1 00:00:00,000 --> 00:00:08,800 It seems this RAM crisis just won't stop. Earlier today, a customer's wallet was brutalized 2 00:00:08,800 --> 00:00:15,400 at the local computer's RS. Live at the scene is special correspondent Riley Murdock. 3 00:00:15,400 --> 00:00:20,320 Thank you Linus. The scene here is chaos. There is affected customers everywhere. Excuse 4 00:00:20,320 --> 00:00:25,240 me, excuse me sir. I see you're holding a high speed DDR5 kit there. Can you explain 5 00:00:25,240 --> 00:00:28,320 why you paid extra for that instead of a lower speed kit? 6 00:00:28,320 --> 00:00:32,440 I don't know man. Everybody tells me I need the more expensive stuff, but it's just so 7 00:00:32,440 --> 00:00:37,800 expensive. I'm just happy they didn't take my other kidney. 8 00:00:37,800 --> 00:00:41,240 Should we tell them? Are you a gamer, sir? 9 00:00:41,240 --> 00:00:48,400 Yes. Then we've got some fantastic news for you because we tested 4 CPUs at 8 different speeds 10 00:00:48,400 --> 00:00:53,960 of DDR5, running as slow as 4800 megatransfers per second, and it turns out you might not 11 00:00:53,960 --> 00:00:57,240 need to spend as much on memory as you thought. 12 00:00:57,240 --> 00:01:03,280 Ooh, that does not look great. Good thing he was born with two of them. But hey, it's 13 00:01:03,280 --> 00:01:09,440 not all bad. Let me cheer you up with this segue to our sponsor. Can we get this guy 14 00:01:09,440 --> 00:01:16,920 some help? The TLDR of our extensive testing is that while some CPUs do obviously benefit from 15 00:01:16,920 --> 00:01:23,360 faster memory, AMD's recent statement that their popular X3D gaming CPUs are super down 16 00:01:23,400 --> 00:01:28,920 to shack up with even the slowest memory seems to be pretty much accurate. We found that 17 00:01:28,920 --> 00:01:33,800 at typical gaming resolutions and settings, there was almost no difference between kits 18 00:01:33,800 --> 00:01:40,400 that run at industry standard or Jdex speeds and faster overclocked gaming RAM. Now, this 19 00:01:40,400 --> 00:01:45,480 is not going to completely solve the problem. Even slower memory has been impacted by recent 20 00:01:45,480 --> 00:01:51,320 price increases. But hey, at least it gives us some options. Let's find out then how 21 00:01:51,360 --> 00:01:57,280 much performance we might be giving up, starting with Intel, where unfortunately the story 22 00:01:57,280 --> 00:02:03,800 is not quite as rosy as AMD's X3D chips. We chose both a last gen and a current gen 23 00:02:03,800 --> 00:02:08,120 representative and we used our standard Labs test matches. Testing took forever by the 24 00:02:08,120 --> 00:02:12,800 way because we had to use the same memory kit across each platform, which meant swapping 25 00:02:12,800 --> 00:02:16,680 out a lot of RAM. Let's start with gaming on Raptor Lake. 26 00:02:16,680 --> 00:02:21,080 Being strictly at our geomean across all games at 1080p, it's pretty clear that our 27 00:02:21,080 --> 00:02:28,240 14900K loves fast DDR5. So if your grandma made a strategic DDR5 investment last year 28 00:02:28,240 --> 00:02:34,120 and you can get some 8400 megatransfer per second CL40, to the moon it goes! At least 29 00:02:34,120 --> 00:02:39,120 in certain games, like Ashes of the Benchmark, Cyberpunk, and Red Dead Redemption 2. The 30 00:02:39,120 --> 00:02:43,960 good news is though, as long as you don't pair it with the slowest kit you can find, 31 00:02:43,960 --> 00:02:49,520 it won't suffer too much. As for Intel's newer Core Ultra 285K, good news! It couldn't 32 00:02:49,520 --> 00:02:53,520 even run our fastest kit. And as for the rest, as long as you're using something that's 33 00:02:53,520 --> 00:02:59,000 faster than JDEC, basically don't worry about it. When we look at the all-important 1% 34 00:02:59,000 --> 00:03:05,120 lows, there is at most a handful of FPS difference across every game and resolution that we tested 35 00:03:05,120 --> 00:03:09,840 to the point where I would say that unless you have an FPS counter in the corner, you 36 00:03:09,840 --> 00:03:14,200 would never know the difference. And this is especially true as your resolution and 37 00:03:14,200 --> 00:03:18,720 your in-game details scale up. Now, you're probably wondering at this point, if these 38 00:03:18,720 --> 00:03:23,240 differences are so small, why were all of you reviewers acting like REM speed mattered 39 00:03:23,240 --> 00:03:29,640 before? It's all about the context, guys. When we were talking about 20 or 30 bucks 40 00:03:29,640 --> 00:03:35,640 to get a few percent performance improvement on your 1500 or 2000 dollar gaming PC, that's 41 00:03:35,640 --> 00:03:39,760 a no-brainer. Who doesn't want bang for the buck? But we're not talking 20 or 30 42 00:03:39,760 --> 00:03:46,400 bucks. We're talking 200 or $300 for a higher speed kit. That changes the math completely. 43 00:03:46,400 --> 00:03:51,440 Or does it? Maybe productivity will tell a different story. Okay, not in Blender. But 44 00:03:51,440 --> 00:03:58,000 our Godot compile test does see a decent improvement with higher speed memory on the 4900K. And 45 00:03:58,000 --> 00:04:02,880 that's not a completely isolated case. While many of our productivity tests do mirror our 46 00:04:02,880 --> 00:04:08,160 gaming tests across both Intel's current and last gen chips, there are a significant 47 00:04:08,160 --> 00:04:13,280 number of results here where spending far more on your memory can yield a meaningful 48 00:04:13,280 --> 00:04:17,480 difference in performance to the point where, if you're making money with your machine, 49 00:04:17,480 --> 00:04:22,640 you might want to consider ponying up for a high speed low latency kit, even at today's 50 00:04:22,640 --> 00:04:28,520 extortionate pricing, which is exactly the problem that consumers are facing right now. 51 00:04:28,520 --> 00:04:34,120 You want to play video games and you gotta budget accordingly. Well, Joe Workstation user, 52 00:04:34,120 --> 00:04:38,920 he's not playing games. He wants to compile his code faster. So he submits a requisition 53 00:04:38,920 --> 00:04:43,200 to the IT department who goes, yeah, if it'll make you more productive, go for it. We're 54 00:04:43,200 --> 00:04:47,240 gonna have a companion LTTLabs.com article, by the way, that dives a little deeper into 55 00:04:47,240 --> 00:04:52,640 which business use cases might benefit and which ones won't. On the subject of business, 56 00:04:52,640 --> 00:04:56,600 check out the new sand colored tech pants at LTTstore.com. They offer loads of utility 57 00:04:56,680 --> 00:05:00,920 with hidden pockets galore while keeping stylish on the job. Now let's change gears 58 00:05:00,920 --> 00:05:05,640 and talk a little bit about AMD. Don't you guys love it when twice a day the corporate 59 00:05:05,640 --> 00:05:12,000 puffery clock gets it right? Just look at those beautifully aligned graphs on our 9800X3D. 60 00:05:12,000 --> 00:05:17,600 Okay, there is a little bit of variance. Our 4800 megatransfer per second kit does sit 61 00:05:17,600 --> 00:05:24,760 in last place and 6000 megatransfer CL30 is still the sweet spot, but turn on ray tracing 62 00:05:24,840 --> 00:05:30,280 or up the resolution and things become even more samelier than before. Like on Intel, 63 00:05:30,280 --> 00:05:35,720 Cyberpunk and Ashes show the biggest spread in our 1080p results, but it's a much smaller 64 00:05:35,720 --> 00:05:41,520 gap and realistically, there are cases here where using 16 gig versus 24 gig modules is 65 00:05:41,520 --> 00:05:47,400 going to make as much of a difference or more. 3DV cash for the win. As for the regular 66 00:05:47,400 --> 00:05:53,600 cash, it might be for the win too. It turns out that while our 9950X did stumble a little 67 00:05:53,640 --> 00:05:58,640 bit with our slowest JDEC kit, overall, as long as you can get your hands on something, 68 00:05:58,640 --> 00:06:05,480 anything with a tuned expo profile, you probably won't notice a difference, especially at risk 69 00:06:05,480 --> 00:06:09,880 of sounding like a broken record, at risk of sounding like a broken record, especially if you 70 00:06:09,880 --> 00:06:14,480 turn up your in-game details or kick your resolution up to 4k. Moving on to productivity, 71 00:06:14,480 --> 00:06:18,760 we've got a similar story to what we saw on Intel, just through a different colored lens. 72 00:06:18,920 --> 00:06:23,320 The JDEC kit is definitely the slowest, demonstrating that overclocked memory does 73 00:06:23,320 --> 00:06:29,320 have a reason to exist, just not at the current prices. Beyond that, there were occasional cases 74 00:06:29,320 --> 00:06:34,680 where faster memory or especially sweet spot memory that runs in sync with AMD's Infinity 75 00:06:34,680 --> 00:06:40,600 Fabric at nice low latencies did result in measurable gains, but outside of those specific 76 00:06:40,600 --> 00:06:45,000 use cases, like Godot, the differences are not going to be enough to ruin your day. 77 00:06:45,080 --> 00:06:51,400 So in conclusion, don't be this guy. If you're a gamer and high-speed memory is available for a 78 00:06:51,400 --> 00:06:57,000 modest price delta, it can make a lot of sense, especially if you're on an older Intel platform. 79 00:06:57,000 --> 00:07:02,520 But right now, I just can't recommend it, especially to the tens of thousands of you 80 00:07:02,520 --> 00:07:07,240 that are buying X3D processors for gaming. By the way, we're going to have X3D processors and 81 00:07:07,240 --> 00:07:10,520 all the stuff that we used on our benches linked in the video description if you guys wanted to 82 00:07:10,520 --> 00:07:14,760 check something out for your next rig, just don't expect any great deals on the RAM down there. 83 00:07:14,760 --> 00:07:19,720 As for the non-gaming folks, man, I really feel for the cottage industry professionals right now, 84 00:07:19,720 --> 00:07:27,160 the folks who do work at home that requires faster RAM or especially a lot of RAM. Like, 85 00:07:27,160 --> 00:07:31,480 it's hard to care too much about corporate's workstation budget, but if you're an at-home 86 00:07:31,480 --> 00:07:38,200 developer or you're a contract creative professional, it is a super sucky time to need an upgrade. 87 00:07:38,200 --> 00:07:43,000 The only good news I have for you is that you can shave a few bucks off of your budget for now, 88 00:07:43,000 --> 00:07:47,400 and then with the money you saved, hopefully get something a little faster once the bubble pops. 89 00:07:51,720 --> 00:07:53,480 Any time now would be nice! 90 00:07:56,680 --> 00:08:01,000 All right, well, I guess it's not happening yet, but while we continue to wait, here's a word 91 00:08:01,000 --> 00:08:05,560 from our sponsor. If you guys enjoyed this video, make sure to check out the last time we did this, 92 00:08:05,560 --> 00:08:12,280 but with DDR4. So if you want to pick up like an older Intel 12 or 13th gen chip and pair that 93 00:08:12,280 --> 00:08:17,080 with some nice fancy DDR4, you can decide for yourself exactly how much to spend.