Glad you clarified the 2 sticks vs 4 thing and replaced it with the more accurate single rank vs dual rank. If your ram modules are single rank, use 4 sticks. If your modules are dual rank, use 2 sticks you won't get any significant benefit by going to 4 (except doubling your available Ram that is).
True. Just wish he went with a more common set up like DDR4 3600 C16 dual rank. I mean, besides super tuned DDR4 3600 C14, DDR4 3600 C16 pretty much IS the sweet spot for Ryzen. Then when you make that set up dual rank, you'd see better gains. Instead, the DDR4 3600 dual rank is conveniently C18? My G.Skill Ripjaws DDR4 3600 C16 dual rank came stock at C16. It's like every example was shown except, IMO, the most important example for Zen 3. DDR4 3600 C16 dual rank. It's why I specifically chose it. I didn't go DDR4 3800 dual rank because I wouldn't have gotten C16. I didn't care about DDR4 4000 because IF rarely is stable @ 2000mhz, again, throwing performance away. But DDR4 3600 C16 dual rank? THAT.... is the sweet spot. And if you are ever so lucky enough to get that to C14, you would have THE BEST RAM set up for Zen 3, hands down. But yeah, why is 3600 C16 dual rank missing? It's like he started with a point to prove and forged the results to suit his presupposed notion. That's why I like Ancient Gameplays. He keeps things realistic and equal. It's the whole reason I went DDR4 3600 C16 dual rank (2x16GB). I hit at LEAST 62GB/s read speeds in AIDA64 memory/cache benchmark. Everyone else hits in the 50GB/s range. Zen 3 is absolutely memory sensitive in all aspects. Timing. Frequency. Ranks. ALL of them. So when you make the best of those three, you get the best results. REAL world results. He went to the trouble of dong DDR4 3200 C14 but when DDR4 3600 came up, it was C18 and I just think that's a bit... well, it's intentionally done to screw with the results. He didn't want dual rank to make a difference so he made it so that it had little difference by choosing "several" set ups, but not one single common set up. DDR4 3600 C16. I mean, ffs lol, everyone knows that's the sweet spot for Zen 3 to begin with. When you dual rank it, it's the best. EDIT: Wow, even at 18:03, he shows the exact kit of RAM I have. There are 3x G.Skill DDR4 3600 kits. The GVKC model number is the dual rank model. That's the exact model he could have and should have used lol. It would have obviously shown why dual rank is important. Instead, it's like he grabbed that kit and upped the CAS to 18 to hinder the results.
@@kingeling I know that on something like Zen 3, 2x dual rank kits (4x sticks, all dual rank) still = dual rank. So it's coming down to MT/s and timings. And we know Ryzen LOVES low timings. Will low timings get you a faster "bandwidth" in AIDA64? No, not if it's DDR4 3200 C14 vs DDR4 3600 C18. But it works wonders in gaming. "Wonders" being 5-10%. And I get what they're saying, it's not that big of a deal. On its own, no, you can't see a 10% difference when it's 50 vs 55 fps. But when you combine the dual rank with your per curve optimization and PBO along with the higher FCLK, it all adds up. You go from an average rig to top 100 in 3DMark. This is why I'm a bit disappointed that they didn't run the exact RAM they showed at the end. G.Skill DDR4 3600 C16 dual rank. It's what I'm running. It's 3600 MT/s AND C16. They totally left that out of the test. If it's 3600 in the test, it's single rank. If it's dual rank, it's either 3200 C14 or 3600 C18 in dual rank. But they showed some DDR4 3600 C16 at the end 😞It's literally the exact kit that I'm using. 2x16GB. There's 3x models of that kind of G.Skill. You have to get the model number that ends in GVKC on the 2x16GB for dual rank. There's GVK and something else, both 2x16GB single rank.
@@kingeling And, wait, that DDR4-3200 C14 just says "dual rank". It's most likely 4x single rank sticks. 4x single rank sticks = dual rank. 2x single rank sticks = single rank.
@@Hardwareunboxed is every benchmark god called Steve?? lol Well this makes me fell bit better of my Newegg purchase, G Skill Trident Z Neo 2x16GB 3600mhz cl16 :P 2x16GB sticks were way cheaper than 4x8GB i didnt wanna wait and have them sell out like everything is going recently..5900X should be here soon! www.newegg.com/global/uk-en/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232861 RIP Intel its been 15 years since I been on AMD :o Not since single core Athlon K7 days lol showing my age
@@Hardwareunboxed but seriously, thanks for this, you saved me $200. I'm going to stick with my 8gb*2 ddr4 3000mhz cl16 if the gains are only 10-15fps with 3600mhz cl16. Especially in my country 3000mhz cl16 32gb ($180) costs less than 3600mhz cl16 16gb ($200)
@@akshathk. And the gains are only 10-15fps IF you are entirely CPU limited. AKA if you have the wildest graphics card but play at 1080p with potato quality settings.
Honestly there is no tech channel that is as informative as HWU. I've learned so much from you guys over the past few years. Thank you genuinely for all your hard work. Love from Ireland.
So, in other words: (if you only need 16GB of RAM), on both AMD and Intel, 4x4GB (single-rank) sticks compared to 2x8GB (single-rank) sticks may give you up to around +15% performance (in specific CPU-limited scenarios), (however it removes the upgrade path to 32GB later on); (if you need 32GB of RAM), 2x16GB (dual-rank, which most are) sticks are fine and there's no need for 4x8GB (single-rank) sticks, (because 2 sticks tend to be more stable and you also retain the upgrade path to 64GB later on); (if you need 64GB of RAM), more research is needed because there is conflicting information on whether quad-rank improves performance further or overloads the controller instead (in which case you'll have to search for rarer single-rank 16GBs for maximum theoretical performance).
Yes, if you are only comparing single rank CL16 to dual rank CL16. There are CL14 kits that will get performance similar to steve's tuned timings, which would be the best in a majority of titles. Of course, it's also a lot more expensive.
@@KsaCommentator Unfortunately many memory manufacturers don't specify so you either have to hope someone else has posted about that particular kit online or buy the kit yourself and use typhoon burner.
Basically after reading about memory you cant access both ranks at once if the memory module is dual ranked. Its because both ranks share the data circuits etc and only the select circuit for each rank is not shared. With 4x8GB each rank has its own. So you can access them all at once. The first two sticks are one rank and the second two sticks are another. The two ranks have their own circuits for data. Also the timings for 2x16GB dual rank are normally lower than single rank memory. Thus with 4x8GB you can keep the low b-die timings and become dual ranked. Quad rank does not help in games because it increases latency. If you think there's no need for 4x8GB (single-rank) sticks, you need to look at your methods of research. I have DDR4 3600MT/s CL14 with a bandwidth of 51GB copy. Four sticks dual rank which gives me ~5% in time spy cpu. The timings are CL14-15-15-35 CR2, single or dual rank. Latency is 47ns dual rank, single rank is 42ns and 49GB copy. I have a 10900k at 15.15k time spy cpu score. About ~800 points are the dual rank increase. In firstrike I am 31.5k physics. The RAM I am using is currently selling for £105 per kit, can be as high as £130. valid.x86.fr/6pbch5 cpu-z I have yet to dial in the single core OC by using the v/f curve. I have 5.2GHz SSE and 5.1GHz AVX.
Thank you for throwing in a 1440p result, I am tech and hardware savy but I was struggling to gauge via 1080p how it might translate to 1440p even with my knowledge of Fps,frame times etc which I've learned from guys like you and tech jesus so thanks for that, gonna have to watch this one a few times.
For sure. I understand why, for CPU reviews, most people focus on 1080p (or even less) to show differences between CPUs. But for people gaming at 1440p or 4K, the reviews aren't very clear when trying to gauge the performance you'll actually get. That's why those GPU scaling videos (1 GPU, different resolutions) are very important to me and I hope HWU continues making them.
I'll tell you more at 1440P ultra settings, even DDR4 2133Mhz is not that bad -)))) I actually tested the speed myself of 5800X and the CPU-Z gives me the higher speed on AUTO settings with 2133 RAM compared to 4000 Mhz CL18. Because 2133 Mhz is CL 13 and it runs at 1.2V - so it gives some extra temperature room for the CPU to overclock even further -))))))
Yeah I was wondering what the total gains would be going from single rank XMP to dual rank tuned memory. Just to see what gains you could realistically expect if you put in both a bit more money and a fair amount of effort.
A note: Ryzen used to have more trouble with overclocking ram in dual rank, that is why it wasn't recommend for first gen. The memory controllers and infinity fabric have changed that a lot
Odd thing however, 3000 and 5000 should have the same IO chip. Are the gains due to the improved infinity fabric, or better IO chip binning/improved fab process?
@Sergio Ribeiro The gains are almost the same. If Ryzen 5000 has any benefits, it's likely due to bios optimizations, or more likely, infinity fabric optimizations and the better design of ryzen 5000. They also are always improving bins and the silicone in general but most of the memory differences between ryzen 3000 and 5000 would be within a few percent at best.
@@Hardwareunboxed not saying it had, but it seems that, allegedly, it's easier to push the Fclk to 2000mhz and achieve 1:1:1 with 4000mhz ram on the 5000 series. Again, supposedly. I've seen people commenting a Der8auer clip on that. This is why I've been wondering if the gains are really coming from improved infinity fabric alone or if the memory controller is also improved. Then again, after watching the clip, I noticed that the gains are nearly the same on 3000 and 5000 series on SR vs DR modules, soooooo, it might be me just confusing things.
@@alexgabriel5877 That's covered closer to the end of this video. Possible benefits entirely depend on your GPU - something like 2070S isn't enough to see the difference
Unfortunately this is really missing a comparison between 2x16gb (dual rank x2) and 4x8gb (single rank x4) at the same speeds, to see if there's actually a difference in the both configurations. Populating all four DIMM slots might have a negative impact on Daisy Chain topology motherboards, but I couldn't find any tests for Ryzen 5000 regarding this.
Wow IMO this was one of your most impressive videos so far. You explained things really clearly and presesnted all the numberes but did´t get carried away with it and gave us much needed context at the end.
So if I understand correctly. A 2 * (2*8) DDR4 3600 CL16 kit is better than a 2*16 DDR4 3600 CL16 kit of the same brand and stuff? Depending on the quality of the gaming settings and the GPUetc, but assuming that it does matter. Or let's put it like this. All options are DDR4 3600 CL16 from the G.Skill Trident Z Neo family. What would be better? 2*16 GB sticks, 4*8 GB sticks or 2*(2*8) sticks? Or have I completely misunderstood how single, dual and quad rank works and perhaps the Trident Z Neo family doesn't have some of those?
@@jelmerbff If you take 2x 8GB or 2x 16GB does not make a difference if it's the same kind of memory. The thing with single and dual rank is that you often do not know whether or not the sticks have single or dual rank on them. So the only way to make sure it behaves as dual rank is to use four memory sticks. I will probably to with 4x 8GB of that Crucial Ballistix 3600 CL16 memory shown near the end of the video now, meaning buying twice the 16GB Kit with 2x 8GB per kit. Before this video I was aiming at the 2x 16GB kit of G.Skill RipJaws V 3600 CL16.
@@jelmerbff most 16GB sticks are dual rank, so performance should be similar between 2x16GB and 4x8GB configurations with the same timings. Both configurations give two ranks per memory channel. Steve even shows that in this video. However, not all 16GB sticks are dual rank, so this is not always the case.
Remember finding a matched quad kit for overclocking the memory is much harder than a dual stick. You'll likely pay more for the quad-kit than two dual-kits for factory matched timings.
@@MrReese Yeah I just bought 2*16 GB 3600 CL16 Trident Z Neo's a while back but a brother of a friend of mine is about to build his pc. I recommended the same sticks I have, but it would be better to get 2 16 GB kits or 1 4*8 GB kit? Or I need to make sure that my own 32 GB Kit is certainly dual rank. Then it won't matter? All in all, dual or quad rank > single rank
Another amazingly detailed video! Thanks so much Steve for all of the testing and comparisons here, it must have taken quite some time. As always, keep up the great work!
Wait, why is there no tuned dual rank memory in the benchmark? Isn't this the whole point of the video, to see scaling from single to dual rank, yet you are only tuning single rank?
This is the first time you've ever lost me with performance results. What's the absolute fastest stick configuration, 2 dual rank sticks, 4 dual rank sticks, or 4 single rank sticks?
2 dual rank sticks would be the fastest when given the same equivalent speed and latency timings. 4 single rank sticks is the most cheap/cost-effective solution to rectifying a ram config that can penalize Ryzen 5k performance, 4 dual rank sticks is just the worst out of the three and will more likely hurt performance - it won't be as good as the two latter.
Loved the explanation of how "paging" works on dual and single rank... and it was fun to see the testing showing that good quality memory matters, but, that extreme quality is only worth it in special cases... Thank you for putting in so much work and effort into producing an excellent video for us tech nerds!
Just want to let you know that we appreciate all the efforts and hardwork you put on every review/videos you guys release. Thank you so much! We also know that this is just the start and the busy days is yet to come but.. have some sleep for now Steve! 😁
This has been my go-to channel for literally anything related to pc build advice and recommendations for years. The amount of work and effort involved in creating these videos is simply massive, and I'm amazed by how much I've learned from this channel.
Thanks for clarifying, I was so confused about it and I wasnt sure if it was a typing mistake or not, since im still quite a noob in memory configurations. Especially at 10:42 I was questioning myself but it turns out my thoughts were right and graph was just misleading.
@@fuzzjunky they have a habit of rambling. I barely watch their videos completely any more. Their content is more of an intro and conclusion type of view.
Yes, their video felt misleading and will most likely be a source of a new myth even if they do offer a follow-up later on. A bit disappointing that they're chasing baity headlines more often now and tend to confuse people with plentiful but not structured enough information.
@@youtubevanced4900 Glad to see I'm not the only one. When GN Steve starts reading the script as fast as possible it gets tiresome and I just tune out. I think they would benefit streamlining the content instead of blurting out charts and charts of data often without any conclusive outcome.
Just one additional reason to get 32GB of ram. - Avoid swapping, SSD reliability by eliminating swapping. - No need to close applications to play games. - Much snappier system overall since commonly used data is in memory. - For some tasks, it would work like CPU would happen to be 10-20% faster. When I made my purchase, the pricing made it a far easier choice. The local price difference between 32GB of DDR 3200 and 16GB of DDR 3600 was smaller than going from DDR 3200 to DDR 3600. I was a bit worried that I left some CPU performance on the table by not going for 3600, but dual rank compensated it fully.
Well that would ring true for just about any main base of the PC audiance. People looking for performance tuning would want to tune to the highest clocks possible that means tight lower groups of ram timings would be a crucial point in decision.
Thanks Steve! Great job as always. Good to know the dual rank setup improves performance but after nearly 1h between yours and the other Steve's videos i would love to know if there is a difference in performance between identically set-up 2x dual-rank and 4x single-rank configurations and how both setups affect the overclocking potential. There's only one man for this job- Buildzoid 😉
Thank you. As a datacenter sysadmin it was maddening to read the discussions spurred from that video. It generated so much confusion, and so many confidently wrong comments.
To sum it up: the best if you can get 2 sticks of dual rank modules. 4 sticks can affect overclockability, while single rank makes your 1% fps much worse. Steve, thank you for this extremely useful material.
Can someone confirm: All in dual channel 2 modules, single rank each module = single rank total memory configuration 2 modules, dual rank each module = dual rank total memory configuration 4 modules, single rank each module = dual rank total memory configuration 4 modules, dual rank each module = quad rank total memory configuration
Jack - I too am now confused. I thought I had a handle on it, but looking at the graph above, at 3:00, it just looks wrong. It's probably not, but looks very confusing to me, and I thought i knew the differences before looking at that! Doh!
@@imadecoy. Fastest 2x8GB dual rank I found were HyperX 3466MHz CL 16. Would that be better than 2x8GB single rank 3600MHz CL16 ? If I get the graphs right it means if you dont tune your memory than dual rank is faster on average. 3200MHz dual rank (with better timings) is beating the faster single rank 3600MHz CL18. Is it beating it because its dual rank or because it has better timings ? CL14 vs CL18...(I am looking at the 8 games average graph)
It was very confusing at first for me. Hearing him say dual rank is a specific spec of ram then never mention that again and then use dual rank term when using 4sticks. I kept looking for 4 sticks dual rank data 🤦🏻♂️lol
@Marcus R It's good article, but they used T-topology board so 4 sticks might work better. It would be useful if they use more popular daisy chain ones.
@Marcus R thank you for that. I've been anxiously awaiting some data on Zen 3 + quad rank configuration and worried my 4x16 setup would be problematic either in stability or performance.
@Marcus R Yeah, I have already B-Dies kits and Asus Hero VIII x570. Not afraid so much, as I should be able to make 3200 CL14 when my CPU arrives, but still it's interesting topic how well rysen 3000/5000 memory controller handles such a load.
@Marcus R OK, well I'm optimistic then. I currently am not having any issues with my 4x16 (quad rank) 3200 CL16 memory, running at stock XMP. I am using a B550 Aorus Master and 3600. My 5950x is on order so we'll see.
Thank you for this video Steve, really informative and easy for someone like me to understand. Been overwhelmed with buying decisions this generation and the whole memory situation was an extra headache I didn't need. When you are considering parting with hard earned cash you don't want to buy parts that are too overkill or are going to hamper other parts. Really good to know that in the real world it won't make a huge impact and that I don't have to rush out and buy a DDR4000 CL14 kit and compromise elsewhere
Thank you Steve! Really informative video! I personally appreciate the 2070 Super data. Now I know for sure I am not leaving any performance on the table by sticking with the XMP profile.
These kinds of videos are pure gold. It doesn't matter how many times someone says "It won't matter in X scenario" without showing it on a graph... people with short attention spans are going to miss the point and miss-quote the video by pointing to a graph of something artificially bottle necked. Absolutely can't wait for the CPU/GPU scaling videos with 10th gen/Zen 3 and Ampere/RDNA2... The pure combinatorial nightmare of that text matrix qualifies you guys as both heroes and saints of public service!
I left a similar comment on GN, but something to consider which may add a confounding variable to the tests is memory topology. Buildzoid tested this with the 3xxx series a year ago. Basically T-Topology and Daisy Chain memory layouts will change performance numbers when it comes to 2 sticks or 4 sticks. He did a video on this. Generally it means a 2 stick daisy chain will outperform a 2 stick t-topology, however a 4 stick t-topology will generally out perform a 4 stick daisy chain. This is particular to which motherboard you have and the layout it has which isn't always easy to discern. If you're interested in this you can take a look at his video on this. I don't have a 5XXX series to test, so I don't know if this is still a thing and I don't think he has made a updated video particularly on this. Video you are looking for is: Rambling about motherboard memory layouts.
Great video. Elephant in the room= at the high end, Intel can top out around 4400mhz C17 but Ryzen is stuck between 3800-4000 due to FCLK limitations (in a gaming context).. that's the comparison I want to see- both platforms pushed to the max. Intel probably still in front here (Gaming only) - which is quite impressive for such a old architecture
Steve the hardware professor! Always love these type of videos you and Tim make. You simplify the explanation, make it easy to understand without a condescending tone 👍👍
Why though? Even a cheap systems can run those titles well above monitor refreshrates, especially if you turn down the setting to "competitive mode". Even those fancy and expensive 360hz monitors are too slow to show a difference. I know it can be beneficial to run games above monitor refreshrate but not when we are talking 900 instead of 800 or 600 instead of 500 fps. The 3600 can run csgo on constant 300+fps no sweat, with much higher fps most of the time. Meaning if you are not one of those fancy folks who spend big bucks on 360hz monitors you are fine with a cheap cpu+ cheap ram. And if you spend the money on such an expensive monitor you also will have enough spare cash to get a zen3 cpu and then you are talking about 500-600fps+. There is just no point in optimizing for competetive esport games. Pick any somewhat uptodate PC and you will be fine on 144hz or 240hz monitors.
@@ralfrudi3963 You are partially right about this. The point is not so much about performance count. A 3060 can deliver the best experience coupled with a good CPU. You can reach 600 fps + and 300-500 on faceit. What you need for CSGO is to lower latency anyway you can, to lower input lag. You need to have the highest Lows 1% and 0.1 % TO AVOID ANY STUTTER. The smoother the experience, the better. You need 6-8 cores (12-16 threads) to avoid lower lows and hyperthreading if you want the absolute experience. Ofc if you're not even remotely competitive or not too serious about the game you won't need that.
For the love of god and all that is holy.. someone get this man some more sticks that can be tuned and run in duel ranked mode! You don't understand... my min max mentality needs to know!! But all joking aside, this is the video I have been looking for when it comes to memory on Ryzen 5000, thank you Steve and the crew at Hardware unboxed for an awesome video! Keep it up.
Ads are out of control, I watched like 6 in this video, so annoying! Great content as always Steve, I never fully understand the dual and single rank thing when talking about memory, now I think I get it.
I don't know if you'll see this, but I'm wondering HOW were you able to make this video? You're running the RAM at up to 4000 and FCLK at up to 1900. I have a 5900X and flat out cannot get it stable at any settings I've tried so far, even 1600 FCLK and 3200 RAM. A R5 3600 in the same system runs perfectly at 1800 FCLK and 3600 RAM. There are threads all over the internet of people having the same problem with 5xx0 CPUs, and the general advice is you have no chance of running above 3200 with current BIOSs, and even then it seems to need a lot of good fortune to end up with a stable system that isn't constantly rebooting. But you make no mention of these problems, how?!!!!!
He basically said, you don't need to know, just buy decent 3600 ram if you want to future proof your ram and forget about it. Unless you play in 1080p on on a 1500$ video card.
Yeah I just ordered them and then wrote the support and they told me they only had the single rank modules in stock. Not sure if they are ever coming back in Dual Rank. Not sure if I should go for 4x8GB which is a bit more expensive in my region at the moment and cancel my order.
Thanks for bringing this up. I was looking to buy that RAM set (2 x 16GB - 3600 CL16) but now that you mentioned it's single rank that means I need to look for other/ different RAM sets then
Thanks for the in-depth analysis. This is something I hadn't thought of in a very long time but makes total sense when I first heard from it. Numbers to back up the theory is always nice to have :)
What your description seems to fail to mention is that at least on Zen 1 and Zen 2 (not sure about Zen 2+) the MMU couldn't effectively drive dual-rank and saw performance drops when using dual-rank or quad-DIMM configurations. What we see now is the scaling we should have seen previously. I don't recall this being investigated when Zen 2 was released so not sure exactly when the improvement occurred.
@@pawerusek363 yeah it's near impossible to get any good g.skill ram in the UK atm. Been searching for the 3800 cl14 kit but that's just a myth at this point
@@arteljus983 starcraft 2 runs at 50 fps and is CPU limited. Stalker causes Zen 2 and Intel CPUs to stutter. Just because it's old doesnt mean it is not demanding. Even Hardware Unboxed tested SC2 last year.
BTW if you don't have Ryzen 5000, this is first upgrade, SC is shit game engine, is not demanding, is write as shit, but Ryzen 5000 with 32MB L3 cache giving a lot of performance, look for testing. It will be better investment than memory.
Awesome work Steve! Any chance of doing some memory testing for other workloads, namely the compile test. Pretty niche I would imagine but would be nice to know personally when upgrading next.
Thanks for a very useful video Steve 😎 I am very glad now that I got the Crucial Ballistix 16Gb x2 kit as it is dual ranked and CL16. I based my purchase on your last video from 11 months ago 😉
The whole point is that none of it matter if you have decent ram and aren't playing at unreasonably high frame rates that are only possible by playing in 1080p on top tier graphics cards...No need for more time wasted on this issue.
@@Ben-ry1py actually I am very interested, for science, it can tell a lot about how the internals work. Also ECC ram is supported in Ryzen, I would actually like to see how it behaves.
@@rudiservo "for science" is a valid reason, but it's a ton of work, and HUB is a gaming channel. I like those edge cases too, but I'd rather Steve spent his time on things relevant to gamers as he's already overworked as it is. I'm just grateful he did this video, cause it's perfect for everyone who got peaked seeing GNs video on dual vs quad sticks of ram...now we know :)
About 2000MHz FCLK, it seems people have misread or forgotten. Hallock said that the BIOS will help those that CAN achieve it. He actually says that those are golden samples and do not represent the thick of the product. 1900 is the default to be expected, 2000 is the extraordinary. Edit: Also, thanks for the video, HWUB
I'm pretty sure you're in my PC. In a recent vid you spoke about the gains of upgrading from a i5 7600k to a Ryzen 5 5600X and now you recommend exactly the RAM-kit I bought. And yes at least I just pop those crucials in, activate XMP and get to gaming. No need to tinker in the BIOS and tighten the timings, especially for only a few % improvements.
Lol I asked this question to Google on the 5th of November. So to see this video makes me smile and happy 😊 to see a lot of people are asking the same question. I'm glad you guys get the time to test everything so the rest of us can make better and more educated spending/buying decisions!
Wow, just WOW! This came out so quick with much more information than I ever wanted to know. I didn't even expected you to touch this topic before the launch of the radeon 6000 series
Im no tech guy and I am slightly confused about which ram should I go for. I got a rough idea in the video but still cant decide whether www.newegg.com/global/ae-en/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232860 provide the best performance or not. (I will post it up again and look for other peoples opinions)
@@Commentrix I using this one and they are dual rank per module, so best performance if you asking about ranks www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232861
I built five Ryzen computers. Two were identical: Corsair Spec Alpha case, EVGA 550 PSU, ASUS PRIME-350 motherboard, Ryzen 3 2200G APU. The red one was noticeably fast, like an i7, flowed like water, pleasingly fast. The yellow one not noticeably fast. In fact, if you opened a lot of tabs, 20-30-40, in two browsers, they would totally stop opening. The only difference was this: the red one had a Samsung 500GB SSD, the yellow one had a Samsung 250GB nvme. The red one had 1 x 8GB Corsair Vengeance LPX ram; the yellow one had 2 x 4GB Corsair Vengeance LPX ram (both 3000Mhz). I know they say 2 sticks of RAM is better, then they went back to 1 is better. All I could figure is that 1 stick of RAM is better. Otherwise it's the SSD that's faster, or a combination. What do you think?
Steve's ability to say really long model numbers is seriously impressive!
I've been learning from Tim.
@@Hardwareunboxed he has taught you well
@@Hardwareunboxed I think lack of sleep helps 😅
His ability sounds even more impressive to me, a not English mother tongue
Got the polish, you mean blooper reel or is Steve a 1 and down kinda guy?
Dude, I love how you explain things. Makes it easier on my undervolted brain.
XD
celeron brain
@@TanvirRaiyan lmao!!!
better explained than gamer nexus
@@heldermachado8739 Nah, GNsteve is good, you probably just don't fit his target audience well. No offense meant.
This is the first explanation of memory ranks I've seen that actually made sense. Thank you for the clear breakdown and visual aids.
Glad you clarified the 2 sticks vs 4 thing and replaced it with the more accurate single rank vs dual rank. If your ram modules are single rank, use 4 sticks. If your modules are dual rank, use 2 sticks you won't get any significant benefit by going to 4 (except doubling your available Ram that is).
True. Just wish he went with a more common set up like DDR4 3600 C16 dual rank. I mean, besides super tuned DDR4 3600 C14, DDR4 3600 C16 pretty much IS the sweet spot for Ryzen.
Then when you make that set up dual rank, you'd see better gains. Instead, the DDR4 3600 dual rank is conveniently C18? My G.Skill Ripjaws DDR4 3600 C16 dual rank came stock at C16. It's like every example was shown except, IMO, the most important example for Zen 3. DDR4 3600 C16 dual rank. It's why I specifically chose it. I didn't go DDR4 3800 dual rank because I wouldn't have gotten C16. I didn't care about DDR4 4000 because IF rarely is stable @ 2000mhz, again, throwing performance away.
But DDR4 3600 C16 dual rank? THAT.... is the sweet spot. And if you are ever so lucky enough to get that to C14, you would have THE BEST RAM set up for Zen 3, hands down.
But yeah, why is 3600 C16 dual rank missing? It's like he started with a point to prove and forged the results to suit his presupposed notion.
That's why I like Ancient Gameplays. He keeps things realistic and equal. It's the whole reason I went DDR4 3600 C16 dual rank (2x16GB). I hit at LEAST 62GB/s read speeds in AIDA64 memory/cache benchmark. Everyone else hits in the 50GB/s range.
Zen 3 is absolutely memory sensitive in all aspects. Timing. Frequency. Ranks. ALL of them. So when you make the best of those three, you get the best results. REAL world results.
He went to the trouble of dong DDR4 3200 C14 but when DDR4 3600 came up, it was C18 and I just think that's a bit... well, it's intentionally done to screw with the results.
He didn't want dual rank to make a difference so he made it so that it had little difference by choosing "several" set ups, but not one single common set up. DDR4 3600 C16. I mean, ffs lol, everyone knows that's the sweet spot for Zen 3 to begin with. When you dual rank it, it's the best.
EDIT: Wow, even at 18:03, he shows the exact kit of RAM I have. There are 3x G.Skill DDR4 3600 kits. The GVKC model number is the dual rank model. That's the exact model he could have and should have used lol. It would have obviously shown why dual rank is important. Instead, it's like he grabbed that kit and upped the CAS to 18 to hinder the results.
So how come 4 dual rank DIMMs perform better than 4 single ranks in the tests? I'm so lost
@@kingeling I know that on something like Zen 3, 2x dual rank kits (4x sticks, all dual rank) still = dual rank.
So it's coming down to MT/s and timings. And we know Ryzen LOVES low timings. Will low timings get you a faster "bandwidth" in AIDA64? No, not if it's DDR4 3200 C14 vs DDR4 3600 C18.
But it works wonders in gaming.
"Wonders" being 5-10%. And I get what they're saying, it's not that big of a deal.
On its own, no, you can't see a 10% difference when it's 50 vs 55 fps. But when you combine the dual rank with your per curve optimization and PBO along with the higher FCLK, it all adds up. You go from an average rig to top 100 in 3DMark.
This is why I'm a bit disappointed that they didn't run the exact RAM they showed at the end. G.Skill DDR4 3600 C16 dual rank. It's what I'm running. It's 3600 MT/s AND C16. They totally left that out of the test. If it's 3600 in the test, it's single rank. If it's dual rank, it's either 3200 C14 or 3600 C18 in dual rank.
But they showed some DDR4 3600 C16 at the end 😞It's literally the exact kit that I'm using. 2x16GB. There's 3x models of that kind of G.Skill. You have to get the model number that ends in GVKC on the 2x16GB for dual rank. There's GVK and something else, both 2x16GB single rank.
@@kingeling And, wait, that DDR4-3200 C14 just says "dual rank". It's most likely 4x single rank sticks. 4x single rank sticks = dual rank. 2x single rank sticks = single rank.
@@kingeling My first reply isn't showing up for me, so, I dunno... I think I'm blacklisted for criticism on this site.
Let's be honest steve... You can't sleep.
You guys don't let me :P :)
@@Hardwareunboxed nyahahahha seriously tho get some sleep...
@@Hardwareunboxed is every benchmark god called Steve?? lol
Well this makes me fell bit better of my Newegg purchase, G Skill Trident Z Neo 2x16GB 3600mhz cl16 :P
2x16GB sticks were way cheaper than 4x8GB
i didnt wanna wait and have them sell out like everything is going recently..5900X should be here soon!
www.newegg.com/global/uk-en/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232861
RIP Intel its been 15 years since I been on AMD :o
Not since single core Athlon K7 days lol showing my age
@@Hardwareunboxed but seriously, thanks for this, you saved me $200. I'm going to stick with my 8gb*2 ddr4 3000mhz cl16 if the gains are only 10-15fps with 3600mhz cl16. Especially in my country 3000mhz cl16 32gb ($180) costs less than 3600mhz cl16 16gb ($200)
@@akshathk. And the gains are only 10-15fps IF you are entirely CPU limited. AKA if you have the wildest graphics card but play at 1080p with potato quality settings.
Your conclusion really sums up why I'm subscribed to this channel.
Half way through i was like iam not getting shit. Saw this jumped to summary that's all I needed 👍
@@Ralf986ms based on ?
Honestly there is no tech channel that is as informative as HWU. I've learned so much from you guys over the past few years. Thank you genuinely for all your hard work. Love from Ireland.
They definitely influence my hardware purchases! Best performance/$ ftw!
Hellooooo....Gamers nexus??
@@amashaziz2212 GN is educational too but to me not even nearly as fun or interesting to watch.
Its because the other channels are geared to kids...
@@amashaziz2212 gamers nexus and jay2cents are aimed for kids... this guy doesnt add all the immature competitive complaining angles.
So, in other words:
(if you only need 16GB of RAM), on both AMD and Intel, 4x4GB (single-rank) sticks compared to 2x8GB (single-rank) sticks may give you up to around +15% performance (in specific CPU-limited scenarios), (however it removes the upgrade path to 32GB later on);
(if you need 32GB of RAM), 2x16GB (dual-rank, which most are) sticks are fine and there's no need for 4x8GB (single-rank) sticks, (because 2 sticks tend to be more stable and you also retain the upgrade path to 64GB later on);
(if you need 64GB of RAM), more research is needed because there is conflicting information on whether quad-rank improves performance further or overloads the controller instead (in which case you'll have to search for rarer single-rank 16GBs for maximum theoretical performance).
Yes, if you are only comparing single rank CL16 to dual rank CL16. There are CL14 kits that will get performance similar to steve's tuned timings, which would be the best in a majority of titles. Of course, it's also a lot more expensive.
@@giglioflex but, how do i know that a memory is single/dual rank?
@@KsaCommentator Unfortunately many memory manufacturers don't specify so you either have to hope someone else has posted about that particular kit online or buy the kit yourself and use typhoon burner.
Excellent summary 👍🏼 thnx
Basically after reading about memory you cant access both ranks at once if the memory module is dual ranked. Its because both ranks share the data circuits etc and only the select circuit for each rank is not shared. With 4x8GB each rank has its own. So you can access them all at once. The first two sticks are one rank and the second two sticks are another. The two ranks have their own circuits for data.
Also the timings for 2x16GB dual rank are normally lower than single rank memory. Thus with 4x8GB you can keep the low b-die timings and become dual ranked. Quad rank does not help in games because it increases latency.
If you think there's no need for 4x8GB (single-rank) sticks, you need to look at your methods of research. I have DDR4 3600MT/s CL14 with a bandwidth of 51GB copy. Four sticks dual rank which gives me ~5% in time spy cpu. The timings are CL14-15-15-35 CR2, single or dual rank. Latency is 47ns dual rank, single rank is 42ns and 49GB copy. I have a 10900k at 15.15k time spy cpu score. About ~800 points are the dual rank increase. In firstrike I am 31.5k physics.
The RAM I am using is currently selling for £105 per kit, can be as high as £130. valid.x86.fr/6pbch5 cpu-z I have yet to dial in the single core OC by using the v/f curve. I have 5.2GHz SSE and 5.1GHz AVX.
I thought Steve promised that he is going to sleep, what are those eyes!
Get used to it mate.
@@Hardwareunboxed g'night mate
@@Hardwareunboxed Please live long enough to review AMD Ryzen 99 999000X vs Intel i99- 999000K
@@Hardwareunboxed Time to start wearing sunglasses each video so we can't see how bloodshot your eyes are
@@3Dant lol his eyes aren't bloodshot fool, they just have bags and dark circles under them
Thank you for throwing in a 1440p result, I am tech and hardware savy but I was struggling to gauge via 1080p how it might translate to 1440p even with my knowledge of Fps,frame times etc which I've learned from guys like you and tech jesus so thanks for that, gonna have to watch this one a few times.
For sure. I understand why, for CPU reviews, most people focus on 1080p (or even less) to show differences between CPUs. But for people gaming at 1440p or 4K, the reviews aren't very clear when trying to gauge the performance you'll actually get.
That's why those GPU scaling videos (1 GPU, different resolutions) are very important to me and I hope HWU continues making them.
@@Cinetyk I think Steve said he's going to make a new one to compare Ampere to Navi 2 after Navi 2 officially launches.
@@johnbuscher that one's gonna be interesting for sure 😄
The higher the resolution the less the CPU matters as everything becomes GPU constrained
I'll tell you more at 1440P ultra settings, even DDR4 2133Mhz is not that bad -)))) I actually tested the speed myself of 5800X and the CPU-Z gives me the higher speed on AUTO settings with 2133 RAM compared to 4000 Mhz CL18. Because 2133 Mhz is CL 13 and it runs at 1.2V - so it gives some extra temperature room for the CPU to overclock even further -))))))
I would love to see in the future updated results with dual rank & tuned timings, and quad rank even just to go overkill ^^
Thanks Steve. Unfortunately this test really lacked a tuned double rank config
Yeah I was wondering what the total gains would be going from single rank XMP to dual rank tuned memory.
Just to see what gains you could realistically expect if you put in both a bit more money and a fair amount of effort.
It's irrelevant unless you play in 1080p on an rtx 3080 or 3090...
No need for him to do any more work to make that clear.
@@Ben-ry1py Actually for different applications like music production this is relevant ;)
A note: Ryzen used to have more trouble with overclocking ram in dual rank, that is why it wasn't recommend for first gen. The memory controllers and infinity fabric have changed that a lot
Odd thing however, 3000 and 5000 should have the same IO chip. Are the gains due to the improved infinity fabric, or better IO chip binning/improved fab process?
@@GrimpakTheMook Zen 2 didn't have memory issues.
@Sergio Ribeiro The gains are almost the same. If Ryzen 5000 has any benefits, it's likely due to bios optimizations, or more likely, infinity fabric optimizations and the better design of ryzen 5000. They also are always improving bins and the silicone in general but most of the memory differences between ryzen 3000 and 5000 would be within a few percent at best.
@@Hardwareunboxed not saying it had, but it seems that, allegedly, it's easier to push the Fclk to 2000mhz and achieve 1:1:1 with 4000mhz ram on the 5000 series. Again, supposedly. I've seen people commenting a Der8auer clip on that.
This is why I've been wondering if the gains are really coming from improved infinity fabric alone or if the memory controller is also improved.
Then again, after watching the clip, I noticed that the gains are nearly the same on 3000 and 5000 series on SR vs DR modules, soooooo, it might be me just confusing things.
@@a.f.r.m...d.e.d.d.2898 Yeah, after rewatching the vid, I noticed that. As for the rest, do check the above answer.
Steve... you said you'd get some sleep!
I know :( I lied!
Both Steves really should get some sleep at this point :D
@@Hardwareunboxed hey man nice video! but what about say 1440p and high settings? will there be no difference in RAM performance ?
He just installed more RAM to keep him going instead
@@alexgabriel5877 That's covered closer to the end of this video. Possible benefits entirely depend on your GPU - something like 2070S isn't enough to see the difference
Unfortunately this is really missing a comparison between 2x16gb (dual rank x2) and 4x8gb (single rank x4) at the same speeds, to see if there's actually a difference in the both configurations. Populating all four DIMM slots might have a negative impact on Daisy Chain topology motherboards, but I couldn't find any tests for Ryzen 5000 regarding this.
You won’t find it because that means companies lose out on money…..
Wow IMO this was one of your most impressive videos so far. You explained things really clearly and presesnted all the numberes but did´t get carried away with it and gave us much needed context at the end.
Thanks for the hard work Steve!
Holy crap, this was super interesting and informative. I need to rethink my entire RAM strategy now, lol.
So if I understand correctly. A 2 * (2*8) DDR4 3600 CL16 kit is better than a 2*16 DDR4 3600 CL16 kit of the same brand and stuff? Depending on the quality of the gaming settings and the GPUetc, but assuming that it does matter.
Or let's put it like this. All options are DDR4 3600 CL16 from the G.Skill Trident Z Neo family. What would be better? 2*16 GB sticks, 4*8 GB sticks or 2*(2*8) sticks? Or have I completely misunderstood how single, dual and quad rank works and perhaps the Trident Z Neo family doesn't have some of those?
@@jelmerbff If you take 2x 8GB or 2x 16GB does not make a difference if it's the same kind of memory.
The thing with single and dual rank is that you often do not know whether or not the sticks have single or dual rank on them. So the only way to make sure it behaves as dual rank is to use four memory sticks.
I will probably to with 4x 8GB of that Crucial Ballistix 3600 CL16 memory shown near the end of the video now, meaning buying twice the 16GB Kit with 2x 8GB per kit. Before this video I was aiming at the 2x 16GB kit of G.Skill RipJaws V 3600 CL16.
@@jelmerbff most 16GB sticks are dual rank, so performance should be similar between 2x16GB and 4x8GB configurations with the same timings. Both configurations give two ranks per memory channel. Steve even shows that in this video.
However, not all 16GB sticks are dual rank, so this is not always the case.
Remember finding a matched quad kit for overclocking the memory is much harder than a dual stick. You'll likely pay more for the quad-kit than two dual-kits for factory matched timings.
@@MrReese Yeah I just bought 2*16 GB 3600 CL16 Trident Z Neo's a while back but a brother of a friend of mine is about to build his pc. I recommended the same sticks I have, but it would be better to get 2 16 GB kits or 1 4*8 GB kit? Or I need to make sure that my own 32 GB Kit is certainly dual rank. Then it won't matter?
All in all, dual or quad rank > single rank
Another amazingly detailed video! Thanks so much Steve for all of the testing and comparisons here, it must have taken quite some time. As always, keep up the great work!
Wait, why is there no tuned dual rank memory in the benchmark? Isn't this the whole point of the video, to see scaling from single to dual rank, yet you are only tuning single rank?
Thank you, I had the same question!
This is the first time you've ever lost me with performance results. What's the absolute fastest stick configuration, 2 dual rank sticks, 4 dual rank sticks, or 4 single rank sticks?
2 dual rank sticks would be the fastest when given the same equivalent speed and latency timings. 4 single rank sticks is the most cheap/cost-effective solution to rectifying a ram config that can penalize Ryzen 5k performance, 4 dual rank sticks is just the worst out of the three and will more likely hurt performance - it won't be as good as the two latter.
4xSR (better overclock) > 2xDR (balanced) > 4xDR (better capacity)
Hardest working tech RUclipsrs out there along with the GN team !
You lied to us about sleeping !!!
They probably have clone.
Loved the explanation of how "paging" works on dual and single rank... and it was fun to see the testing showing that good quality memory matters, but, that extreme quality is only worth it in special cases... Thank you for putting in so much work and effort into producing an excellent video for us tech nerds!
Hello, I am one of those people that spends hours of time tuning my memory for next to no benefit. It is fun.
@@Ralf986ms I tune memory for benchmarking on hwbot, for fun. I dont even remember what was said in this video lol. My comment is 2 months old.
@@Ralf986ms Okay dude. Have a good time with that.
Just want to let you know that we appreciate all the efforts and hardwork you put on every review/videos you guys release. Thank you so much! We also know that this is just the start and the busy days is yet to come but.. have some sleep for now Steve! 😁
This has been my go-to channel for literally anything related to pc build advice and recommendations for years. The amount of work and effort involved in creating these videos is simply massive, and I'm amazed by how much I've learned from this channel.
This is really a mind-boggling amount of benchmarking, plus some wise words at the end. Much appreciated
Just a quick correction, the 2x8 single rank ram in the benchmarks (second from bottom) says 4x8 not 2x8.
You are right! In the bar it is incorrect but in the graph key it is correct
Thanks for clarifying, I was so confused about it and I wasnt sure if it was a typing mistake or not, since im still quite a noob in memory configurations. Especially at 10:42 I was questioning myself but it turns out my thoughts were right and graph was just misleading.
THIS COMMENT NEEDS TO BE PINNED FOR ANY NEW PC PEOPLE!!
@@rabidzojo1073 exactly !! I thought I lost the plot for a second😅
what a great and informative video, good job mr. steve
Dang, this video makes the GN video look about as in-depth and knowledgeable as a Jayz2cents video.
Gn also rambled and gave too much info without any real clear conclusion.
I'm glad they brought it up though because others have joined in to clarify.
@@fuzzjunky they have a habit of rambling. I barely watch their videos completely any more. Their content is more of an intro and conclusion type of view.
Lol
Yes, their video felt misleading and will most likely be a source of a new myth even if they do offer a follow-up later on. A bit disappointing that they're chasing baity headlines more often now and tend to confuse people with plentiful but not structured enough information.
@@youtubevanced4900 Glad to see I'm not the only one. When GN Steve starts reading the script as fast as possible it gets tiresome and I just tune out. I think they would benefit streamlining the content instead of blurting out charts and charts of data often without any conclusive outcome.
Just one additional reason to get 32GB of ram.
- Avoid swapping, SSD reliability by eliminating swapping.
- No need to close applications to play games.
- Much snappier system overall since commonly used data is in memory.
- For some tasks, it would work like CPU would happen to be 10-20% faster.
When I made my purchase, the pricing made it a far easier choice. The local price difference between 32GB of DDR 3200 and 16GB of DDR 3600 was smaller than going from DDR 3200 to DDR 3600. I was a bit worried that I left some CPU performance on the table by not going for 3600, but dual rank compensated it fully.
So pretty much as I understood it from GN before, if Im looking at getting 32gb then Im better getting 4x8 than 2x16, all other things being equal.
Well that would ring true for just about any main base of the PC audiance. People looking for performance tuning would want to tune to the highest clocks possible that means tight lower groups of ram timings would be a crucial point in decision.
Thanks Steve! Great job as always. Good to know the dual rank setup improves performance but after nearly 1h between yours and the other Steve's videos i would love to know if there is a difference in performance between identically set-up 2x dual-rank and 4x single-rank configurations and how both setups affect the overclocking potential. There's only one man for this job- Buildzoid 😉
It only matters if you are getting stupidly high frame rates, as in, playing in 1080p on an rtx 3080 or 3090...which is ridiculous.
Steve, thank you for "wasting your time" to save everyone else's! Much appreciated
Steve, THANK you for the effort. It would be so interesting to see a part 2 of this with tuned dual rank / 4x DIMM's. The performance must be epic.
Thank you. As a datacenter sysadmin it was maddening to read the discussions spurred from that video. It generated so much confusion, and so many confidently wrong comments.
7:42 Rap god 🔥
Gawddamn in first run it was pretty normal to me
Awesome job. That cleared all questions. Thanks!
To sum it up: the best if you can get 2 sticks of dual rank modules. 4 sticks can affect overclockability, while single rank makes your 1% fps much worse.
Steve, thank you for this extremely useful material.
What is missing: comparison between T-Topology and Daisy Chain motherboards, as these work differently with 2-4 sticks.
Need an anwser to this asap!
Can someone confirm:
All in dual channel
2 modules, single rank each module = single rank total memory configuration
2 modules, dual rank each module = dual rank total memory configuration
4 modules, single rank each module = dual rank total memory configuration
4 modules, dual rank each module = quad rank total memory configuration
Jack - I too am now confused. I thought I had a handle on it, but looking at the graph above, at 3:00, it just looks wrong. It's probably not, but looks very confusing to me, and I thought i knew the differences before looking at that! Doh!
Thanks for your getting this up so quickly, much appreciated!
I had to watch this video two times to understand. Well made!
Thank you Steve.
Thank you for this, Steve! I just saved this video in preparation for my Ryzen 5000 build starting in a month or so!
Buildzoid would approve of your memory recommendations.
No, he would say get dual rank.
@@imadecoy. Fastest 2x8GB dual rank I found were HyperX 3466MHz CL 16. Would that be better than 2x8GB single rank 3600MHz CL16 ? If I get the graphs right it means if you dont tune your memory than dual rank is faster on average. 3200MHz dual rank (with better timings) is beating the faster single rank 3600MHz CL18. Is it beating it because its dual rank or because it has better timings ? CL14 vs CL18...(I am looking at the 8 games average graph)
Wished Steve included a Quad rank configuration, 4x16GB (dual rank) modules. It would probably be slower than dual rank.
It was very confusing at first for me. Hearing him say dual rank is a specific spec of ram then never mention that again and then use dual rank term when using 4sticks. I kept looking for 4 sticks dual rank data 🤦🏻♂️lol
@Marcus R It's good article, but they used T-topology board so 4 sticks might work better. It would be useful if they use more popular daisy chain ones.
@Marcus R thank you for that. I've been anxiously awaiting some data on Zen 3 + quad rank configuration and worried my 4x16 setup would be problematic either in stability or performance.
@Marcus R Yeah, I have already B-Dies kits and Asus Hero VIII x570. Not afraid so much, as I should be able to make 3200 CL14 when my CPU arrives, but still it's interesting topic how well rysen 3000/5000 memory controller handles such a load.
@Marcus R OK, well I'm optimistic then. I currently am not having any issues with my 4x16 (quad rank) 3200 CL16 memory, running at stock XMP. I am using a B550 Aorus Master and 3600. My 5950x is on order so we'll see.
Thank you for this video Steve, really informative and easy for someone like me to understand. Been overwhelmed with buying decisions this generation and the whole memory situation was an extra headache I didn't need. When you are considering parting with hard earned cash you don't want to buy parts that are too overkill or are going to hamper other parts. Really good to know that in the real world it won't make a huge impact and that I don't have to rush out and buy a DDR4000 CL14 kit and compromise elsewhere
The video I've been waiting for. Thanks, this is will help to soften my OCD in search of the optimal ram for my new Zen 3 build
Thank you Steve! Really informative video! I personally appreciate the 2070 Super data. Now I know for sure I am not leaving any performance on the table by sticking with the XMP profile.
Thanks Steve appreciate your hard work.
These kinds of videos are pure gold. It doesn't matter how many times someone says "It won't matter in X scenario" without showing it on a graph... people with short attention spans are going to miss the point and miss-quote the video by pointing to a graph of something artificially bottle necked. Absolutely can't wait for the CPU/GPU scaling videos with 10th gen/Zen 3 and Ampere/RDNA2... The pure combinatorial nightmare of that text matrix qualifies you guys as both heroes and saints of public service!
I left a similar comment on GN, but something to consider which may add a confounding variable to the tests is memory topology. Buildzoid tested this with the 3xxx series a year ago. Basically T-Topology and Daisy Chain memory layouts will change performance numbers when it comes to 2 sticks or 4 sticks. He did a video on this. Generally it means a 2 stick daisy chain will outperform a 2 stick t-topology, however a 4 stick t-topology will generally out perform a 4 stick daisy chain. This is particular to which motherboard you have and the layout it has which isn't always easy to discern. If you're interested in this you can take a look at his video on this. I don't have a 5XXX series to test, so I don't know if this is still a thing and I don't think he has made a updated video particularly on this.
Video you are looking for is: Rambling about motherboard memory layouts.
'whatever leaderboard that gets them excited these days'
ooooo shots fired at buildzoid :)))
Amazing level of dedication! High quality graphs! Love this channel!
Jesus Christ 12 commercial brakes for a 20min video.. RUclips has gone to shit... Grateful for the content tho Hardware Unboxed!
Great video. Elephant in the room= at the high end, Intel can top out around 4400mhz C17 but Ryzen is stuck between 3800-4000 due to FCLK limitations (in a gaming context).. that's the comparison I want to see- both platforms pushed to the max. Intel probably still in front here (Gaming only) - which is quite impressive for such a old architecture
But they work slightly differently, and memory speed is all but irrelevant to games and most laymen's workloads
Watching this 3 months later...
19:45 "There's some new GPUs, just around the corner."
... What GPU?? ...
Well done! Very well done! And I appreciate your "truth bomb" conclusions at the end about the reality of all this. Excellent work!
Amazing video. Thank you for taking the time to test this in real world scenarios.
Steve the hardware professor! Always love these type of videos you and Tim make. You simplify the explanation, make it easy to understand without a condescending tone 👍👍
How would 2 dual rank 16GB sticks work? How about 4 dual rank 16 GB sticks?
Buildzoid came to the conclusion that quad rank has no benefit. It doesn't hurt either.
Would like to see this in games with competitive graphic settings like CS GO.
Why though? Even a cheap systems can run those titles well above monitor refreshrates, especially if you turn down the setting to "competitive mode". Even those fancy and expensive 360hz monitors are too slow to show a difference. I know it can be beneficial to run games above monitor refreshrate but not when we are talking 900 instead of 800 or 600 instead of 500 fps.
The 3600 can run csgo on constant 300+fps no sweat, with much higher fps most of the time. Meaning if you are not one of those fancy folks who spend big bucks on 360hz monitors you are fine with a cheap cpu+ cheap ram. And if you spend the money on such an expensive monitor you also will have enough spare cash to get a zen3 cpu and then you are talking about 500-600fps+.
There is just no point in optimizing for competetive esport games. Pick any somewhat uptodate PC and you will be fine on 144hz or 240hz monitors.
there's no point in that unless you're playing on 800mhz monitor
@@ralfrudi3963 You are partially right about this.
The point is not so much about performance count. A 3060 can deliver the best experience coupled with a good CPU. You can reach 600 fps + and 300-500 on faceit.
What you need for CSGO is to lower latency anyway you can, to lower input lag.
You need to have the highest Lows 1% and 0.1 % TO AVOID ANY STUTTER.
The smoother the experience, the better.
You need 6-8 cores (12-16 threads) to avoid lower lows and hyperthreading if you want the absolute experience.
Ofc if you're not even remotely competitive or not too serious about the game you won't need that.
For the love of god and all that is holy.. someone get this man some more sticks that can be tuned and run in duel ranked mode! You don't understand... my min max mentality needs to know!!
But all joking aside, this is the video I have been looking for when it comes to memory on Ryzen 5000, thank you Steve and the crew at Hardware unboxed for an awesome video!
Keep it up.
Ads are out of control, I watched like 6 in this video, so annoying! Great content as always Steve, I never fully understand the dual and single rank thing when talking about memory, now I think I get it.
I don't know if you'll see this, but I'm wondering HOW were you able to make this video? You're running the RAM at up to 4000 and FCLK at up to 1900. I have a 5900X and flat out cannot get it stable at any settings I've tried so far, even 1600 FCLK and 3200 RAM. A R5 3600 in the same system runs perfectly at 1800 FCLK and 3600 RAM. There are threads all over the internet of people having the same problem with 5xx0 CPUs, and the general advice is you have no chance of running above 3200 with current BIOSs, and even then it seems to need a lot of good fortune to end up with a stable system that isn't constantly rebooting. But you make no mention of these problems, how?!!!!!
The moment he called 2070 Super a "Mid-Ranged GPU" I died
Unless we counting 70 class cards as high-end now, the RTX 2080Ti is the upper end of midrange this generation with the 3070
1650 Super is not even a GPU then 😔
@@imo098765 60 is always mid-range.
I guess my 1060 6GB is now a low end card. F.
They are not in touch with reality. 1660 Super is a mid range GPU. RX 570 and below are lower end ones. 2070 Super is easily high end GPU.
hardware unboxed is nearing i mil subs!
still feels like 200k was just yesterday
Very thorough and well laid out. This will be the go to reference video for this topic. Nice job.
I need to overclock my brain to understand ram ranking.
Don't forget to tune and tighten those timings of your brain 😊
I use coffee for that lol
Two double sided sticks or four single sided sticks will be dual rank memory.
Basically:
RAM Stick have funny square blocks on it that makes it FAST
He basically said, you don't need to know, just buy decent 3600 ram if you want to future proof your ram and forget about it. Unless you play in 1080p on on a 1500$ video card.
Ohh... shots fired!
Right? ;)
pew pew
MAN DOWN MAN DOWN
For anyone reading this information, the Crucial Ballistix 32gb (16x2) set can come in single rank.
Yeah I just ordered them and then wrote the support and they told me they only had the single rank modules in stock. Not sure if they are ever coming back in Dual Rank. Not sure if I should go for 4x8GB which is a bit more expensive in my region at the moment and cancel my order.
Thanks for bringing this up. I was looking to buy that RAM set (2 x 16GB - 3600 CL16) but now that you mentioned it's single rank that means I need to look for other/ different RAM sets then
Love the "...it's just a waste of time." at 19:10 lol
*Uh Oh.... This is what EVERYBODY has been waiting for.* 🔥
Wow, another variable to be considered. I wonder how my 4-stick quad rank config would perform (and OC) on Zen 3
Thanks for the in-depth analysis. This is something I hadn't thought of in a very long time but makes total sense when I first heard from it. Numbers to back up the theory is always nice to have :)
What your description seems to fail to mention is that at least on Zen 1 and Zen 2 (not sure about Zen 2+) the MMU couldn't effectively drive dual-rank and saw performance drops when using dual-rank or quad-DIMM configurations. What we see now is the scaling we should have seen previously. I don't recall this being investigated when Zen 2 was released so not sure exactly when the improvement occurred.
Please test 2x16GB dual sided.
Send me some please and I will :D
2x16gb dual rank seems to be the sweet spot from various posts I've seen
@@darthsidious1959 New G.Skill 2x16GB 4000+ rocks, but they are out of stock right now.
@@pawerusek363 yeah it's near impossible to get any good g.skill ram in the UK atm. Been searching for the 3800 cl14 kit but that's just a myth at this point
But isn't that exactly Steves ADATA XPG 2x16 on the chart?
I am eyeing a 32GB 2x16 dual-rank 3200 mhz CL14 kit now. I wonder if it would help Stalker and Starcraft.
Starcraft? You can probably run that on one 8GB stick and you will still have plenty of performance. Stalker is also old so no.
@@arteljus983 starcraft 2 runs at 50 fps and is CPU limited.
Stalker causes Zen 2 and Intel CPUs to stutter. Just because it's old doesnt mean it is not demanding.
Even Hardware Unboxed tested SC2 last year.
There are 2x16GB 3600 CL14 available on market, I would look at them. (most 16GB modules are dual rank)
BTW if you don't have Ryzen 5000, this is first upgrade, SC is shit game engine, is not demanding, is write as shit, but Ryzen 5000 with 32MB L3 cache giving a lot of performance, look for testing. It will be better investment than memory.
@@m_sedziwoj Ryzen 5000 is supposedly very fast in STALKER and Starcraft 2. Will see.
hands down the best channel puts stats into context of real world applicability
Awesome work Steve! Any chance of doing some memory testing for other workloads, namely the compile test. Pretty niche I would imagine but would be nice to know personally when upgrading next.
GN had code compilation in their test suite. Don't remember the results though...
Now show the Australian prices Steve so our friends in the US see our dilemma
It is two different currencies,
was about to go to sleep....
Same!
Thanks for a very useful video Steve 😎 I am very glad now that I got the Crucial Ballistix 16Gb x2 kit as it is dual ranked and CL16. I based my purchase on your last video from 11 months ago 😉
Thank you for alll your work, i was really curious about this!
Watched this whole thing. Still don't understand if I should buy 4x8 or 2x16... lol
Can you do a test with 3600 and 3800 dual rank tuned ram?
And for science make a test with quad rank and ECC.
The whole point is that none of it matter if you have decent ram and aren't playing at unreasonably high frame rates that are only possible by playing in 1080p on top tier graphics cards...No need for more time wasted on this issue.
@@Ben-ry1py actually I am very interested, for science, it can tell a lot about how the internals work.
Also ECC ram is supported in Ryzen, I would actually like to see how it behaves.
@@rudiservo "for science" is a valid reason, but it's a ton of work, and HUB is a gaming channel. I like those edge cases too, but I'd rather Steve spent his time on things relevant to gamers as he's already overworked as it is. I'm just grateful he did this video, cause it's perfect for everyone who got peaked seeing GNs video on dual vs quad sticks of ram...now we know :)
About 2000MHz FCLK, it seems people have misread or forgotten. Hallock said that the BIOS will help those that CAN achieve it. He actually says that those are golden samples and do not represent the thick of the product. 1900 is the default to be expected, 2000 is the extraordinary.
Edit: Also, thanks for the video, HWUB
I'm pretty sure you're in my PC. In a recent vid you spoke about the gains of upgrading from a i5 7600k to a Ryzen 5 5600X and now you recommend exactly the RAM-kit I bought. And yes at least I just pop those crucials in, activate XMP and get to gaming. No need to tinker in the BIOS and tighten the timings, especially for only a few % improvements.
Answered in just 5 minutes
Bravo HUB 👍
When you have a 4k screen and none of this matters at all but still i kinda wanne know xD
Lol I asked this question to Google on the 5th of November. So to see this video makes me smile and happy 😊 to see a lot of people are asking the same question. I'm glad you guys get the time to test everything so the rest of us can make better and more educated spending/buying decisions!
Wow, just WOW!
This came out so quick with much more information than I ever wanted to know. I didn't even expected you to touch this topic before the launch of the radeon 6000 series
He rushed it IMO, he used 2 single rank tuned sticks and compared it to different brand/speed 4 untuned sticks.
Steve said Wendell said 2x16GB is the new sweetspot :)
Good. Cause that's what I bought
Because most 16GB modules available on market are dual rank, so 2x16GB = 4 ranks, 2 per channel.
Remember 2x16GB can both be single rank or dual rank. As this year memory manufacturers released their new 16gbit modules.
Im no tech guy and I am slightly confused about which ram should I go for. I got a rough idea in the video but still cant decide whether www.newegg.com/global/ae-en/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232860 provide the best performance or not. (I will post it up again and look for other peoples opinions)
@@Commentrix I using this one and they are dual rank per module, so best performance if you asking about ranks www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232861
HW upload:
Pressing video as fast as ryzen smoked intel
So in about 2-3 years?
Super informative. Exactly what I was looking for. We appreciate your dedication. You may rest now.
I built five Ryzen computers. Two were identical: Corsair Spec Alpha case, EVGA 550 PSU, ASUS PRIME-350 motherboard, Ryzen 3 2200G APU. The red one was noticeably fast, like an i7, flowed like water, pleasingly fast. The yellow one not noticeably fast. In fact, if you opened a lot of tabs, 20-30-40, in two browsers, they would totally stop opening. The only difference was this: the red one had a Samsung 500GB SSD, the yellow one had a Samsung 250GB nvme. The red one had 1 x 8GB Corsair Vengeance LPX ram; the yellow one had 2 x 4GB Corsair Vengeance LPX ram (both 3000Mhz). I know they say 2 sticks of RAM is better, then they went back to 1 is better. All I could figure is that 1 stick of RAM is better. Otherwise it's the SSD that's faster, or a combination. What do you think?
Faster than notification!
Best ryzen 5000 memory behavior explanation video so far. Nice work.
One of those channels when I hit 👍🏻 before I even start watching the video. And, obviously, for a good reason.
Thanks for clearing my confusion on this subject.