Hi, this is Wayne again with a topic “Is DDR5 FINALLY Worth It On Alder Lake?”.
Really ddr5: how interesting can that be? Well, we can ramble get it ramble. [ Applause, ], with the launch of a new socket lga 1700 intel’s alder lake has come to life. It’S hard to believe that it’s been almost six months and that brings with it ddr5 ddr5 uh the track record. So far, it’s not been great supply issues. What has been supplied in the market? Not really all that fast and performance, and not a lot of different over ddr4 that is finally finally starting to change this kit of a data memory.
Xpg is the fastest ddr5 kit that i’ve tested so far, ddr5 6000 and if you were shopping for ddr4 memory in the past, you’d know that sometimes they played a little fast and loose with the timings with ddr4. It was possible to get a ddr4 3600 kit that was actually slower than a ddr4, 3200 kit, at least when you benchmarked it when it came to the games, and the reason for this is the timings most of the time. The memory is advertised with at least the first timing. Number gamers nexus actually did a good video on the complexities of the timing, and you know it’s like you – have a row change time and a command time and what’s the command rate cr1 cr2.
I don’t really want to rehash that, although that might be an interesting video, because ddr5 is a little bit more nuanced, because there’s two channels that are half as wide on each memory. Stick, it’s not really exactly correct to call it. I mean it is four channel, but each channel is only half as much bandwidth as ddr4, so it’s effectively the same bandwidth and they do that for latency reasons, but yeah anyway, i’m getting off track, ddr5 6000. So when i tested this, i was expecting it to really not be a lot different than other kits of memory.
You know we’ve sort of launched with uh ddr4. You know 2400 4800, 2400, megahertz, 4, 800 mega transfers per second uh is sort of what we launched with so from 4 800 to 6000 is a pretty big uplift, and i was kind of surprised when we’re working with you know a 30 90 or a 6900 Xt gpu the very highest end of the market right now and we’re talking about things like 1080p gaming, the extra memory does actually help it can be as much as two three percent difference for those high frame rates, so i decided to do some comparisons. I’Ve got some kingston hyperx memory, that’s 5200.
I’Ve got some 5600 g skill, trident z, and this running at 6. 000.. Now i really expected it.
You know, like i say, for there not to be a lot of difference between, say. You know: 5, 600 and 6. 000, but surprisingly, there was because in ddr4 i’ve come to expect shenanigans, but there wasn’t a date has actually got a good product here, the secondary timings and the tertiary timings on this look pretty good, as evidenced by the actual transfer rates in a to 64.. So if you look at this kind of a benchmark and you side by side by side each memory, you can see exactly what you’re getting in terms of performance uplift and you can see exactly what you’re getting in terms of latency latency. I think is also very important and a lot of people don’t really talk about latency when they’re doing these kinds of benchmarks, but we got all the way down to about 71 nanoseconds from over 80 nanoseconds in the baseline configuration now for the baseline testing. That was a stock jet x standard jet is the standard that you know, sort of ddr5 is implemented at is the completely standard 4800 mega transfers, and if you notice 1.1 volts, when you want to run faster, you need more voltage.
So in the spd in the profile, as reported by cpu-z, we can see. Oh look. It’S going to bump it up to 1.2 volts in terms of the ability to tune the memory and get a little bit better performance. You can get down to about 68.
69 nanoseconds, i don’t think it’s worth your time to spend a lot of time trying to dial into your your memory, especially if you use your machine for something other than gaming, because it can introduce instability, one of the main places that the instability will get introduced At is wake from sleep. So when the machine goes to sleep, the memory is not exactly off, but it’s not exactly on either and that critical sort of wake up thing. Uh, it’s really hard to get right and ddr5 was a little rushed and so especially early kits like away from wake from sleep. Your computer doesn’t do anything you might be pushing the overclock just a little too much. Sometimes the bios update will help with that. Sometimes there’s new timing tables for manufacturers that will help with that, but in general yeah. Now our test system was with the msi supreme 3090, one of the highest end, graphics cards you can get and on the team red side we were testing with the asrock 6900 xt. No stability issues, no performance issues, no weird glitches, no gotchas. So i’m very satisfied with this kit of memory from a data there’s nothing untoward going on under the hood, and it really is ddr5 6000. It is worthy of the ddr5 6000 label even offering a little bit of a performance uplift over our ddr5 5600 kit. I mean it’s only 400 extra mega transfers – another 200 megahertz, but it was actually surprisingly stable on our 12 900 kf. This is a recently produced 1200 kf as well.
So it’s not quite the launch day. 12. 900.. I thought it would mix things up a little bit and we’re testing on an aorus master motherboard, but i also tested with the msi ek carbon x, motherboard, no problems there. Actually, msi has memory, try it and i was able to get a little bit better performance out of it overclocking and doing things on that motherboard than this motherboard.
I’Ve got an upcoming build based around the asus strix z690, and i’m going to use this kit of memory in that build we’re going to probably not have as high end of a gpu as a 30 90 in that build. So i’m not expecting there to be a you know, a huge performance difference, but we’ll take a look. So look for that video, so the short version is i’m pleased with this kit of memory from a data it is as advertised pleasantly surprised.
I look forward to testing even more kits of memory, because things are only just now starting to get interesting with ddr5 5600. 5400 5600. Something like that is basically the floor and we’ve got so much more room to go from here.
Maybe we’ll get back to sub 65 nanoseconds as the norm. I can dream, can’t i uh one thing i wasn’t able to do with the overclocking was clear: a hundred gigabytes per second transfer. I got really close. You know when we started out with 2400, and we look at the first number here from 8-64. It’S basically how much? How much information can you transfer in a second and the 100 gigabyte per second barrier will be an important one when it’s finally broken out of the box.
This kit doesn’t break that barrier, but it comes dangerously close and i think, with a little tuning, i could get it hitting over 100 gigabytes per second and have it be stable for more than an hour but uh again, you know how what’s your time worth. Is that is that time well spent or or is that just bragging rights in a you know in a benchmark? But still this is almost 20 above baseline when we’re talking about ddr5 and quite the gap between the highest end, ddr4 memory kits that are available and with just two dimms we’re not really talking about a huge difference over ddr4 except the transfer rate, the transfer speed. So big benefit there. If you do decide to overclock the memory a little bit i’ll, give you a little bit of a warning.
You can achieve an overclock that is stable, but the overclock is worse performance than something that is slightly worse numbers than the overclock, and the reason for that is because, even though the ddr5 memory is not error, correcting there are error, correcting mechanisms that are implemented in Non-Error, correcting ddr5 ddr4 introduced some basic, you know: retransmit retry functionality, ddr5 improves that retransmit retry functionality, and so you can have an overclock that is perfectly stable because it’s having errors, but it’s retrying. You lose time when you’re retrying, the latency, will be a little bit worse. Your game might hitch or stutter a little bit more, even though you run something like 8-64, because 8-64 takes so long to run on average. Over that long run, it looks like you’re doing really well, but when you’re playing a game, those spikes as you go you’ll feel them and it will be a very bad experience.
So don’t overclock too much. You can get yourself into trouble, make your system unstable and it can be hard to diagnose, because the air correction mechanism is doing error, correction and it doesn’t really have a clean way to bubble those errors up to the top to say: hey, stupid, you’re. Getting a lot of memory errors, because then people would freak out, because under normal operation, you’re still going to get memory errors every now and then and people would think something’s wrong. So it’s like! Oh, we need to hide it from the consumers.
Ah, like i said, even though this is not air correcting memory, air correcting memory adds another layer on top of this. An air correcting ddr5 is a thing that exists and it may even be unlocked on alder lake in certain chipsets. I’M investigating that. I think that’s a bios oversight, but we’ll see i’m well.
This is level one. This has been a fun chat or ramble. If you will about a data, ddr5 xpg memory running at six thousand mega transfers per second i’m signing out.
You can find me in the level one forums .