Latency in DDR5 seems excessive.
Latency in DDR5 seems excessive.
I reviewed some articles about DD5 modules after they were released. A key issue that affects performance is the excessive latency—36ns is quite high for system RAM, even with newer memory types. I’m not constructing a new rig, but I still anticipate decent speed. I’ll need to revisit the material, remembering DDR5-6400 specs. Given the expensive price and limited availability, it seems unlikely anyone will purchase soon.
You haven't worked with a Ryzen CPU from the 5000 series before, and a 55ns latency is roughly the best you can achieve with stable daily configurations. Where did you find the 36ns figure? The estimated initial latency for slower 4800MT/s CL40 models is around 16.7ns, while actual core latency tends to be near 60ns or more on the slower units. Part of the slowdown for DDR5 comes from the memory controller operating at half speed (Gear 2), meaning data moves only half as much compared to DDR4 in Gear 1.
It's brand new. DDR4 cost the same when it launched. Kits were around $600, while I own a set for $350. Now it's available at much lower prices. Because DDR4 is running out of bandwidth, it can't keep up with the needs of newer CPU designs. Check how it performs from past to present. A CPU with more cache prefers high-speed memory. "Latency issues" happen in nanoseconds—measuring that takes just a thought. 50 ns means one billionth of a second. Worrying about that is overkill. "Oh no, my latency is 50 billionths of a second..." Haha, you get the idea. Focus on speed rather than latency. Latency connects to CPU design, and overclocking helps too. A 40 MHz clock at 6400 MT/s equals 3200 MHz, which DDR4 handles at half that power. It's interesting how 12th gen supports both DDR4 and DDR5, but companies must balance sales and performance. That was a smart move in the market.
The 36ns figure comes from the documentation and videos about DDR5-6400. Keep in mind its slower speed compared to DDR4-3200. It's wise to hold off until DDR5 matures, prices drop, and it gains broader adoption.
Upcoming platforms in the next two years... DDR4 will reach end-of-life.
13th generation will probably stick with DDR4, but after that it’s likely to move to DDR5. I’m pretty sure when the 14th generation arrives, DDR5 will become more valuable because of its higher bandwidth, even though latency stays similar. DDR5 is mostly the same as DDR4 in delay but offers much faster data transfer. If bare PCBs start selling well, maybe we should upgrade to DDR5 earlier in the 14th gen.
It will take a while to process. I still prefer DDR4 since I have other things to invest in, like window tint and wheel spacers for my 02 MK4. I just put in a new Double Din 10" Android 11 quad core stereo. It has a nice little 10" sub in the trunk and also helps with weight reduction.
I certainly can't read the future, but the past tells quite a bit. I have a lot of old hardware. The point where generations lapsed. Even had an Old AsRock board that was s939 and with a Riser card could install AM2 processors and ddr2 memory. It really was a creature of sorts. I blew my FX-55 and the board up with a volt mod. lol. It was AsRock Dual Sata II 939. What a killer board and time to have been into PCs and overclocking. A lot of interesting stuff has happened during transitions. People really should try and embrace all they can. Understand it, not just know it. Here's the transition. s939 to AM2 4000+ Double the L2 cache, leaving DDR1 for DDR2 and secure the bandwidth for dual cores. They decided, double the cores, double the bandwidth. Frequency from 200mhz to 400mhz. Latency, shit about the same as it's always been. Within a certain threshold they design the Cache to operate. Be it frequency or architecture change or both. Architecture stays the same below - Doubled the cache, double the core count, doubled the memory bandwidth. s939 https://www.cpu-world.com/CPUs/K8/AMD-Athlon 64 4000+ - ADA4000DKA5CF (ADA4000CFBOX).html sAM2 https://www.cpu-world.com/CPUs/K8/AMD-Athlon 64 X2 4000%2B - ADO4000IAA5DD (ADO4000DDBOX).html So why not go back to X58 triple channel? It's not cost effective to justify the performance gains. So it didn't take off. When a pro gamer spends 5,000 to over 10,000 US dolla, This small elite company might buy into ThreadRipper. Cause well, money is plentiful and the hardware is available. But you don't see a lot of talk about thread ripper processors much. Niche market. DDR5 will be mainstream. Probably for 4 or 5 years if I where to guess. Just depends when the CPU needs cap out the available bandwidth. Oh, and same goes for your video cards and hard drives. PCI-E lanes, nobody talks about the latency here.... they just want PCI-E 4.0+++ cause FASTERER. 90% just simply don't know why.... Really all this is the same as it's been in the past. Hasn't changed much. The frequency clubs keep changing though. 4ghz club. 6ghz club, 8ghz club. DDR4 3000mhz ddr4 6000 club. Now you want to join the DDR5 8000mt/s 4000Mhz club.... But yeah. pretty much that stuff. sorry for the ramble.
I'm still trying to understand where that figure came from. It seems way off compared to the expected latency values. In the tools I use—like Intel MLC and Aida64 tests—I see 36ns as a solid, optimized benchmark. Those are where stable overclocking really shines. It would help to reference those specific studies you mentioned. I know DDR4 3200MHz CL16 and DDR5 6400MHz CL32 should match up closely in latency, but DDR5 is expected to be significantly faster. Mostly because the Alder Lake controller isn't as smooth with DDR5 yet, and prices are dropping. Once AM5 arrives, we should see better memory management and more advanced designs, making DDR5 adoption more practical. It's still early, but it's definitely progressing. The main hurdle was DDR4's exceptional performance; its timing was nearly perfect, and its speed dropped quickly. DDR5 will eventually surpass it, but it's going to take time before it becomes a clear choice for high-speed builds.