on AMD EPYC:

How Much Faster on 32–64 Cores

Section 1 · MT5 Strategy Tester

Why Genetic Optimization in MT5 Is So CPU-Hungry

MT5’s genetic optimization feels “smart”, but under the hood it is still brute-forcing an insane number of parameter combinations. The more inputs you add, the more your CPU starts to look like a tiny engine trying to pull a freight train of backtests.

1.1 · Brute force vs genetics

MT5 is still testing thousands of variants, just in a smarter order

In classic brute-force optimization MT5 walks through every single parameter combination you define. Ten parameters, ten steps each — that is already 1010 passes. Completely unrealistic for a single desktop. So the strategy tester offers a shortcut: the genetic algorithm.

Genetic optimization does not magically reduce the workload to “a few runs”. It simply prioritizes promising areas of the search space and drops obviously bad regions faster. You still run ~tens of thousands passes for any serious EA, especially when you add realistic data ranges and multiple symbols.

In practice, traders often underestimate this: they see “genetic” and expect something almost instant. MT5 still has to simulate every trade for every pass.

1.2 · Combinatorial explosion

Every new parameter can add days to your optimization

A typical EA does not optimize only one thing. You tweak: lot size, stop loss, take profit, trailing rules, filters for volatility, session times, and so on. Each dimension you add multiplies the total work.

Example, just as a rough feeling:
  • 4 parameters × 10 steps ≈ 10 000 combinations
  • 7 parameters × 10 steps ≈ 10 000 000 combinations
  • with walk-forward or multiple symbols — it quickly gets worse
The genetic algorithm can skip part of that, but it cannot break math. If you want robust, not curve-fitted results, you keep more runs, not fewer. That is why people move from a single desktop to multi-core CPUs and EPYC farms: just to keep total optimization time within something close to human life.
1.3 · Real bottlenecks

It looks like “just CPU”, but RAM and disk quietly join the party

When you watch Task Manager during genetic optimization, you see CPU at 100 % and assume this is the only limit. But MT5 is also:
  • keeping tick history and EA state in RAM,
  • reading and writing results to disk,
  • sending tasks to local/remote agents over the network.
On a weak machine, slow SSD or low memory can throttle your cores. On a strong machine (for example, 32–64 cores on AMD EPYC) the CPU scales nicely until one of these other subsystems becomes your new ceiling.

The short version: MT5 is CPU-bound most of the time, but to unlock all those cores you must give them fast RAM, NVMe and a clean local network. Otherwise your expensive hardware spends time waiting instead of crunching passes.

Final step

Ready to stop waiting 2–3 days for every optimization?

Tell us which EA you are running and what kind of optimizations you launch. We will propose an MT5 farm configuration that fits your workload and budget.

Chat via Telegram / WhatsApp
Share a short description of your EA, symbols, history depth and typical number of passes.
Section 2 · Cores and Threads in MT5

How MT5 Strategy Tester Uses Cores and Threads

MT5 does not simply “run faster” when you buy a bigger CPU. It spawns local and remote agents, maps them to cores and threads, and then tries to keep them busy with passes. Understanding this logic is the key to choosing between a fast desktop CPU and a many-core AMD EPYC server.

2.1 · Local agents, remote agents, MQL5 Cloud

MT5 turns each core into an “agent” that eats optimization tasks

When you click Start in the Strategy Tester, MT5 does not run everything inside a single process. It creates testing agents: small worker processes that take passes from a queue and execute them.
Local agents Remote agents over LAN/VPN MQL5 Cloud Network
On a typical machine you get one local agent per logical core (hardware thread). On a 32-core / 64-thread CPU that is up to 64 local workers eating passes in parallel . Remote agents are installed on other PCs or servers and behave the same way, except the task queue travels over the network.

From MT5’s point of view, there is no big philosophical difference between a local core and an agent on a remote EPYC box. If the agent is online and not overloaded, the tester will try to keep it busy.

2.2 · Scaling from 4 to 64 threads

More threads mean more passes in flight – up to a point

Each agent receives a single pass (one specific set of input parameters), runs the full backtest, sends back the result, then takes the next pass. With 4 threads you have at most 4 passes running at once. With 32–64 threads, dozens of passes are executed in parallel.

This is why a many-core CPU like AMD EPYC feels like a different league: instead of waiting for one EA/backtest at a time, you are chewing through batches of passes in parallel and watching the optimization map fill up much faster.

The catch: the more agents you spawn, the more pressure you put on memory, disk, and the scheduler. Past a certain point you get smaller and smaller gains per extra thread.

2.3 · Non-linear speedup

Why 64 cores are not 8 times faster than 8 cores

In theory, doubling the number of agents should cut the optimization time in half. In reality, several things get in the way:
  • Single-threaded pieces inside MT5 and your EA logic.
  • Shared resources – the same disk and memory bus for dozens of agents.
  • Overhead of task scheduling and result aggregation.
  • Energy and thermals – long runs at 100 % load can trigger throttling on consumer CPUs.
On i9-class hardware you may see good scaling up to 8–16 threads. On server CPUs like AMD EPYC, which are built to run flat-out for hours, you can push to 32–64 threads before hitting the “diminishing returns” zone.

The point of moving to EPYC is not perfect linear scaling. It is raw throughput over long horizons: finishing in hours what a single desktop would need days or weeks to complete.

Section 3 · Hardware Contenders

Hardware Contenders: i9-12900K, Ryzen 9 7950X3D and AMD EPYC 9454P

Before we talk about “how much faster” genetic optimization can be, we need to understand the three main types of CPUs traders usually consider: a high-end desktop like the i9-12900K, a cache-heavy Ryzen 9 7950X3D, and a server-grade AMD EPYC 9454P that looks more like a small MT5 data center.

Desktop baseline Intel Core i9-12900K

Intel Core i9-12900K

8P + 8E cores · 24 threads · LGA1700 desktop
Cores / threads
16 / 24
Max boost
up to ~5.1–5.2 GHz
L3 cache
≈ 30 MB
Power envelope
high under full load
The i9-12900K is a strong single-core performer with enough threads to run several MT5 agents in parallel. For many traders it is the first “serious” CPU they use for optimization: you can keep your main terminal, charts and a few agents on one machine without feeling slow.

The downside is that once you start doing heavy multi-symbol genetic optimization, 24 threads are quickly saturated and the chip runs hot for long sessions.

Best as a personal workstation and entry into multi-core optimization, not as a long-running MT5 farm.

High-end desktop Ryzen 9 7950X3D

AMD Ryzen 9 7950X3D

16 cores · 32 threads · 3D V-Cache desktop
Cores / threads
16 / 32
Max boost
up to ~5.6–5.7 GHz
L3 cache
very large (3D V-Cache)
Power envelope
high, but tuned for efficiency
The 7950X3D combines fast cores with a huge L3 cache. For MT5 this means less time spent going to RAM for the same historical data and EA state, especially when you repeat similar passes over and over.

Compared with the i9, you get more threads and often better passes-per-watt, making it attractive for traders who want to run optimizations almost every day but still on a single desktop.

Great for a power-user MT5 rig that handles both optimization and live trading with low latency.

Server-grade AMD EPYC 9454P

AMD EPYC 9454P

48 cores · 96 threads · Genoa server CPU
Cores / threads
48 / 96
Base / boost
server clocks · sustained
L3 cache
large shared cache
Platform
ECC DDR5 · multi-channel
EPYC 9454P belongs to a different world: it is made to run at high utilization 24/7. With up to 96 threads you can spawn dozens of MT5 agents on a single machine or split the CPU into several VMs and build your own optimization farm plus RDP terminals.

Single-core speed is lower than a gaming desktop, but once you load 32–64 cores with genetic optimization, EPYC starts to win simply by brute throughput.

Ideal as a dedicated optimization server or a small in-house MT5 cluster (especially in hosting / data-center scenarios).

Relative strengths for MT5 genetic optimization
Qualitative comparison
i9-12900K (desktop baseline)
Ryzen 9 7950X3D (high-end desktop)
EPYC 9454P (server farm)
Backtest throughput
i9-12900K reference
Ryzen 9 7950X3D ≈ 1.3–1.5× vs i9
EPYC 9454P multi-times faster with 32–64 agents
Stability under 10+ hour runs
i9-12900K good, but hot
Ryzen 9 7950X3D better per watt
EPYC 9454P built for 24/7 load
Cost per million passes
Single desktop cheap to buy, slower
Dedicated EPYC server higher monthly cost, lower cost per pass

Numbers above are qualitative, but the pattern is clear: desktops are fine for experiments and single-user setups, while EPYC makes sense once you treat optimization as a continuous, industrial-scale process.

CHECK PRICES
Section 5 · Desktop vs Server

Desktop vs Server: What Really Matters for MT5 Optimization Speed

For MT5 genetic optimization, the key difference between a powerful desktop (i9, 7950X3D) and a server (EPYC) is not just core count. It is about how long you can run at high utilization and how many passes you can complete per day without babysitting the machine.

Desktop CPUs (i9 / 7950X3D)
  • Very fast single-core for manual trading and UI.
  • Good scaling up to 16–32 threads.
  • Limited by thermals in long 100% runs.
  • Usually 64–128 GB RAM, 1–2 NVMe drives.
Server CPUs (EPYC 9454P)
  • Many cores (32–64+) for parallel agents.
  • Platform designed for 24/7 full load.
  • More RAM channels and ECC memory.
  • Multiple NVMe drives and better I/O.

In short: a desktop is perfect for running MT5 in front of you; a server is perfect for running optimizations behind your back all day and night.

Section 6 · Methodology

Benchmark Methodology: How We Measure “Faster”

To compare CPUs fairly, you need the same workload on each machine. The exact numbers will differ per EA, but the process should be consistent.

Test setup
  • Same MT5 build and settings.
  • Same EA with 5–7 optimized parameters.
  • Same symbols and timeframe (e.g. EURUSD H1).
  • Same history period (e.g. 3–5 years of data).
Metrics we care about
  • Total optimization time (wall clock).
  • Passes per hour (throughput).
  • Average CPU load and stability.
  • Time-to-first-good-result (how quickly you see decent candidates).

Once the methodology is fixed, you can scale agents: first 8, then 16, 32, 64, and see where each CPU stops giving useful extra speed.

Section 7 · Raw Speed

Raw Speed: 8 vs 16 vs 32 vs 64 Cores in MT5 Genetic Optimization

In theory, doubling cores halves optimization time. In practice you see good but sub-linear scaling, because of single-threaded parts and shared resources.

8 logical cores

Typical mid-range CPU or small VM. Enough for simple EAs and short histories. Genetic optimization works, but long runs easily stretch into many hours or days.

16–32 logical cores

Sweet spot for many users. MT5 can run many agents in parallel, and most workloads feel 2–4× faster than on 8 cores, assuming fast NVMe and enough RAM.

32–64+ logical cores

This is the territory of EPYC. Scaling continues, but gains per extra core are smaller. The payoff: whole optimizations that used to take a weekend now fit into one evening or even a few hours.

The exact multiplier depends on your EA and data, but the pattern is stable: once MT5 can keep dozens of agents busy, the calendar time to test a new idea drops dramatically.

Section 8 · 7950X3D vs EPYC

7950X3D vs EPYC Farm: Which One Should You Choose?

Both the Ryzen 9 7950X3D and an EPYC 9454P server are powerful for MT5, but they solve different problems.

Pick 7950X3D if…
  • You want a single, very fast trading PC.
  • You combine manual trading + coding + optimization.
  • You run optimizations often, but not 24/7.
  • You care about latency to broker as much as about throughput.
High-end desktop Up to 32 threads Great single-core speed
Pick EPYC 9454P (farm) if…
  • You manage many EAs / portfolios in parallel.
  • You need constant, background optimization.
  • You want to offload heavy work from your main PC.
  • You prefer renting dedicated hardware instead of buying.
Dedicated server / farm 48+ cores, 96 threads Designed for 24/7 load

Many traders start with a strong desktop (7950X3D), and once they hit a wall in R&D speed, they add an EPYC-based optimization server and keep the desktop only for trading and monitoring.

Section 9 · EPYC Optimization Farm

Building an MT5 Genetic Optimization Farm on AMD EPYC

The idea is simple: let your desktop trade and let your EPYC server suffer. You push heavy MT5 genetic optimization to a many-core box with fast RAM and NVMe, wire agents over LAN or VPN – and suddenly a “weekend optimization” shrinks to a few hours.

1
Choose your EPYC base configuration
For a serious MT5 farm, you want a single, fat node first, not ten weak ones. A typical starting point:
  • AMD EPYC 9454P (48 cores / 96 threads, Genoa).
  • At least 128–256 GB ECC DDR5 RAM.
  • 2× NVMe SSD in RAID1 (for speed + safety).
  • 1 Gbit/s network with low jitter.
This gives you enough headroom to spawn 32–64 MT5 agents and still have resources left for the OS and monitoring.
EPYC 9454P · 48c / 96t 256 GB ECC RAM 2× NVMe Datacenter
2
Decide: bare metal or multiple VMs
Bare metal Windows Server is simplest: one OS, one MT5 agents service, maximum performance. But many users prefer to split the EPYC into a few dedicated VMs:
  • 1–2 VMs purely for MT5 optimization agents.
  • 1 small VM for live MT5 terminals or monitoring.
This way you can reboot or reconfigure the farm without touching critical trading sessions.
Bare metal = max speed VMs = isolation & flexibility
3
Install MT5 agents and connect your farm
On each VM or bare-metal OS:
  • Install MetaTrader 5 and the MetaTester agents.
  • Open a port range in the firewall (for example 3000–3015).
  • Set passwords for agents and note IP + ports.
On your local MT5 terminal (desktop or laptop):
  • Open Strategy Tester → Agents.
  • Add remote agents (EPYC IP, port, password).
  • Group them into a separate “EPYC farm” profile.
After that, every optimization run can be sent to your farm with a single click – while your trading machine remains cool and quiet.

In a typical setup, the only traffic between your PC and the EPYC server is optimization jobs + results. Your broker connection stays local, with your usual low latency.

CHECK PRICES
Section 10 · Final Choice

7950X3D vs EPYC Farm: Which One Should You Choose?

Both options are fast. The real question is not “who wins the benchmark”, but “how do you actually work with MT5”: at one screen, or like a small research desk with constant optimization in the background.

Ryzen 9 7950X3D High-end desktop

When 7950X3D is the right answer

One powerful PC that does everything: charts, coding, manual trades, and optimizations.

Choose 7950X3D if:
  • You are a single trader or small team using one main workstation.
  • You run optimizations often, but not 24/7 on multiple EAs in parallel.
  • You care about low latency to broker from the same machine.
  • You want top single-core performance for manual trading and UI.

In practice: 7950X3D is a perfect “all-in-one” MT5 rig. Great choice if you are still in the phase of building and refining your own strategies, not running a full farm yet.

EPYC 9454P Farm Dedicated server / cluster

When EPYC farm is the right answer

Many cores, remote agents, and MT5 optimizations running like a background service.

Choose EPYC farm if:
  • You manage many EAs / portfolios or work with client strategies.
  • You want optimizations to run constantly in parallel, day and night.
  • You prefer to keep your trading PC light and offload heavy work to a server.
  • You’re ready for a monthly budget for a dedicated EPYC server instead of buying more desktops.

In practice: EPYC shines when optimization is not a weekend event, but an always-on process that feeds new settings and portfolios back to your trading environment.

Quick decision: where are you today?
“I’m optimizing my own systems”

Mostly one person, 1–3 EAs, experiments after work or on weekends. You want to speed things up, but you do not need a full farm yet.

Go 7950X3D Single powerful workstation
“I constantly test many ideas”

Multiple EAs, portfolio-level testing, walk-forward, Monte Carlo. Optimizations run almost every day and you feel that your desktop is always at 100%.

Add EPYC farm Desktop + dedicated server
“I’m a small research/prop team”

Several people, shared EAs, client capital or prop trading. You need predictable optimization throughput and central infrastructure.

Start with EPYC Treat optimization as infrastructure

The most natural upgrade path is: first a strong desktop (7950X3D) as your main MT5 rig, then – when optimizations start blocking your work – add an EPYC-based farm (on-prem or hosted) and move heavy genetic runs there. Your trading stays fast and responsive, while the server quietly does the hard work in the background.

Final step

Ready to stop waiting 2–3 days for every optimization?

Tell us which EA you are running and what kind of optimizations you launch. We will propose an MT5 farm configuration that fits your workload and budget.

Chat via Telegram / WhatsApp
Share a short description of your EA, symbols, history depth and typical number of passes.
FAQ · MT5 Genetic Optimization & High-Core CPUs

FAQ: Genetic Optimization in MT5 on High-Core-Count CPUs

A few practical answers to the questions that usually appear once you start thinking about 32–64 cores, MT5 agents and whether an EPYC-based farm really makes sense.

Q
Is it worth paying extra for 64 cores?
Sometimes yes, sometimes no. It depends on how you actually use MT5.

If you optimize one or two EAs occasionally, a good 16-core desktop (like 7950X3D) already feels very fast. You will not get a proportional benefit from 64 cores, because MT5 and your EA still have single-threaded parts and shared bottlenecks (disk, RAM, network).

Paying for 64 cores starts to make sense if you:
  • run many optimizations in parallel (multiple EAs, symbols, portfolios),
  • use long histories, walk-forward, Monte Carlo, multi-symbol tests,
  • care about total passes per day, not just speed of one run.
For that kind of workload, moving from 16 cores to 32–64 cores can easily turn “multi-day” jobs into “overnight” or “afternoon” runs.
32 cores = sweet spot for many 64 cores = for heavy, continuous R&D
Q
How many agents are “too many” for one machine?
A simple rule: do not blindly max everything out. MT5 can create one local agent per logical thread, but that does not mean you must run them all at 100%.

Practical guidelines:
  • Start with 0.5–1× of your logical threads as agents (e.g. 32–64 agents on a 48c/96t EPYC).
  • Watch CPU, RAM and disk I/O while optimizing.
  • If the system starts swapping, freezing or I/O is pegged, reduce agents.
For many workloads, running a bit below the hard maximum (for example 50–70% of possible agents) gives better, more stable throughput than trying to keep every single thread at 100% all the time.
Stability > “100% everywhere” Tune by watching real load
Q
Do I still need MQL5 Cloud Network if I have an EPYC farm?
Not always, but it can still be useful.

With a decent EPYC server you already have a lot of local power that you control. For many traders this is enough, and they stop using MQL5 Cloud completely for cost and privacy reasons (no EA code leaving their own infrastructure).

The MQL5 Cloud Network still makes sense when you:
  • need a short, massive burst of power for a huge experiment,
  • do not want to expand your own hardware for rare peak loads,
  • are okay with paying per task and sharing compiled code with the cloud.
In other words: an EPYC farm becomes your baseline, and MQL5 Cloud turns into an optional “turbo button”, not your only source of speed.
EPYC = predictable capacity Cloud = burst, not foundation
Q
Can I run optimizations and live trading on the same server?
Technically yes. Strategically: better not, if you can avoid it.

Heavy genetic optimization can:
  • push CPU and RAM to the limit,
  • create disk spikes due to logs and result files,
  • increase latency and risk small freezes or re-quotes.
If you must run both on one machine:
  • set a hard limit on agents during trading hours,
  • separate trading and optimization into different VMs or terminals,
  • keep enough CPU/RAM headroom for live sessions.
The safer pattern is: trading on a clean VPS / desktop close to the broker, and all heavy optimization on a separate EPYC box that can be pushed to 100% without touching your live orders.
Live trading: keep it “boring” Optimization: separate and aggressive
Ready to speed up your MT5 optimization?

Turn your genetic optimization into an EPYC-powered service

If this article made you think “my desktop is not enough anymore”, that is a good sign. The next step is simple: move your heavy MT5 genetic optimization to a dedicated AMD EPYC server and keep your trading machine fast, quiet and focused on orders.

✓
Hardware tuned for MT5: EPYC-based dedicated servers configured for multi-agent genetic optimization, walk-forward and Monte Carlo runs.
✓
No sysadmin headache: OS, agents, monitoring and security are prepared for you — you connect your MT5 and start testing.
✓
Monthly pricing you can plan: predictable cost for a fixed amount of compute, instead of unpredictable cloud bills per million passes.
View MT5 dedicated server plans →
or start with a smaller EPYC setup and scale later — details on winservers.net