Bottleneck in RAM speed reaching 30,000 MB/s for AMD FX processors
Bottleneck in RAM speed reaching 30,000 MB/s for AMD FX processors
NB(IMC) at 2200mhz achieves a maximum read speed of 26000mb/s on the AIDA benchmark. At 2400mhz it reaches up to 30000mb/s, while 2600mhz gives 30000mb/s. This seems inconsistent, possibly due to limitations. DDR3 1600 offers around 23000mb/s, DDR3 1866 can hit 26000-27000mb/s, and DDR3 2133 reaches 30000mb/s. At 2400mhz it should be near 35000mb/s, but results are consistently 30000mb/s even at higher frequencies. It’s unclear what factor is restricting performance and whether adjustments can be made. System details: AMD FX-8350 4.5Ghz, 1.425v (medium llc), ASUS Crosshair V Formula Z. NB/HT at 2600mhz with CPU at 1.3v and regular LLC. G.Skill Trident X DDR3-2400mt/s (F3-2400C10-8GTX, Dual Rank). Source: https://www.gskill.com/specification/165...cification. BIOS timings: DRAM CAS# latency 10, RAS# to CAS delay 12, RAS# pre time 12, RAS# act time 31, RAS# read to PRE delay 7, RAS# write to read delay 6, CAS# write latency 8, write recovery 14, refresh cycle 300ns, row cycle 46, read to write delay 9, write to read delay DD 1, refresh rate 7.8ms, command rate 2.
It doesn't make sense here. The chip's official spec is 1866, not 2400 stable. The memory bandwidth for DDR3 on the 8350 is around 30 GB/s, which likely explains your performance cap. AMD doesn't offer a direct equivalent to Ark, so I found it in a B&H Photo listing.
it's very stable, I use 2400 10-12-12-31 with other settings on auto 10-hour y-crunch VT3. I can reach 2500mhz using junk timings, but that's the cap of this build, I didn't check stability at that speed. Occasionally after many restarts it runs at 32000mb/s with write, copy and improved latency. But it's uncommon. Another option to hit 32000mb/s is using 2700mhz NB, but I'm not sure if I'm stable on that. Oh right... on overview it mentions: up to 29.9GB/s memory bandwidth for DDR3 at spec it says: Maximum Memory Bandwidth 37 GB/s anyway where was I? Since sometimes with random memory traffic it reaches 32000mb/s, I probably hit the limit. Plus 2700mhz did also reach 32000mb/s so maybe combining them could push it to 34000mb/s? I tried HT-Link from 2600 to 2900mhz but it didn't help with RAM at all, HT-Link in general doesn't seem to improve things, which makes sense since PCIe 2.0 is limited by the spec (2.0). NB(IMC) from 2400-2600 doesn't seem to help with read speed, but for some odd reason at 2700mhz I get consistent 32000mb/s maybe if I fine-tune my RAM timings the motherboard will adapt better?
the details are not reliable, especially for cpus. The mentioned specs seem mismatched with actual performance numbers. older specifications often don’t match current realities, and many resellers push outdated kits. frequency overclocks can be misleading—true specs differ from what’s advertised. if you’re looking for a solid build, consider more recent parts rather than relying on older models.
I attempted to install DDR3-2666 or 2600 but couldn't locate any options beyond 2400, even with the latest versions. My only choice is to reduce the secondary timings at least, while aiming for good timing and keeping voltages under 1.65V.
Skip the benchmarks and focus on stability above all. Those high-speed Hynix chips will handle it fine at lower voltages, especially with a voltage above 11. I don’t believe you should rely on 4Gbit models unless you’re confident about the specs. Most 1Gbit and 2Gbit versions will improve noticeably with higher voltages, as seen in the overclock examples. It’s surprising you haven’t considered the real-world impact of RAM interfaces—your current setup seems a bit naive on that front.
For my systems, I’m sticking to the 2832C10 and 3400C11 overclocks. They outperform the cheaper options significantly, especially in timing precision for tasks like TRFC. These sticks are quite rare, but they’re worth it if you need better performance.
If you’re after more RAM, the 2832C10 Stable and 3400C11 Freq overclocks are definitely better than the generic XMP sticks you’re using. They run much higher and offer tighter timing, which is crucial for certain workloads.
Regarding the capacitors, they’re not reliable below 3000V. The cheap ones won’t last long. I prefer the ones I already have, but if you’re looking for more capacity, the Hynix CFRs are a solid alternative—they’re better than your XMP sticks, even though they don’t match the performance of the 4Gbit models.
For a daily build, you’ll likely find used AM4 boards at good prices. You could get a 3600+ B450+16GB setup for a similar cost, which would be worth reselling if you’re interested in the market.
If you want something more fun and capable, go with a 5500 or 5600G. The HMA81gu6djr8n (8Gbit) can reach over 5000 MHz, and the M378A1G43EB1 (4Gbit) is solid too. These will handle overclocking well, especially with modern RAM.
For better RAM, consider the X58, X79, X99, or LGA1150 options. They support decent overclocking and have good CPU performance for their age. If you’re looking for a daily build that’s both enjoyable and functional, these are solid choices.
Keep in mind, though—don’t set arbitrary volt limits unless you’re sure about the stability. That can ruin your gains before you even start. This hardware is affordable, and if you manage to find something better, it’ll be worth the effort.
This setup didn't perform well at higher voltages, so I moved on from that. I stayed with 2400mt/s since I don’t want anything below 8gb per stick. I’m not keen on anything under that speed, and honestly, I enjoy testing if stuttering (around 0.1% FPS) can be fixed by adjusting RAM timings. It’s more fun to see if a game runs smoothly on this build compared to others. I’m not interested in upgrading hardware unless necessary.
I searched for this kit, but most discussions focus on pushing it to 2500MHz with primary timings and ignore stability. No one mentioned secondary timings or 4GB per stick. I thought about getting a better case like a 420mm AIO to hit 4700-4800MHz, but my CPU can’t reach that high across all cores. That’s why I chose to keep the CPU at 4500MHz and experiment with different RAM options—like NB(IMC) and HT-Link—and now I’m getting solid results at 2600MHz.
I stress-tested it nonstop for a week, and nothing weird happened in games. The biggest challenge was figuring out the right timings. I tried tweaking them, like changing the R3 clock speed to 3100 and adding B550 + DDR4 16GB, but it didn’t really help. I ended up settling on 2400MHz with optimized settings.
I attempted to overclock the RAM to 2600MHz, but it wouldn’t go higher no matter the voltage. I could get it to 2500 with weird timings, but it didn’t seem worth it. So I decided to stick with 2400MHz and fine-tune the settings.
The main problem is I don’t fully understand the timing specs, and there’s no clear guidance on where to set them for stability. I’m a bit lazy, so if you have any tips or suggestions on the best timings, it would be really helpful. Changing just one setting—like RAS from 46 to 43—caused stuttering in games like CS2. With default settings, everything runs smoother.
Lastly, I tried to boost the RAM speed to 2600MHz, but it didn’t work consistently. If you can recommend a good pairing or timing configuration, that would make a big difference.
2400 runs a bit sluggish and the RAM is outdated. Try exploring the mobile options with a Hyundai CFR (HMT351U6C) chip—it should perform better than older Samsung chips and you might find it cheaper than the resale value of that 16GB kit your X58A U3 can handle at 2600 with multiple sticks. Stability checks will follow once I have a complete Hyundai CFR setup, since Samsung dies quickly on this platform for some reason. While it may detect mixed configurations, it tends to avoid Windows likely because of instability. Probably won’t test very low voltages unless you really need it, as the lower voltage won’t hurt performance much and could shorten CPU or RAM life. But with CFR it should boot at 2600 in 2DPC, which is a good sign. I’ll run benchmarks once I have a full set and see how it holds up.
For tuning, 1.8V is the bare minimum for decent DDR3 overclocking with 2.0V+ precharge. 1Gbit and 2Gbit modules work well at 1.65V, but 4Gbit Hynix should run smoothly at 1.65V for high-frequency use. Performance will be average due to subcomponents, but you can still enjoy solid results without major CPU or RAM wear.
If you disable half the cores (FX mode), benchmarks often improve and single-core stability increases. This could help with your setup, especially if you’re aiming for better gaming or productivity.
You might want to avoid all those low-voltage options—they usually hurt overclocking and can cause premature failure. Stick to higher voltages unless you’re certain about the RAM and CPU tolerances.
asus cvfz belongs to T-Topology. From what I recall, ASUS conducted tests showing support for 4x8GB RAM at 2400mhz. Based on that, I believe I can handle 4x8GB at 2400mhz with CR2 _________________________. I attempted to reduce the clock speed from 10 to 9, but it didn't work—I couldn't reach the first three timings (10-12-12) or even start. The fourth timing tRAS:31 works if I lower it as per CAS# guidelines; Latency:8 seems achievable. Regarding arbitrary voltages, I use 1.45-1.5V for CPU, 1.3-1.35V for CPU/NB, and 1.65V for DRAM. People often suggest 24/7 operation at 8V, but that increased heat is a concern. For voltage settings, I follow 1.45-1.5V, 1.3-1.35V, and 1.65V respectively.
My motherboard doesn't disable cores per module, which surprises me. Even if it could, disabling would affect performance on multi-core tasks. For example, with four cores, disabling one per module helps stability, but I'm not sure about the exact impact.
Each module has L1 96KB, L2 2MB, a single FPU, and shared tools I don't fully understand. If Windows runs a demanding program, it might struggle regardless of core count.
I used CPU power +130% in Phase Mode Standard, but extreme settings only caused more heat without improvement. For DRAM, I stick to 1.65V.
My motherboard doesn't support regular core disablement, though I'm unsure why. If it did, disabling one per module would help with multi-core workloads. For workloads like 4 cores, I'd prefer 8 cores for smoother operation.
Windows handles 4-core programs well—using one core per module is sufficient for most tasks. However, for heavy multitasking or gaming, 8 cores deliver better performance. I believe Windows manages cores efficiently, so a single core per module works fine in many cases.
For comparison, Intel CPUs like i5-3470 show solid single-core scores (127-129 cps, 467-472 cps at 3600mhz), while AMD offerings like FX-8300 or FX-6300 shine in multi-core scenarios. The FX series offers larger caches and better L2/L3 sizes, but power consumption remains a challenge.
In summary, one core per module is adequate for everyday use, especially with modern Windows optimizations. However, for demanding applications or gaming, dedicating more cores improves overall experience.