Alder Lake Quad DDR5 Overclocking Explored
The first commercial DDR5 platform to reach consumers, Intel’s Alder Lake has achieved quite the reputation in overclocking…as long as only two DIMMs are used. Conversely, our tests have shown the performance benefit of additional memory ranks. Could more be better than faster? Hoping to find something useful, we tried adding a second set of Corsair’s DDR5-5600 last winter…only to learn that it wouldn’t even run at its XMP values.
Of course we’ve learned a little about overclocking since. Yet even today, we’re only getting those two kits as far as DDR5-5000 before they collectively crap out. At least that lower data rate left enough stability to drop the timings to the XMP ratings with all four DIMMS installed.
Moving on, we noticed that Sabrent’s supremely overclockable DDR5-4800 used the same ICs as our Patriot Viper Venom RGB DDR5-6200. How far could we go if we paired these?
As it turns out…not too far. But we did get timings down to CAS 32 at DDR5-5200, so we’re looking forward to the performance bump.
We should probably let everyone know that we limited our overclocking attempts to 1.35V VDD and VDDQ and did try various “in between” voltage levels, so there’s plenty of room for advanced overclockers to try something else. Having said that, the best we’ve seen so far of an XMP-rated quad-DIMM kit is DDR5-5200, so we’re fair comfortable with the results we achieved in our Patriot/Sabrent mix of Hynix H5CG48MEBD-X014 ICs.
|CPU||Intel Core i9-12900KF: 16C/24T, 3.2-5.2 GHz, 30 MB L3 Cache, LGA 1700|
|CPU Cooler||Alphacool Eisblock XPX CPU, VPP655 with Eisbecher D5 150mm, NexXxoS UT60 X-Flow|
|Motherboard||ASRock Z690 Taichi LGA 1700|
|Graphics||Gigabyte GeForce RTX 2070 Gaming OC 8G: 1410-1725 MHz GPU, GDDR6|
|Power||be quiet! Dark Power Pro 10 850W: ATX12V v2.3, EPS12V, 80 PLUS Platinum|
|Hard Drive||TOSHIBA OCZ RD400 256GB NVMe 1.1b SSD|
|Sound||Integrated HD Audio|
|Graphics Driver||GeForce 496.49|
The first thing to note is that our ASRock Z690 Taichi sample depreciated both the Corsair and mixed Hynix kits to DDR5-4000 when all four slots were filled, and retained C40 for both when manually configured to DDR5-4800. No further tweaks were required to run these at DDR5-4800, so we reserved our efforts for the DDR5-5000 (Corsair) and DDR5-5200 (Hynix) overclocks.
SiSoftware Sandra shows miniscule bandwidth gains for the presence of four DIMMs at the same DDR5-4800 setting, but the latency didn’t get worse either. AIDA64 shows an increase in write bandwidth.
3DMark’s CPU score shows the benefits of faster memory, but not much else.
7-Zip shows the benefits of four ranks of memory in its file decompression test. Handbrake also finishes its encode a bit more quickly with four DIMMs, and Corona Benchmark even produces a few more rays per second.
The basics in our timed benchmark average are seen at DDR5-4800 C40, where four DIMMs are 5% faster than two. Overclocking gets us even farther.
That timed benchmark average chart brings us to our most-important conclusion: More memory is better, but not better than a higher data rate. The best solution we can think of, then, is one that we didn’t have the hardware to test: G.Skill offers a 64GB kit that contains two double-sided DIMMs, rather than the four single-sided DIMMs that we tried here, at DDR5-6000. You’ll be the first to know if/when we’re able to procure it.