- Thread Starter
- #301
At lower clocks, the results of dual-rank kits are about the same as those of 2 memory ratio higher single-rank kits. In the comparison in my last reviews, 2x32GB DDR5-6000 CL32 is about the same as 2x16GB DDR5-6400 CL32. You can expect something similar at higher clocks, so DR 7600 will be about the same as SR 8000. This is assuming that the software can make use of dual-rank. Most software doesn't react to that at all. On the other hand, most software barely reacts to anything above ~7200, so the average will be DR 7600 = SR 8000.
Used benchmarks cover various scenarios, from low to high display resolutions, rendering, office applications, and some more. Out of synthetic bandwidth tests in AIDA64 and games at 1080p, results are +/- 7%. As long as getting ~7200 RAM is a good idea, then going for the fastest kits is a waste of money.
As you said, the difference was more significant in DDR4. I guess it's because the CPU cache is getting larger and faster, so you can't see the delays if RAM is slower. In AMD X3D, where there is a huge cache, you can barely see any difference, no matter what RAM is used.
I was comparing 2x24GB vs 2x32GB on 7800X3D, and the results were the same. Even the difference between tight and relaxed timings was barely visible in anything. In Intel, dual-rank is faster but not fast enough to waste a lot of money on 96GB kits. I also know some other users who bought dual-rank kits for slightly better results on Intel.
The last time I asked V-Color for their top 96GB kit for review, they said yes, but it wasn't available for 2-3 months. It's still the same IC as in the 6000 CL32, 6400 CL32, or 6800 CL34 kits. Maybe there is a matter of binning, but I see that in DDR5, it's not so important, and as long as RAM uses Hynix A/M, then it overclocks high. All popular brands use the same Richtek PMIC with Hynix A/M-die. All have high-quality PCB. In these top-speed kits, the only difference is the XMP profile (additional timings) and RGB/heatsink design.
It's hard to get a 96GB kit for review. Most brands have limited marketing budgets, so they would rather send two to three 32/48GB kits to different redactions than one 96 GB kit.
Most brands moved with all new series to 24/48GB modules. If they had new 2x32GB kits, then I would say it's the best option as it would be cheaper than 96GB and overclock slightly better ( the same clock but easier to set tighter timings). On the other hand, every new BIOS feels optimized for 24/48GB modules. The same on AMD and Intel; I could easily set higher clocks on these non-binary Hynix M kits than on Hynix A. The difference is not so significant, but on the last 3 Hynix A/SR kits, I couldn't run tests at more than 8200-8400 when every Hynix M/SR kit goes up to 8400-8600.
Used benchmarks cover various scenarios, from low to high display resolutions, rendering, office applications, and some more. Out of synthetic bandwidth tests in AIDA64 and games at 1080p, results are +/- 7%. As long as getting ~7200 RAM is a good idea, then going for the fastest kits is a waste of money.
As you said, the difference was more significant in DDR4. I guess it's because the CPU cache is getting larger and faster, so you can't see the delays if RAM is slower. In AMD X3D, where there is a huge cache, you can barely see any difference, no matter what RAM is used.
I was comparing 2x24GB vs 2x32GB on 7800X3D, and the results were the same. Even the difference between tight and relaxed timings was barely visible in anything. In Intel, dual-rank is faster but not fast enough to waste a lot of money on 96GB kits. I also know some other users who bought dual-rank kits for slightly better results on Intel.
The last time I asked V-Color for their top 96GB kit for review, they said yes, but it wasn't available for 2-3 months. It's still the same IC as in the 6000 CL32, 6400 CL32, or 6800 CL34 kits. Maybe there is a matter of binning, but I see that in DDR5, it's not so important, and as long as RAM uses Hynix A/M, then it overclocks high. All popular brands use the same Richtek PMIC with Hynix A/M-die. All have high-quality PCB. In these top-speed kits, the only difference is the XMP profile (additional timings) and RGB/heatsink design.
It's hard to get a 96GB kit for review. Most brands have limited marketing budgets, so they would rather send two to three 32/48GB kits to different redactions than one 96 GB kit.
Most brands moved with all new series to 24/48GB modules. If they had new 2x32GB kits, then I would say it's the best option as it would be cheaper than 96GB and overclock slightly better ( the same clock but easier to set tighter timings). On the other hand, every new BIOS feels optimized for 24/48GB modules. The same on AMD and Intel; I could easily set higher clocks on these non-binary Hynix M kits than on Hynix A. The difference is not so significant, but on the last 3 Hynix A/SR kits, I couldn't run tests at more than 8200-8400 when every Hynix M/SR kit goes up to 8400-8600.
It overclocks up to 7600MT/s CL40 1.35V, so significantly better than Micron-based UDIMMs. On the other hand, it costs almost twice as much, so it's hard to recommend right now.
