Which preferred overclock technique?
Which preferred overclock technique?
Are there any techniques that outperform others in overclocking? My thought is that using BIOS for everything offers the most stability, but I find other methods interesting like XTU or gigabytes software (the least powerful option). In my case, would it even make sense to overclock? The main purpose is a gaming rig with occasional 3DMark Timespy sessions. A friend recommended a simple under-voltage tweak that could provide some benefits, though if the hardware is there, why not utilize it? I typically play on a 4K 144Hz monitor, and FPS in COD and Forza Horizon 5 stay consistently around 144fps, rarely dropping below 140.
Average temperatures during gaming: GPU 55-60°C, CPU 51-55°C (with occasional spikes to 60°C). Timespy scores around 36,400k for the GPU, while the CPU usually sits between 19,200-19,400 with an average temperature of 61°C.
Rig specifications:
13700k
Gigabyte Z690 Aorus Ultra
DDR5 5200
1200W Evga Platinum PSU
H150i Elite
Gigabyte OC 4090
Samsung 980 Pro/970 Plus M.2
Phantek G500A
Monitors Aorus FV43U
Gigabyte G27QC
Not really beneficial. The speed boosts on Intel and AMD are almost at their maximum. You can increase power usage and reach 100 or 200Mhz, but you usually lose single core performance for higher multi-core speeds.
If gaming is the main purpose, keep it as is or lower the power goal to reduce extra heat.
It's important to recognize that various overclocking techniques yield different outcomes. As illustrated by OC Strategy #1: Unleashed Turbo combined with XMP 3.0, the results are quite limited.
SuperPI 4M: +0.50%
Geekbench 5 (single): +2.24%
Geekbench 5 (multi): +8.06%
Cinebench R23 Single: +0.43%
Cinebench R23 Multi: +2.24%
CPU-Z V17.01.64 Single: +0.58%
CPU-Z V17.01.64 Multi: +0.18%
V-Ray 5: +3.92%
AI Benchmark: +7.09%
3DMark Night Raid: +0.90%
CS:GO FPS Bench: +1.44%
Tomb Raider: +4.40%
Final Fantasy XV: +0.48%
3DMark CPU Profile scores at stock
CPU Profile 1 Thread: +0.18%
CPU Profile 2 Threads: +0.27%
CPU Profile 4 Threads: +0.45%
CPU Profile 8 Threads: +0.45%
CPU Profile 16 Threads: +1.54%
CPU Profile Max Threads: +2.11%
Choosing the optimal overclocking path can significantly boost performance, though this depends on the specific workload. RAM remains limited to XMP settings. The Aorus DDR5-6200 model presents a promising option for further overclocking and achieving substantial gains. It may be worthwhile for those willing to invest considerable time and effort to achieve more than 3% improvement in gaming. Reducing the CPU's latency as much as possible will enhance overall results, offering lower latency similar to DDR4 configurations.
Keep in mind this analysis is based on a single sample. For some users, the effort may not justify the gains.
Share my insights on current hardware performance. Overclocking seems to be less effective these days, especially when it comes to power use and little actual gain. It wasn’t that different back in the day—10% improvements were seen as impressive, unlike now. Personally, I haven’t had much success with CPU or memory overclocking through software.
I believe there might still be a role for it in certain scenarios, though not necessarily for me right now. My current setup handles most games well, except for a few that aren't optimized. The 4k monitor I have is only 144hz and 43 inches, which can feel a bit too much for some titles (a lot of screen space to manage). My next choice would be a smaller screen like a Samsung G7 at 165hz or a G8 at 240hz with 4k resolution. I haven’t made any purchases yet; I’ll wait for price reductions. In the meantime, I’m interested in how far the 4090 @ 4k can go. It makes little sense to buy a 240hz 4k if the 4090 isn’t capable, but with new monitors coming out, a 175hz or 190hz model might be possible and I’d have to slightly overclock it.
Not really now. You can push around 500W through a GPU, but you only see a small boost in performance and it creates a lot of extra heat inside the GPU. They’re essentially offering overclocked GPUs, which is why the power needs are so high. Lowering the power limits seems more practical for me. I’d get about 90% of the performance with nearly half the power usage.
In practice, even with a 10900k, you're adjusting the v/f curve parameters. You can lower the LLC setting to 4-5 (maximum 8). It's a slow process, requiring testing each adjustment. For every change, you must run prime 95 small ffts and a full pass takes about an hour and 40 minutes.
This method allows you to reduce power consumption. However, maintaining cache at its default speed increases the demand for vcore and overall power usage.
The outcome is improved performance, but it comes with significant time investment. Combining this with voltage offset tuning, I managed to achieve a SP63 10900k running at 5.2GHz SSE and 5.1GHz AVX. I had to fine-tune every setting myself, minimizing power draw as much as possible. This enabled me to perform prime 95 small ffts with AVX without exceeding 100c. More stable performance in the 80s and low 90s.
This setup is suitable for a 200MHz overclock on the motherboard's MCE and a 100MHz boost over the AI. The gains from the 100MHz are offset by a drop in cache speed from 47 to 43, resulting in reduced vcore usage—bringing it back to stock levels.
I have a benchmark overclock that consistently achieves Prime95 small fft/large fft stability (though I can't maintain it continuously). The CPU-z results remain consistent. While this approach offers performance gains, it demands considerable time and effort.