Corsair Vengeance 6000C30 OC on Arrow Lake
Corsair Vengeance 6000C30 OC on Arrow Lake
CPU performance: 245KF RAM: Corsair Vengeance 6000C30 2x16GB CMK32GX5M2B6000Z30 Mobo: Asus Prime B860M-K (2 RAM slots) Goal: how quickly can I activate the RAM? Speed matters more than exact timings at the moment. For future tests, I plan to use Prime95 8192k FFT benchmarks. I searched for guides on adjusting RAM speed on Arrow Lake, but found nothing useful. It seems like either advanced users already know everything, or the information is too basic. This is my first attempt at overclocking DDR5 here. My only other system is AMD, which doesn’t justify the effort. Also, this is my first time working with Arrow Lake. Tools indicate the chips are SK Hynix. I haven’t gone deeper than that, but the EXPO setup shows 6000C30. So far, I’ve tested RAM speeds and timings in EXPO mode, aiming for 1.40V XMP. The Prime95 results (6 cores, 64MB each) show scores around 4800-7000 depending on configuration. Each task uses about 64MB, so even one will hit cache limits. I ran each at least three times and recorded the best. SPD 4800 240.74 202.07 197.93 XMP II 6000C30 265.75 248.32 244.90 XMP I 6600 34-40-40-84-527-2T 272.19 259.34 254.21 XMP I 7000 38-48-48-90-558-2T (note: typo in bios, intended C36 but irrelevant now) 275.83 272.54 276.68 This is where I stand. I also completed a few minutes of Prime95 stress test without errors, so I’ll continue after sharing this. I need to focus on other tasks right now. Any suggestions on typical limits for this RAM type? I think there are ARL-specific tweaks with the tiles that could help later, but they’re less urgent. If anyone asks why I switched from XMP II to XMP I, it’s because XMP II loads the full profile and uses those values, while XMP I only uses primary timings and Asus handles the rest. Since I’m outside the profile, it seems unnecessary to load everything fully. Outside the profile, ASUS’ default worked best. I’m aware not everyone cares about Prime95 right now. Once I lock in settings I like for long-term use, I’ll run broader tests. Edit: I’ve read somewhere suggesting I’m running in gear 4—maybe I should try gear 2 later?
Retrieve a stick to check the version number. There are two versions available: A die (ver. 5.43.01) offering high memory speeds, and M die (ver. 5.43.13) which generally reaches about 7000MT/s. In the screenshot you shared, you're using Gear 2. On many motherboards, Gear 4 may not even boot.
Took a snapshot recently to preserve it for later retrieval. Might be nearing its capacity? https://rog-forum.asus.com/t5/technologi...-p/1054808 I followed the details on the link. It indicated mem controller clock equals RAM clock = gear 2. Mem controller clock is 1/2 RAM clock = gear 4. The real timing isn’t measured in MT/s. I experienced a similar situation with my AM5 setup. Mobo uses half controller clock with 2R modules, matching only with 1R when needed. I had to adjust it manually. I should verify the version of the 2R modules now! Edit: My two 32GB modules are likely the same type—should they be 2R? Should I take them from my AM5 system and replace them with the M die 2x 16GB? The aim is to boost RAM performance on the Intel platform. Since the AM5 runs X3D, it may be less affected by a rank reduction. I haven’t done a thorough test yet, which is why I have these two configurations.
Yup, thats M die then. Unless you've got a really high end board and do a lot of tuning, getting above 7000 is very difficult basically impossible. Back when I owned an M die kit I couldn't even get it to boot above DDR5 7000, let alone be anywhere near stable, with the highest I've ever seen anyone get it 24/7 stable being 7200MT/s. Weird, with every other Intel system I've seen, that logic would describe Gear 1 and Gear 2 respectively. My 13900K system I know is in Gear 2 it reports the memory controller frequency as half the DRAM frequency EDIT: That might be interesting to test then. Yeah, your 32GB modules are going to be A die, though I don't actually know if you would be able to get much higher frequency out of the 32GB modules because they're dual rank. I have gotten those kits to boot all the way up to 8000MT/s on an APU system, but highest frequency I've been able to even make it into Windows with was 6800 so I don't have high hopes that you'd get much past 7000MT/s with it. That said, I haven't really seen anyone test dual rank on Arrow Lake, so it could be that the memory controller is stronger enough that you can acually get dual rank sticks to decent frequencies.
I haven't performed detailed stability testing, but it appears stable for over ten minutes on Prime95 when stressing RAM (as seen at the 7000 settings previously). It might be related to Asus' design choices. If you didn’t notice, I made some adjustments at the end. I also have the identical 2R kit from an earlier version, which could indicate a different die. What might that result in? A forum user noted that an A die can push performance higher—around 7600-8000 on Intel chips, while a M die usually caps near 7000-7200 but often with superior timing.
Asus handling their projects wouldn’t come as a surprise, though it might just reflect reporting differences from Arrow Lake. I’d still download MemTweakIt to verify if you’re using Gear 2, just in case there’s some uncertainty. I noticed the edit too—I made changes to my post right before you posted this if you’re interested. Yes, that aligns with what I expected. The performance boost from the die is a bit exaggerated; only TRC2 shows improvement on the die, while everything else remains comparable or better on the A die (though there’s still some variation between kits).
I’ll look for that tomorrow and check the outcome. Done! I read somewhere about timing variations not just between chips but also how AMD and Intel controllers interact. It’s more complex than I expected. The main goal here is to get a better RAM option for my Intel setup instead of the AMD one I had available. I also saw a note that a 24Gb M die outperforms a 16Gb M die, though I don’t have the 24Gb one. Remembering when 24GB modules first showed promise in overclocking, I won’t make any changes right now. My to-do list for tomorrow: install MemTweakIt, test pushing the M die kit past 7000, and compare with the A die 2R kit if possible.
These are the components currently in my setup, timing-wise it resembles a standard A die but with lower tRFC and tRAS values. It tends to run a bit faster, though that’s not enough to surpass a regular A die in performance. If you’d like, I can share timing screenshots for reference so your build can be based on them. I have these for all major Hynix memory chips—most are 13th generation, which could affect results compared to Arrow Lake. In short, AMD adjusts tertiary timings differently and uses tRFC1 instead of tRFC2. While it sounds like Intel fine-tunes RAM and AMD does too, in practice they often match internal subtimings, leading to similar results across systems.
I have some background on RAM replacement, though I haven’t worked with DDR5 before today. My knowledge is limited to what’s already established, and DDR5 chips are still emerging, especially in the Arrow Lake lineup. At the end of the matter, it’s the same components. I’m focusing more on EXPO and XMP settings. The module was primarily designed for AMD systems, with a slower setting of 6000C30 as the main adjustment point due to speed constraints. My main concern is whether I can achieve higher bandwidth while maintaining similar latency. For real-world performance, I rely on practical tests with Prime95, which gives more insight than generic benchmarks like Aida64 that don’t reflect actual usage patterns. From my earlier experiments, raising the settings from 6000C30 to 7000C38 in XMP seems to improve Prime95 results by about 4%. There’s still room for further optimization, and adding E-cores might unlock even better gains. The built-in benchmarks typically use only P-cores, so including E-cores could provide additional benefits. Interestingly, the EXPO and XMP profiles in both kits match in their parameter settings, yet they differ slightly—EXPO includes a minor adjustment that AMD adapted from Intel’s approach but tweaked it slightly.
I've adjusted several screenshots to 7000 and focused on comparing like primaries. The M die I used performed much better overall; my previous set was quite poor. I'll share a screenshot for your 24Gb M die when you're ready, though the performance numbers are pretty close to my A die with looser tRRD_sg and tRFC.