Yup, that's the question I immediately wanted to ask. BTW it seems like 3200 cl14 2x16 are almost all gone, is there any other way to find a dual rank kit?
I wish there were b-die only bins included so we could know about ranks on those at least. That said, even what is included is a ton of pretty time-consuming testing, and at least it spreads the word to NOT just allow the board to set Gear Mode and leave it. It's criminal that this board defaults to Gear 2 at 2400Mbps.
TL:DR DDR4 CL 16 3200 Mhz is the best value for ram in this current day, and DDR5 will only be worth it when it becomes cheap and more optimized like DDR4 is right now
His settings are horrible, even the DDR4s. No one uses intel with higher than 50ns latency. Infact most goes below 40ns. Nowonder his intel benchmarks especially gaming is so poor.
@@blindguardian8599 nah.. it's just the lack of binned chips at the moment due to volume. Some tuners with better silicon lottery on value rams are already doing 6700c27 with 40~ ns latency on hynix DDR5 dies.
Big thanks for this video! There hasn't been that many resources on DDR4 memory scaling online for Alder Lake and was wondering if the old Intel minimum gains with memory scaling is still true. As I'm looking for 2x32=64GB, there is a much more expensive price gap from 3200CL16 to 3600CL16 and it barely seems to be worth the extra money for almost no gain in the two producitivty apps tested here and just a few frames in games.
How is $5-10 a large gap? That is for a 16GB kit, so $20-60 for you. If you are already paying several hundred, I don't see why $20 or even $60 is an issue. On a Ryzen 5000, there is a decent increase with 3600 over 3200. Maybe not with Intel, but it is a small dollar difference when you are ready spending $2000+.
@@farmeunit in some regions there is a 50 dollar gap. Here,I can get 32gb 3200 cl 16 for 90. However for 3600 cl 16, the minimum is 150-160. And I'm broke after getting a gpu. So...
@@farmeunit As others have also mentioned, in some regions 3200 is the more saturated and competitive market, while anything beyond that is priced by local retailers as a premium product.
So I play a lot of simulation games. Factorio, KSP, Rimworld, Stellaris etc. None of these games need high frame rates. Most of them are more likely to be CPU limited rather than GPU limited (at least assuming you have something better than integrated graphics). So I'm disappointed that this roundup only included 2 real world application benchmarks, along with a selection high frame rate games that tend to be mostly GPU limited. I would have liked to have seen graphs for at least code compilation and Factorio as well. I understand why you include the GPU limited games, but this sort of testing would have benefited from showing more diverse applications that are CPU/memory bound. Personally for me, a memory scaling benchmark implies a more detailed testing process. Eg using your DDR4 3200 CL14 (or whatever) memory, you have a latency scaling section where you test at CL14, CL16, CL18 and CL20, all at 3200 data rate, and we get to see which applications and games are sensitive to lower latency, and then a section where you using keep the memory timings the same (eg all CL18 or CL20 or whatever), and test at 3000, 3200, 3600, 3800, 4000 etc so we can see which games/applications are sensitive to memory bandwidth. Yeah, you can't do that across DDR4 and DDR5. But doing a good job of isolating which games/applications are more sensitive to bandwidth, and which are more sensitive to latency means that when you do show the DDR4 and DDR5 graphs in this video it is more obvious why a game/application favours DDR4 or DDR5.
Yeah I am also wishing for a better testing set when looking at ram specifically. I am curious how this performs in other 'simulation' workloads, like matlab, mathematica, etc.
Dude I get what you mean but come on just to add a single new kit of memory or different cl timings is so much work and it's honestly unfair on Steve he can't possibly test every single thing in every single scenario and while you are 100% correct that he could have included more cpu bound games or apps but honestly the vast majority of gamers generally are GPU bound playing AAA games at 4k high settings for eg with a 3090 or 6900xt (probably relevant GPUs for anyone buying a 12900k and especially ddr5) so I can kinda see what Steve's rationale is.
@@theigpugamer isnt that his job? Personally the test wouldve been done better with the same kit of ram tuned to different speed and latency with a fixed sub timing table on both DDR4 and DDR5 setups for better consistency. As of now, its a mixture of dies and even ranks. This video sounds more like a ram kit xmp review rather than a DDR4 vs DDR5 performance comparison. Pretty sure if you throw in a gskill kit or kingston kit with the same xmp profile, the result will be different. Even the aida benchmark result is all over the place.
It's interesting that in the 4000 kit loses to 3600 in many cases (and sometimes even 3200) even though it has both higher bandwidth and lower latency.
Yeah only think missing from the video was if Steve could have mentioned if the kits where single or dual rank. Because I'm guessing that is what made the 3600 kit beat the 4000 kit. But we don't know. Could also just be better training with the motherboard so the secondary and tertiary timing ended up better on the 3600 kit? We need a bit more info, Steve :D But at the same time: huge thanks for the video, Steve. Very good and interesting as always!
That's some pretty horrible latency numbers even on DDR4. Gigabyte boards maybe? Usually and ideally it should be set below 40ns for DDR4 and below 50ns for DDR5. Maybe that's why his gaming benchmark is kinda poor.
It's not interesting at all its a v ery bad set. i run a 9900k with a 4000 cl16-16-16 and i have a latency of 40ns which is way lower then this crappy set. The true latency of this 3600 cl16 is better as 4000 cl18 (8.8 vs 9 ns) Frankly he weasted time testing this kit as its not anything someone should buy.
@@pino_de_vogel indeed. Mine was 4400c16 35ns on a 10600K lol... also his numbers were all over the place. He should fix the sub timing table then the result wouldve been more consistent... this is more like a ram kit xmp and motherboard review rather than a DDR4 vs DDR5 comparison. Afterall, even the motherboard is different.
Thanks Steve for a really good detail comparison, I never use my upcoming PC for any rendering work. Seems like 3200Mhz CL16 is still not too bad for my 12700f + b660 with RTX 3070. Especially play games in 1440p, I see very minimal difference.
I really hope they do an updated video with current ddr5 vs ddr4 and 13th gen intel. I say this with the selfish desire to know if I can upgrade my CPU from a 12700k to a 13900k and actually gain something from it with ddr4 3200 cl16 or if any gains would require ddr5
So, it seems that as it was will all of the previous memory technology transitions that during the first generation of the new technology it makes some small wins in some applications, some small losses in others, and costs a lot more while doing it. As usual I'll gladly wait for the next iteration where the gains should be more universal and the costs more in line with the previous generation.
Hey Steve, could you please make a video of the best B660 mobos to buy for the budget builds. The 12400f & this RAM guide will also be very helpful in making the budget best value builds! Thanks for everything! Looking forward to the upcoming videos
I'm making one up now. 12400f €120 2x8gb Crucial Ballistix BL2K8G32C16U4B, apparently the "B" at the end is for Samsung die. €60 ASRock B660M PRO rs €130 Thinking about the Cooler Master box Nr200 with ATX psu adapter. €110
@@fufu1128 I would also go for the 12400f but I don't have a GPU and can't afford one in this market so will probably go for 12400/12500 based on my regional pricing
@@chandans9846 Yeah I feel your pain. I got a second hand rx570 to give me output for now. I think the 12400 is around €160-170 with igpu, so expect €50 more, but you can get video out at least✌🏻
@@chandans9846 Yeah I feel your pain. I got a second hand rx570 to give me output for now. I think the 12400 is around €160-170 with igpu, so expect €50 more, but you can get video out at least✌🏻
@@fufu1128 the "B" at the end doesnt stand for B die it literally stands for "Black" as in black kit. Try typing the same part number but with W instead and you ll get the white kit or with R and you ll get the red kit. Cruical is owned by micron and iirc they only sell micron kits so the sumsung Bdie is off the table. Nowadays it doesnt make sense to gamble for Bdie. If you want a decent one go for gskill 3200 C14 14 14 34 since that one is guaranteed to be Bdie and shouldnt be too hard to oc
Thanks for this Video. It gives one a general overall perspective of DDR4 vs DDR5, at this time. When DDR5 performance goes up,(and it will) and pricing comes down,(and it will) we may see a better incentive to buy into DDR5. I am sure that when the time comes, you will have a video on it. Great Job, Thanks again.
Would love to see Total War games included in RAM benchmarks in future. They are some of the most memory bound games in benchmarks. Star Citizen is also very memory bound, but good luck figuring out a way to test it and with all the changes it undergos.
I swear you guys read my mind and always come out with videos of what I'm looking for literally the day after I start researching stuff thanks for the info you guys rock!
you can buy extremely cheap bdie (4400 c19 patriot vipers) for even less and get waay better performance. unfortunately people will buy big name components and have no idea
@@Milkykt I mean, sure, if your Mobo and/or CPU support it. Personally I "only" have G.Skill Aegis 2400 and it works well enough for what 90% of users will ever need.
Thanks for this video! I miss 3600 CL14, which I find the best kit for DDR4 build with either a old 8700K/9900K or 12900K or even a 5950x. Simply the best kit for XMP.
As mentioned, GSkill Ripjaws V 3600 CL16 seems to be the go-to & best buy DDR4 memory right now...going for another 2x8 kit next week for the new i5 12500 Z690 build.
My strategy when building right now would be to buy the cheapest single stick of DDR5 available (160 bucks here) so you can buy a DDR5 motherboard already. When DDR5 prices come down you can do a bios update, upgrade your ram to something sweet and sell the old stick. Over the years a second-hand DDR5 board will be worth more than a DDR4 board.
That's a highly flawed stragedy that will get you stuck in upgrade libido. By the time you wait for the prices to come down, you'll have already migrated to a better motherboard series and much superior processor. I'd give it a nice 2-3 years before we see reasonable prices and good latency numbers
covering the IGPUs with different memory kits could be interesting as well especially when alder lake laptops are easily available which would maybe have enough graphics performance that you might care about the uplift.
I would like to see some tuned DDR4 in this chart like 3600mhz cl14 or 4000mhz cl16. It would give a good representation how tuned DDR4 compares to stock (XMP one) and stock DDR5.
All that was available when I put my 12th gen build together was 5200Mhz kits and with the high CL40 timings it wasn't really ideal. After a lot of benchmarking, tweaking and testing I've been able to tighten the timings on my RAM and increase integer and float point performance giving me a boost to my max frame rate, 1.0% and 0.1% lows in games that's almost on par with DDR5-6000 performance. I did this because no matter what voltages I tried I couldn't get the RAM to run at a higher frequency, sometimes I would get it to boot but then run into crashes once I got to my desktop. So it is possible to squeeze better performance out of the lower frequency DDR5 kits without increasing the frequency but it was a lot of messing around in my BIOS to get stable tighter timings.
When I posted on Ars that I'd gotten 3600 to go with a 12400f, someone said that wouldn't work well and pointed me to your testing. Much relieved by your Final Thoughts when I realized I'd actually ordered 3200 16-18-18-38.
Could you include some audio applications (especially VST plugins and complex audio processing setups) in these kinds of tests as well? The scenario is entirely different from all these other tests since there is a very strict realtime constraint placed on the system. DSP is very sensitive to both memory latency and bandwidth depending on the exact scenario, and would be extremely valuable information for music producers and other audio professionals (and hobbyists).
I'm disappointed that we don't see any fast B-die bins in the DDR4 results. 3600 16-16-16-36 or 4000 16-16-16-36 would have been nice to see because they are cheaper than most DDR5.
They really are. One of the cheapest RAM kits in Australia right now is $150 for 16gb of 3600c16 B-die. Almost misleading. Then again, it pretty much is just an ad for corsair
Please do a video on the difference between 2 and 4 sticks of memory for Alderlake. I've just got a 12700F and reused my old 16GB DDR4 3200 CL14 memory. It's in 2 sticks. (G.Skill DDR4-3200 16GB Dual Channel [Flare X] F4-3200C14D-16GFX (F4-3200C14D-16GFX) - $158.82) I decided to expand to 32GB just because I could and I saw the exact same memory kit for sale for around AU$160 which I think is about half what i paid for the original kit back in 2017. I saw a video somewhere (maybe GN) that showed around 10% improvement in gaming performance when switching between 2 and 4 sticks. I would like to know if that carries over to Alderlake as well.
You get 10% improvement is not because of 2 and 4 sticks of RAM. It is because your existing RAM kit is a 8GB stick X 2 . Single rank memory per memory channel. There are only 8 memory chips on each stick. If you put 2 sticks of single rank memory on 1 memory channel, effectively you get dual rank memory, total 16 memory chips per channel. Your 10% gain is from the increased memory copy speed. However, you will still get the same improvement if you use 2 sticks of 16GB DDR3200 CL14 memory . In short, you gain performance from transitioning from single rank memory to dual rank memory.
The difference is in memory ranks ,not the overall amount of sticks. 2 sticks of dual ranked memory will be roughly equal to 4 sticks of single ranked memory.
@@Zerian_B13 Yeah I guessed that was probably the case. Given it's almost impossible to know whether any given kit will be dual rank or single rank before buying and pretty much all 16GB kits being single, I still think it's fair to just say 4 sticks vs 2 sticks.
@@youtubevanced4900 It's possible to work out how many ranks you have if you know what memory chip/die your modules use. Samsung B-die chips are 8Gb (1GB) each, so are always 8GB per rank. DDR4-3200 CL14 is pretty much always Samsung B-die, because no other chip can reliably run at such low latency without having a much higher frequency. Actual CAS latency in ns = CL divided by internal clock frequency (half of advertised frequency) in GHz. < 8.75ns is probably B-die (though might be a high bin of Rev E/B or CJR/DJR. These can usually be indentified by poor primary timings other than CL) < 8.5ns is definitely B-die 14/1.6GHz = 8.75ns
I just reused my 6 year old 4x8GB 3200CL16 kit, getting another round of use from my old RAM was pretty much the reason I went for a upgrade in the first place. For old, 4 dimm dual rank e-die kit on a pretty basic board (Z690-P D4), I'm not unhappy it's running fine at 3600 16-18-18-36. There might be some room to improve the timings, but that's as much frequency as they'll post with.
This channel is indeed very informative, without a doubt, but I watch their videos to fall asleep. There is something about their voice that is just too soothing.
An interesting video idea you guys could do is compare DDR3, DDR4 and DDR5 for the fun of it. I know there's certain limitations but I think it'd be cool to see it!
DDR3 over 2000mhz below CL10 can still hang pretty well honestly. The thing is it can't be tested with any cpu newer than a 5775C so the tests couldn't really be 1:1
System RAM bandwidth FASTEST DDR4 @3600 - 54GB/S M1 - around 60GByte/s Steam deck (4channel) - 88Gbyte/s M1 pro lpddr5(4 channel)- 200 Gbyte/s M1 max lpddr5(8 channel)- 409 GByte/s All these products use lpddr5 with different amount of channels. Now my point. These are SOC's & APU's that have only iGPU to work with so the system RAM bandwidth matters greatly. Since the VRAM is just the System RAM so it needs to be faster hence the need for higher bandwidth. So the ultra thin notebooks, 2in1, small tablets using these type of SOC's will not only greatly improve the performance of gaming on IGPU But also every other system tasks requiring high bandwidth like CAD, animation, scientific softwares that use upwards of 500GB of RAM etc.
I know this is for the new intel alderlake cpu's but anyone looking at this and thinking they might get similar results with AMD should note anything above 3600 on AMD will not get this kind of scaling due to the infinity fabric limitations. Pretty sure Hardware unboxed have another video showing amd memory scaling so go check that out.
G.SKILL Trident Z Royal Elite Series 16GB (2 x 8GB) 288-Pin DDR4 SDRAM DDR4 4000 (PC4 32000) Desktop Memory Model F4-4000C14D-16GTES i got the G.SKILL Trident Z Royal Series 32GB (4 x 8GB) 288-Pin DDR4 SDRAM DDR4 4000 (PC4 32000) Intel XMP 2.0 Desktop Memory Model F4-4000C16D-32GTRS
Thank you for the high quality work! I don't know how you guys manage to put out so much great content between testing, writing, filming, getting sponsors, keeping in touch with industry contacts, keeping in touch with the community like being on the Moore's Law is Dead podcast... do you guys even get to game in a satisfying way? It bears repeating: I REALLY APPRECIATE the hard work you put in. I love your channel just as much as the other big ones. Keep it up!
Thanks Steve, great mainstream memory comparison, but I feel sorry for you when many of the comments will ask for why you didnt max each kit manually and spend 100 more hours on this. Thank you for the XMP Kit comparison as it stands with Corsair kits.
I think most people dont understand the point of this faster ddr5/lpddr5/lpddr5x memory. It may not improve a typical gaming performance by much but the main improvement is in the iGPU performance,any software that is highly dependent on system RAM like cad,video rendering, animation etc. These type of things will benefit from the higher speeds. This is the same reason why ryzen mobile APUs like 5800u with the vega 8 gpu has huge improvements when used with a faster system RAM. It's simple logic. Since all mobile & desktop apu have only iGPU to use then the system Ram becomes the VRam for igpu & more the bandwidth higher the performance. Take steam deck for an example. It has lpddr5 @5500 in quad configuration i.e. 88Gbyte/s same as the ddr5 6400 fpr desktop shown here. Also one reason for the high performance of apple m1 pro & M1 max is their memory bandwidth of 200Gbyte/s & 409Gbyte/s & since both the SOC's have iGPU hence them benefiting from the higher bandwidth & more importantly the whole system could use the same memory to do other tasks much more faster. A ps5 with windows installed on it will benefit greatly from the memory bandwidth that is used for the whole system & not like a pc where the gpu has equally fast gpu memory bandwidth but slower system RAM bandwidth.
I'm 37 now (doing this tech stuff since I'm 7) and I can predict the results for DDR-27 99999999Ghz in 250 years right now: It's going to be the same result. It's always the same for every generational switch. It doesn't make sense to buy the new Ram Version for at least the first year, because it always costs at least 50% more (often times 100% more) for 5-10% more performance.
Pity Corsair didn't send you the Dominator CL14 DDR4-3600 I just snagged 32gb's of from Newegg. Would have liked to see how it rates vs DDR5. (the 3600 you guys tested was relatively high latency)
Thanks for including Photoshop, the graphs was very helpful in making me choose. I was thinking initially that in PS a 3600 cl16 would be a faster and cheaper choice then existing DDR5 at the moment (but not future proof). Now I'm inclined to choose a 4800 cl40 at double the price of ddr4 (I would not go for 6200 cl38 atm cause the prices are 3-4 times higher then DDR4, and the performance compared to lower tire DDR5 is little).
At twice the price you're still better off just getting DDR4 and then reselling it later when DDR5 is cheaper. You'll likely be able to get a better DDR5 kit at that time for a reasonable price, as well.
@@nagorak666 Exacly my solution , i got 64 gigs of 3600 , with a 12900k see no point in spending 700$ just to get DDR5 with a mombo , when I only spend 350$ for my ddr4 ram plus the mother board
I started my 12900k build with dominator 5200 c38 and it ran great. I then moved to dominator 5600 c36 and there was a slight increase in performance. I then moved to gkill trident z5 6000 c36 and it was not stable at all even on just the normal xmp profile. Was not happy with the gskill kit. I now have the dominator 6200 c36 as seen in the video and it runs wonderfully. I could not be happier with the performance. It's awesome!
Being a Zen 2 3600 owner and gaming at 1440p, DDR5 wont be something I'm looking at until a future AM5 build (or Intel) and a reason for me to get 8c/16t upgrade.
With AMD Ryzen I believe 3600 Mhz is the sweet spot due to infinity fabric limitations. My son and I had similar 5950X rigs, and with Asus motherboards and Corsair H115i coolers - his 16GB x2 CL 14 3600 G.skill RAM gave him about 5% better 3DMark CPU performance than my 16GB x4 G.Skill CL 16 3600 MHz RAM. IIRC I got about 2% higher benchmarks going from 2 sticks to 4 sticks of C16-3600, but did not catch up to him. But, I think I got about 10% improvement when I went from CL16 3200 RAM (8GBx2) on my Ryzen system to CL16 3600 (16GBx2 before I added 2 more sticks).
I'm in need for a 2x16GB kit for my 12700k and I had too much tunnel vision on Samsung B-Die kits like 3600C14 and better. I now found out this Corsair kit is only $266 on NewEgg. Is there any reason to get the Corsair 4600C18 instead of a 3600C14 BDie that costs roughly roughly the same?
@@Ben-ld1qi you wont be able to run XMP in gear1 with the 4600c18kit. instead you will have to manually clock down the ram and tighten down the timings. so if you already have to do that, you gonna want to get a B-Die kit like the 3600c14 and simply overclock it
wouldn't a cpu with lesser performance be better for a test of this nature since ram speed / latency seems to make more of a difference when the cpu is the bottleneck? something like the 12600k or 12400 come to mind🤔.
I still have a single 2133 MHz stick in my computer. I think it's a limitation of H110, but either way the terrible speed is evident in games when going over my mediocre 2 GB of VRAM.
In my country these awesome DDR4 are basically same price, less then 50$ difference, there is no reason to ignore the upgrade to DDR5 if you are building a setup like me right now.
I want to see the same test with CL14-14-14-34 DDR4 at 4000mhz or 4133mhz in gear 1... And with the upcoming Ryzen 7000AMD/13th gen intel CPU with only DDR5 support i think it's a very important test to do.
I bought the ddr4 version so I could just keep my ddr4 3200 cl16. Because ddr5 is and was insane. Glad to see that's pretty much the budget sweet spot overall.
Also for any cpu and memory testing it would be really nice to see more esports titles at low res and details, this is area where cpu and gpu matter the most and you want to pump as many frames. The games you tested there are irrelevant for this kind of testing, except for rainbow 6
Yeah these tests are not great because the memory isn’t manually tuned, which makes a huge difference with Alder Lake. On my 12900k at 3800Mhz C15 with the Aida64 benchmark I get 45ns of latency, whereas the best hardware unboxed could do was like 56ns. At 4000Mhz c15 I can push 43.5ns, but I wasn’t stable with my kit (every single timing tuned, from primary to tertiary to turn around)
These XMP kits were bottom of the barrel as another poster said. Not trying to promote anything but IMHO another RUclipsr called FrameChasers did a way better maxed out DDR4 vs DDR5 gaming comparison (only gaming tho). The DDR4 kit was 4000C14 with tweaked subtimings and I can't remember what the DDR5 was, but also tightly tuned.
@@xblur17 Very nice mate, I'm waiting for my 12700k DDR4 build to arrive. I'd be quite happy with 3800C15 and C14 would be a dream haha. Which kit are kit using bro? I'm thinking of pulling the trigger on a TridentZ Neo 2x16GB Dual Rank 3600 14-15-15-35 1.45v, going at $270 right now. Alternatively I can save a bit and get Corsaor Dominator Platinum RGB (DR) 3600 14-16-16-36 1.45v for $215, but from what I've read on forums, I don't have too much faith in current Corsair's B-Die kit, it seems that most people wishing to tune & OC their kit opt for GSkill.
@@Ben-ld1qi TridentZ Neo 2 x 8gb, incredible kit! Manually tuned b-die and a manually OC’d processor do so much for performance, especially in competitive game titles. Best of luck with your build man, it’s gonna be great!
Built a new system over the holidays, haven't built one since 2012. Went with the 12700k and 64gb of DDR4 3600. Mobo's are also more affordable if you go with the DDR4. Don't get me wrong, I wanted the ASUS formula for my all white build, but the ROG Strix was just fine.
I have 3000 MHz with i7 12700k worth upgrade If I used DLSS in a monitor at 1440p I know with DLSS goes to rescaling at 969p something l played Marvel Guardians of the Galaxy and I notice using DLSS with Ray tracyng OFF GPU its at 70 % with Rtx 3070 ti but in other games like cyberpunk always had 99% GPU usage even using DLSS so can somebody explain me if its the game or its the system I didnt see no one explain some games arent optimized with DLSS ? Hope can help me
May i just ask you something? So i have a b350 tomahawk and 1600x for about 4.5 years now, i wanted to upgrade to ryzen 5600x or the 5800x, but obviously it wasnt supported, so the only real choices were 3600 and 3700x. Until a few weeks ago when there have apparently been some bios updates that unnoficially add support for ryzen 5000 cpus. Could you verify and make a video about it? Its kind of a big deal but don't see anyone covering it. Cheers!
PSA! Alder Lake non K CPUs have locked voltage control (VCCSA) for Memory Controller (IMC) to 0.89-98v. If you are going for anything above 3200Mhz Gear 1 or 5400Mhz Gear 2 on non K processor, you may be unable to run at those speeds, it's a lottery! This should had been mentioned in the review, my 12700F which I returned was not able to run 3600Mhz CL18, best value RAM. I got 12700K instead and even on b660 the VCCSA is unlocked, so the CPU are locked!
Hmm it would be really good to know if the PCIe bus/RAM interface is the limitation for the CPU in some of these cases. DDR5 offering a lot more bandwidth might be allowing the CPU to run code better while DMA is transferring textures to/from graphics cards.
Help me please . I just buy a onn monitor for my son is a 1080p resolution . The problem is when i connect with a display port the monitor showing 1080p but when i plus my hdmi cable showing me 4k resolution any ideas please?
Hi, are you planning to do a B660 mainboard test like you did with the B560 series boards? I‘m kind of afraid to buy anything because B560 was such a mess. Thanks!
This is a dumb question but my google fu failed me. My asus mobo didn't have a gear mode choice that I could see or notice. Z690 tuf d4. Where in the bios is the gear mode option? I have seen gear mode clearly labeled in a friend's ryzen uefi so I was wondering if it will be called something else on an intel mobo?
I'm going to wait to upgrade from my current system until DDR5 is not so ridiculously over-priced relative to the performance. I will be tweaking my current DDR4 installation for more memory and better latency.
Man, so many questions. If you had some 3200/14 memory do you think it would have been the best? Have you ever lowered a ram's speed to get better timings? Can you do a similar test across all Ryzen generations? 😅
"Have you ever lowered a ram's speed to get better timings?"... Uh, that doesn't give you better timings. The big problem here is that timings are given in clock ticks and not nanoseconds like they should be. A 6000CL38 kit has the exact same command latency as a 3000CL14 when measured in nanoseconds.
I could not afford a DDR5 motherboard and DDR5 RAM. I have 16GB x 4 with timings 16, 19, 19, 39. Best I could afford with my Z390 motherboard. Thank you for this review.
I'd also wonder, how some DDR4-4000 CL14 Kits would compare against the DDR5-6200 CL36 Kits. Because in my country, the first costs around 530 USD compared to only 540 USD for the DDR5 stuff.
Is DDR5 worth it just for future proofing? Getting some lower tier sticks now, and then once you feel like you need to upgrade in the future when it’s gotten cheaper and better, to get some better sticks without having to change the motherboard as well?
You guys are awesome. so many info and the way you present it, fantanstic! I suppose all that data refer also to 16 ( 2x8 ) kits? because I dont see a massive reason to upgrade to 32 for now, and I own a 2x8 3600 cl16 ballistix set.
Something looks wonky with the Hitman 3 charts. The 1440p chart is visually what I expected based on the other games, but the FPS measurements are higher than 1080p.
are these memory kits dual or single rank? because these days 2x16 DDR4 kits can be both.
Yup, that's the question I immediately wanted to ask.
BTW it seems like 3200 cl14 2x16 are almost all gone, is there any other way to find a dual rank kit?
Any upcoming memory OC records? XD
I wish there were b-die only bins included so we could know about ranks on those at least. That said, even what is included is a ton of pretty time-consuming testing, and at least it spreads the word to NOT just allow the board to set Gear Mode and leave it. It's criminal that this board defaults to Gear 2 at 2400Mbps.
The 4600mhz kit he used ( 4000mhz in the test) is single rank
Isn't here a lack of propper DDR4 Kits? there are no propper B-Die kits included on the DDR4 side but something highend like a 6200Cl36 kit on DDR5
I can't imagine how long this testing took! We'll done, big kudos to you! 👍
TL:DR DDR4 CL 16 3200 Mhz is the best value for ram in this current day, and DDR5 will only be worth it when it becomes cheap and more optimized like DDR4 is right now
even got one pair of 16gb sticks with these specs for my laptop for 150 euros:)
His settings are horrible, even the DDR4s. No one uses intel with higher than 50ns latency. Infact most goes below 40ns. Nowonder his intel benchmarks especially gaming is so poor.
and that time will be in like 2 years from now
@@blindguardian8599 nah.. it's just the lack of binned chips at the moment due to volume. Some tuners with better silicon lottery on value rams are already doing 6700c27 with 40~ ns latency on hynix DDR5 dies.
Big thanks for this video! There hasn't been that many resources on DDR4 memory scaling online for Alder Lake and was wondering if the old Intel minimum gains with memory scaling is still true. As I'm looking for 2x32=64GB, there is a much more expensive price gap from 3200CL16 to 3600CL16 and it barely seems to be worth the extra money for almost no gain in the two producitivty apps tested here and just a few frames in games.
How is $5-10 a large gap? That is for a 16GB kit, so $20-60 for you. If you are already paying several hundred, I don't see why $20 or even $60 is an issue. On a Ryzen 5000, there is a decent increase with 3600 over 3200. Maybe not with Intel, but it is a small dollar difference when you are ready spending $2000+.
Quiet happy with my 48gb of 3200 cl16 on my 5900x :)
@@karlreading3201 I have the same chip, but I went for 32GB of 3600 cl14.
@@farmeunit in some regions there is a 50 dollar gap. Here,I can get 32gb 3200 cl 16 for 90. However for 3600 cl 16, the minimum is 150-160. And I'm broke after getting a gpu. So...
@@farmeunit As others have also mentioned, in some regions 3200 is the more saturated and competitive market, while anything beyond that is priced by local retailers as a premium product.
First-gen of just about any new memory standard has always been a wash overall vs solid yet still affordable mainstream memory.
With the exception of perhaps the original DDR memory.
So I play a lot of simulation games. Factorio, KSP, Rimworld, Stellaris etc. None of these games need high frame rates. Most of them are more likely to be CPU limited rather than GPU limited (at least assuming you have something better than integrated graphics).
So I'm disappointed that this roundup only included 2 real world application benchmarks, along with a selection high frame rate games that tend to be mostly GPU limited. I would have liked to have seen graphs for at least code compilation and Factorio as well.
I understand why you include the GPU limited games, but this sort of testing would have benefited from showing more diverse applications that are CPU/memory bound.
Personally for me, a memory scaling benchmark implies a more detailed testing process. Eg using your DDR4 3200 CL14 (or whatever) memory, you have a latency scaling section where you test at CL14, CL16, CL18 and CL20, all at 3200 data rate, and we get to see which applications and games are sensitive to lower latency, and then a section where you using keep the memory timings the same (eg all CL18 or CL20 or whatever), and test at 3000, 3200, 3600, 3800, 4000 etc so we can see which games/applications are sensitive to memory bandwidth. Yeah, you can't do that across DDR4 and DDR5. But doing a good job of isolating which games/applications are more sensitive to bandwidth, and which are more sensitive to latency means that when you do show the DDR4 and DDR5 graphs in this video it is more obvious why a game/application favours DDR4 or DDR5.
Yeah I am also wishing for a better testing set when looking at ram specifically. I am curious how this performs in other 'simulation' workloads, like matlab, mathematica, etc.
Dude I get what you mean but come on just to add a single new kit of memory or different cl timings is so much work and it's honestly unfair on Steve he can't possibly test every single thing in every single scenario and while you are 100% correct that he could have included more cpu bound games or apps but honestly the vast majority of gamers generally are GPU bound playing AAA games at 4k high settings for eg with a 3090 or 6900xt (probably relevant GPUs for anyone buying a 12900k and especially ddr5) so I can kinda see what Steve's rationale is.
@@theigpugamer I think fewer timings on more tests would be better but maybe that's the next video in the oven.
@@QuantumConundrum you never no Steve he's a madlad he can benchmark anything
@@theigpugamer isnt that his job? Personally the test wouldve been done better with the same kit of ram tuned to different speed and latency with a fixed sub timing table on both DDR4 and DDR5 setups for better consistency.
As of now, its a mixture of dies and even ranks. This video sounds more like a ram kit xmp review rather than a DDR4 vs DDR5 performance comparison. Pretty sure if you throw in a gskill kit or kingston kit with the same xmp profile, the result will be different.
Even the aida benchmark result is all over the place.
It's interesting that in the 4000 kit loses to 3600 in many cases (and sometimes even 3200) even though it has both higher bandwidth and lower latency.
Yeah only think missing from the video was if Steve could have mentioned if the kits where single or dual rank. Because I'm guessing that is what made the 3600 kit beat the 4000 kit. But we don't know. Could also just be better training with the motherboard so the secondary and tertiary timing ended up better on the 3600 kit?
We need a bit more info, Steve :D
But at the same time: huge thanks for the video, Steve. Very good and interesting as always!
That's some pretty horrible latency numbers even on DDR4. Gigabyte boards maybe? Usually and ideally it should be set below 40ns for DDR4 and below 50ns for DDR5. Maybe that's why his gaming benchmark is kinda poor.
It's not interesting at all its a v ery bad set. i run a 9900k with a 4000 cl16-16-16 and i have a latency of 40ns which is way lower then this crappy set.
The true latency of this 3600 cl16 is better as 4000 cl18 (8.8 vs 9 ns) Frankly he weasted time testing this kit as its not anything someone should buy.
@@bingbing3464 Yes (40 ns), with tuned sub-timings and B-dies. But not out of the box or with just XMP enabled.
@@pino_de_vogel indeed. Mine was 4400c16 35ns on a 10600K lol... also his numbers were all over the place. He should fix the sub timing table then the result wouldve been more consistent... this is more like a ram kit xmp and motherboard review rather than a DDR4 vs DDR5 comparison. Afterall, even the motherboard is different.
Thanks Steve for a really good detail comparison, I never use my upcoming PC for any rendering work. Seems like 3200Mhz CL16 is still not too bad for my 12700f + b660 with RTX 3070. Especially play games in 1440p, I see very minimal difference.
I really hope they do an updated video with current ddr5 vs ddr4 and 13th gen intel. I say this with the selfish desire to know if I can upgrade my CPU from a 12700k to a 13900k and actually gain something from it with ddr4 3200 cl16 or if any gains would require ddr5
So, it seems that as it was will all of the previous memory technology transitions that during the first generation of the new technology it makes some small wins in some applications, some small losses in others, and costs a lot more while doing it. As usual I'll gladly wait for the next iteration where the gains should be more universal and the costs more in line with the previous generation.
6700k pulled away from the 4790k while starting out almost even in gaming benchmarks after the faster ddr4 showed up.
@@WayStedYou With all the scalping and other factors it will take a long time to DDR5 be the replacement.
In (west) Europe you can buy the very famous crucial 3600 cl16 for 75 80€, for me it's a no brainer 🤔
Thanks for review.
Ordered G-Skill Ripjaws 3200 MHz with my i3-12100F, looks like I'll be fine.
Omg I was looking everywhere for info on this!
Hey Steve, could you please make a video of the best B660 mobos to buy for the budget builds.
The 12400f & this RAM guide will also be very helpful in making the budget best value builds!
Thanks for everything! Looking forward to the upcoming videos
I'm making one up now.
12400f €120
2x8gb Crucial Ballistix BL2K8G32C16U4B, apparently the "B" at the end is for Samsung die. €60
ASRock B660M PRO rs €130
Thinking about the Cooler Master box Nr200 with ATX psu adapter. €110
@@fufu1128 I would also go for the 12400f but I don't have a GPU and can't afford one in this market so will probably go for 12400/12500 based on my regional pricing
@@chandans9846 Yeah I feel your pain.
I got a second hand rx570 to give me output for now. I think the 12400 is around €160-170 with igpu, so expect €50 more, but you can get video out at least✌🏻
@@chandans9846 Yeah I feel your pain.
I got a second hand rx570 to give me output for now. I think the 12400 is around €160-170 with igpu, so expect €50 more, but you can get video out at least✌🏻
@@fufu1128 the "B" at the end doesnt stand for B die it literally stands for "Black" as in black kit. Try typing the same part number but with W instead and you ll get the white kit or with R and you ll get the red kit.
Cruical is owned by micron and iirc they only sell micron kits so the sumsung Bdie is off the table.
Nowadays it doesnt make sense to gamble for Bdie. If you want a decent one go for gskill 3200 C14 14 14 34 since that one is guaranteed to be Bdie and shouldnt be too hard to oc
Thanks for this Video. It gives one a general overall perspective of DDR4 vs DDR5, at this time. When DDR5 performance goes up,(and it will) and pricing comes down,(and it will) we may see a better incentive to buy into DDR5. I am sure that when the time comes, you will have a video on it. Great Job, Thanks again.
Thanks for the video, made my mind up for my next 2 builds as far as memory purchases, very useful.
Would love to see Total War games included in RAM benchmarks in future. They are some of the most memory bound games in benchmarks. Star Citizen is also very memory bound, but good luck figuring out a way to test it and with all the changes it undergos.
I swear you guys read my mind and always come out with videos of what I'm looking for literally the day after I start researching stuff thanks for the info you guys rock!
Halfway through the video and it seems like DDR4-3200 is far and away the most optimal choice in terms of performance and pricing.
you can buy extremely cheap bdie (4400 c19 patriot vipers) for even less and get waay better performance. unfortunately people will buy big name components and have no idea
@@Milkykt I mean, sure, if your Mobo and/or CPU support it. Personally I "only" have G.Skill Aegis 2400 and it works well enough for what 90% of users will ever need.
but there are so few ddr4 motherboards, at least from asus, can't even find a cheap one with 6 sata6 ports.
@@bikingchupei2447 DDR4 Mobos are all I can find locally, lol. 6 SATA ports?
Thanks for this video! I miss 3600 CL14, which I find the best kit for DDR4 build with either a old 8700K/9900K or 12900K or even a 5950x. Simply the best kit for XMP.
As mentioned, GSkill Ripjaws V 3600 CL16 seems to be the go-to & best buy DDR4 memory right now...going for another 2x8 kit next week for the new i5 12500 Z690 build.
shame those kits arent B dies tho, and most likely just some mid-range hynix or micron chips.
@@AdaaDK yeah it's hynix and has kinda loose timings 16 19 19 39
how times have changed... just picked up Trident z ddr5 6000 cl36 for $50 used...
My strategy when building right now would be to buy the cheapest single stick of DDR5 available (160 bucks here) so you can buy a DDR5 motherboard already. When DDR5 prices come down you can do a bios update, upgrade your ram to something sweet and sell the old stick. Over the years a second-hand DDR5 board will be worth more than a DDR4 board.
That's a highly flawed stragedy that will get you stuck in upgrade libido. By the time you wait for the prices to come down, you'll have already migrated to a better motherboard series and much superior processor. I'd give it a nice 2-3 years before we see reasonable prices and good latency numbers
covering the IGPUs with different memory kits could be interesting as well especially when alder lake laptops are easily available which would maybe have enough graphics performance that you might care about the uplift.
I would like to see some tuned DDR4 in this chart like 3600mhz cl14 or 4000mhz cl16. It would give a good representation how tuned DDR4 compares to stock (XMP one) and stock DDR5.
I have the T-Force 32GB (16x2) 3600 cl14 kit. 14-15-15-35. It's awesome.
The studio looks amazing Steve!
04:05 Time in *nanoseconds*.
All that was available when I put my 12th gen build together was 5200Mhz kits and with the high CL40 timings it wasn't really ideal. After a lot of benchmarking, tweaking and testing I've been able to tighten the timings on my RAM and increase integer and float point performance giving me a boost to my max frame rate, 1.0% and 0.1% lows in games that's almost on par with DDR5-6000 performance. I did this because no matter what voltages I tried I couldn't get the RAM to run at a higher frequency, sometimes I would get it to boot but then run into crashes once I got to my desktop. So it is possible to squeeze better performance out of the lower frequency DDR5 kits without increasing the frequency but it was a lot of messing around in my BIOS to get stable tighter timings.
How in Hitman 3 at 1080p you get around 150 fps and at 1440p you get around 200 fps? Doesn't make any sense.
The graphs are swapped around for sure.
Thanks for the deep dive on this!
When I posted on Ars that I'd gotten 3600 to go with a 12400f, someone said that wouldn't work well and pointed me to your testing. Much relieved by your Final Thoughts when I realized I'd actually ordered 3200 16-18-18-38.
Could you include some audio applications (especially VST plugins and complex audio processing setups) in these kinds of tests as well? The scenario is entirely different from all these other tests since there is a very strict realtime constraint placed on the system. DSP is very sensitive to both memory latency and bandwidth depending on the exact scenario, and would be extremely valuable information for music producers and other audio professionals (and hobbyists).
I second this. No one ever does DAWbench and DSP oriented benchmarks.
I'm disappointed that we don't see any fast B-die bins in the DDR4 results. 3600 16-16-16-36 or 4000 16-16-16-36 would have been nice to see because they are cheaper than most DDR5.
They really are. One of the cheapest RAM kits in Australia right now is $150 for 16gb of 3600c16 B-die. Almost misleading. Then again, it pretty much is just an ad for corsair
Please do a video on the difference between 2 and 4 sticks of memory for Alderlake.
I've just got a 12700F and reused my old 16GB DDR4 3200 CL14 memory. It's in 2 sticks.
(G.Skill DDR4-3200 16GB Dual Channel [Flare X] F4-3200C14D-16GFX (F4-3200C14D-16GFX) - $158.82)
I decided to expand to 32GB just because I could and I saw the exact same memory kit for sale for around AU$160 which I think is about half what i paid for the original kit back in 2017.
I saw a video somewhere (maybe GN) that showed around 10% improvement in gaming performance when switching between 2 and 4 sticks.
I would like to know if that carries over to Alderlake as well.
go with 32gb x2 and sell off these two
You get 10% improvement is not because of 2 and 4 sticks of RAM. It is because your existing RAM kit is a 8GB stick X 2 . Single rank memory per memory channel. There are only 8 memory chips on each stick. If you put 2 sticks of single rank memory on 1 memory channel, effectively you get dual rank memory, total 16 memory chips per channel. Your 10% gain is from the increased memory copy speed. However, you will still get the same improvement if you use 2 sticks of 16GB DDR3200 CL14 memory . In short, you gain performance from transitioning from single rank memory to dual rank memory.
The difference is in memory ranks ,not the overall amount of sticks.
2 sticks of dual ranked memory will be roughly equal to 4 sticks of single ranked memory.
@@Zerian_B13 Yeah I guessed that was probably the case. Given it's almost impossible to know whether any given kit will be dual rank or single rank before buying and pretty much all 16GB kits being single, I still think it's fair to just say 4 sticks vs 2 sticks.
@@youtubevanced4900 It's possible to work out how many ranks you have if you know what memory chip/die your modules use. Samsung B-die chips are 8Gb (1GB) each, so are always 8GB per rank. DDR4-3200 CL14 is pretty much always Samsung B-die, because no other chip can reliably run at such low latency without having a much higher frequency.
Actual CAS latency in ns = CL divided by internal clock frequency (half of advertised frequency) in GHz.
< 8.75ns is probably B-die (though might be a high bin of Rev E/B or CJR/DJR. These can usually be indentified by poor primary timings other than CL)
< 8.5ns is definitely B-die
14/1.6GHz = 8.75ns
Super informative. Well done.
I just reused my 6 year old 4x8GB 3200CL16 kit, getting another round of use from my old RAM was pretty much the reason I went for a upgrade in the first place. For old, 4 dimm dual rank e-die kit on a pretty basic board (Z690-P D4), I'm not unhappy it's running fine at 3600 16-18-18-36. There might be some room to improve the timings, but that's as much frequency as they'll post with.
I love RUclips captions.. Today it was welcome to hammer unbox!
Great video as always. I appreciate your effort for making those video
This channel is indeed very informative, without a doubt, but I watch their videos to fall asleep. There is something about their voice that is just too soothing.
An interesting video idea you guys could do is compare DDR3, DDR4 and DDR5 for the fun of it. I know there's certain limitations but I think it'd be cool to see it!
DDR3 over 2000mhz below CL10 can still hang pretty well honestly. The thing is it can't be tested with any cpu newer than a 5775C so the tests couldn't really be 1:1
System RAM bandwidth
FASTEST DDR4 @3600 - 54GB/S
M1 - around 60GByte/s
Steam deck (4channel) - 88Gbyte/s
M1 pro lpddr5(4 channel)- 200 Gbyte/s
M1 max lpddr5(8 channel)- 409 GByte/s
All these products use lpddr5 with different amount of channels.
Now my point.
These are SOC's & APU's that have only iGPU to work with so the system RAM bandwidth matters greatly. Since the VRAM is just the System RAM so it needs to be faster hence the need for higher bandwidth.
So the ultra thin notebooks, 2in1, small tablets using these type of SOC's will not only greatly improve the performance of gaming on IGPU But also every other system tasks requiring high bandwidth like CAD, animation, scientific softwares that use upwards of 500GB of RAM etc.
I know this is for the new intel alderlake cpu's but anyone looking at this and thinking they might get similar results with AMD should note anything above 3600 on AMD will not get this kind of scaling due to the infinity fabric limitations. Pretty sure Hardware unboxed have another video showing amd memory scaling so go check that out.
G.SKILL Trident Z Royal Elite Series 16GB (2 x 8GB) 288-Pin DDR4 SDRAM DDR4 4000 (PC4 32000) Desktop Memory Model F4-4000C14D-16GTES
i got the G.SKILL Trident Z Royal Series 32GB (4 x 8GB) 288-Pin DDR4 SDRAM DDR4 4000 (PC4 32000) Intel XMP 2.0 Desktop Memory Model F4-4000C16D-32GTRS
You got the Royals 4000 16-16-16-36? Have you had a chance to tweak them yet?
Good video thanks, answered alot of questions people had 👍
Thank you for the high quality work!
I don't know how you guys manage to put out so much great content between testing, writing, filming, getting sponsors, keeping in touch with industry contacts, keeping in touch with the community like being on the Moore's Law is Dead podcast... do you guys even get to game in a satisfying way?
It bears repeating: I REALLY APPRECIATE the hard work you put in. I love your channel just as much as the other big ones. Keep it up!
Thanks Steve, great mainstream memory comparison, but I feel sorry for you when many of the comments will ask for why you didnt max each kit manually and spend 100 more hours on this. Thank you for the XMP Kit comparison as it stands with Corsair kits.
I just wish they included the 3200CL14 B-die kit they seem to be using in most of their other CPU/GPU reviews.
"Welcome back to Hammer Unbox!" 😄 Hammertime!
I would be interested in a similar comparison for a budget build (e.g. 12100F+6500 XT) where the VRAM is often exceeded
I think most people dont understand the point of this faster ddr5/lpddr5/lpddr5x memory.
It may not improve a typical gaming performance by much but the main improvement is in the iGPU performance,any software that is highly dependent on system RAM like cad,video rendering, animation etc. These type of things will benefit from the higher speeds.
This is the same reason why ryzen mobile APUs like 5800u with the vega 8 gpu has huge improvements when used with a faster system RAM. It's simple logic. Since all mobile & desktop apu have only iGPU to use then the system Ram becomes the VRam for igpu & more the bandwidth higher the performance.
Take steam deck for an example. It has lpddr5 @5500 in quad configuration i.e. 88Gbyte/s same as the ddr5 6400 fpr desktop shown here.
Also one reason for the high performance of apple m1 pro & M1 max is their memory bandwidth of 200Gbyte/s & 409Gbyte/s & since both the SOC's have iGPU hence them benefiting from the higher bandwidth & more importantly the whole system could use the same memory to do other tasks much more faster.
A ps5 with windows installed on it will benefit greatly from the memory bandwidth that is used for the whole system & not like a pc where the gpu has equally fast gpu memory bandwidth but slower system RAM bandwidth.
I have a 12700k with 2x16 4400mhz Cl 18 ram, and got it for $129. I'm not worrying about DDR5 until Meteor Lake
I'm 37 now (doing this tech stuff since I'm 7) and I can predict the results for DDR-27 99999999Ghz in 250 years right now: It's going to be the same result. It's always the same for every generational switch. It doesn't make sense to buy the new Ram Version for at least the first year, because it always costs at least 50% more (often times 100% more) for 5-10% more performance.
Pity Corsair didn't send you the Dominator CL14 DDR4-3600 I just snagged 32gb's of from Newegg. Would have liked to see how it rates vs DDR5. (the 3600 you guys tested was relatively high latency)
G-Skill ddr4 4000 cl-18 is $119 USD on Amazon at the moment. Just scored some for the new 12th gen build. Good to see at 1440p it's decent
Thanks for including Photoshop, the graphs was very helpful in making me choose. I was thinking initially that in PS a 3600 cl16 would be a faster and cheaper choice then existing DDR5 at the moment (but not future proof). Now I'm inclined to choose a 4800 cl40 at double the price of ddr4 (I would not go for 6200 cl38 atm cause the prices are 3-4 times higher then DDR4, and the performance compared to lower tire DDR5 is little).
At twice the price you're still better off just getting DDR4 and then reselling it later when DDR5 is cheaper. You'll likely be able to get a better DDR5 kit at that time for a reasonable price, as well.
@@nagorak666 Exacly my solution , i got 64 gigs of 3600 , with a 12900k see no point in spending 700$ just to get DDR5 with a mombo , when I only spend 350$ for my ddr4 ram plus the mother board
I started my 12900k build with dominator 5200 c38 and it ran great. I then moved to dominator 5600 c36 and there was a slight increase in performance. I then moved to gkill trident z5 6000 c36 and it was not stable at all even on just the normal xmp profile. Was not happy with the gskill kit. I now have the dominator 6200 c36 as seen in the video and it runs wonderfully. I could not be happier with the performance. It's awesome!
@@jamesdoe7605 yeah I returned them.
Being a Zen 2 3600 owner and gaming at 1440p, DDR5 wont be something I'm looking at until a future AM5 build (or Intel) and a reason for me to get 8c/16t upgrade.
With AMD Ryzen I believe 3600 Mhz is the sweet spot due to infinity fabric limitations. My son and I had similar 5950X rigs, and with Asus motherboards and Corsair H115i coolers - his 16GB x2 CL 14 3600 G.skill RAM gave him about 5% better 3DMark CPU performance than my 16GB x4 G.Skill CL 16 3600 MHz RAM. IIRC I got about 2% higher benchmarks going from 2 sticks to 4 sticks of C16-3600, but did not catch up to him. But, I think I got about 10% improvement when I went from CL16 3200 RAM (8GBx2) on my Ryzen system to CL16 3600 (16GBx2 before I added 2 more sticks).
Pricing really changed since a year for the better: 6000Mhz CL30 or 6400Mhz CL32 kit and both cost 133€, both are g.skill 32GB kits.
Wow. I wish I fortunate enough to be able to watch these videos.
Excellent videos as always 👌
I have the Vengeance RT 4600 kit and I can say that it´s OP for Ryzen. Ich can get it to 3800 cl15 easely! Way better than the Dominator kit.
I'm in need for a 2x16GB kit for my 12700k and I had too much tunnel vision on Samsung B-Die kits like 3600C14 and better. I now found out this Corsair kit is only $266 on NewEgg. Is there any reason to get the Corsair 4600C18 instead of a 3600C14 BDie that costs roughly roughly the same?
@@Ben-ld1qi you wont be able to run XMP in gear1 with the 4600c18kit. instead you will have to manually clock down the ram and tighten down the timings. so if you already have to do that, you gonna want to get a B-Die kit like the 3600c14 and simply overclock it
Memory! Thanks! I'll have to check my MSI Tr4 board. Me thinks it's DDR-4. I'm planning on an upgrade soon.
hitman 3: how did you manage to get around 55 MORE avg fps in 1440p then 1080p? 200ish vs 145 - 155ish respectively.
I think he made a mistake with the graphs
Prices have changed and there are other models and lower CLs in DDR5, now. This needs an update.
Steve why don't you test something like the RX 6500xt on ddr4 vs dd5 running out of VRAM that would be interesting if ddr5 can pick up some slack
The 6500xt is trash, and shouldn't be tested with anything at this point.
wouldn't a cpu with lesser performance be better for a test of this nature since ram speed / latency seems to make more of a difference when the cpu is the bottleneck? something like the 12600k or 12400 come to mind🤔.
You need to do more CPU intensive games or based benchmarks, this is where the real performance difference will show not more GPU limited games.
Would be interesting with a 2666 kit. They are very often seen at a discount or as private sales, not so much any 2400 or 2933 kit.
20% price increase for 3600 cl16 kits, is a better deal than crappy memory. Dont cheap out where you shouldnt..........
@@TTks124 Yeah, for just buying a memory kit. But there can be some tempting deals for used complete systems and they often include 2666.
I still have a single 2133 MHz stick in my computer. I think it's a limitation of H110, but either way the terrible speed is evident in games when going over my mediocre 2 GB of VRAM.
Im still on DDR3 my G.
In my country these awesome DDR4 are basically same price, less then 50$ difference, there is no reason to ignore the upgrade to DDR5 if you are building a setup like me right now.
I want to see the same test with CL14-14-14-34 DDR4 at 4000mhz or 4133mhz in gear 1... And with the upcoming Ryzen 7000AMD/13th gen intel CPU with only DDR5 support i think it's a very important test to do.
I bought the ddr4 version so I could just keep my ddr4 3200 cl16. Because ddr5 is and was insane. Glad to see that's pretty much the budget sweet spot overall.
Looks like Steve has half of Australias ddr5 supply
Also for any cpu and memory testing it would be really nice to see more esports titles at low res and details, this is area where cpu and gpu matter the most and you want to pump as many frames. The games you tested there are irrelevant for this kind of testing, except for rainbow 6
@@wojtek-33 not even close. In such games and scenarios memory makes a much larger difference
A bit of a missed opportunity was to test a lower end model with only the igpu. That will be the combo that will end up in gazillion laptops.
thanks for this efficient comparison.
Non k CPUs also have stability problems with ddr4 3600 or higher gear 1, due to locked SA voltage.
Can’t you just buy a kit of 4000 cl19 vipers for like $100usd...? Kinda wild there wasn’t any B-die in here
Yeah these tests are not great because the memory isn’t manually tuned, which makes a huge difference with Alder Lake. On my 12900k at 3800Mhz C15 with the Aida64 benchmark I get 45ns of latency, whereas the best hardware unboxed could do was like 56ns. At 4000Mhz c15 I can push 43.5ns, but I wasn’t stable with my kit (every single timing tuned, from primary to tertiary to turn around)
These XMP kits were bottom of the barrel as another poster said. Not trying to promote anything but IMHO another RUclipsr called FrameChasers did a way better maxed out DDR4 vs DDR5 gaming comparison (only gaming tho). The DDR4 kit was 4000C14 with tweaked subtimings and I can't remember what the DDR5 was, but also tightly tuned.
@@xblur17 Very nice mate, I'm waiting for my 12700k DDR4 build to arrive. I'd be quite happy with 3800C15 and C14 would be a dream haha. Which kit are kit using bro? I'm thinking of pulling the trigger on a TridentZ Neo 2x16GB Dual Rank 3600 14-15-15-35 1.45v, going at $270 right now. Alternatively I can save a bit and get Corsaor Dominator Platinum RGB (DR) 3600 14-16-16-36 1.45v for $215, but from what I've read on forums, I don't have too much faith in current Corsair's B-Die kit, it seems that most people wishing to tune & OC their kit opt for GSkill.
@@Ben-ld1qi TridentZ Neo 2 x 8gb, incredible kit! Manually tuned b-die and a manually OC’d processor do so much for performance, especially in competitive game titles. Best of luck with your build man, it’s gonna be great!
@@Ben-ld1qi Aye ben, we’re from the same community lmfaooo I just rewatched Jufes’ 10900k mem scaling bench with 4000ringbus then this video dropped😂
Built a new system over the holidays, haven't built one since 2012. Went with the 12700k and 64gb of DDR4 3600. Mobo's are also more affordable if you go with the DDR4. Don't get me wrong, I wanted the ASUS formula for my all white build, but the ROG Strix was just fine.
ROG Strix A?
@@baskapes762 yeah that's the one I went with. Don't get me wrong, the price is still double what a higher end board was 3 years ago.
How does igpu performance change?
I have 3000 MHz with i7 12700k worth upgrade If I used DLSS in a monitor at 1440p I know with DLSS goes to rescaling at 969p something l played Marvel Guardians of the Galaxy and I notice using DLSS with Ray tracyng OFF GPU its at 70 % with Rtx 3070 ti but in other games like cyberpunk always had 99% GPU usage even using DLSS so can somebody explain me if its the game or its the system I didnt see no one explain some games arent optimized with DLSS ? Hope can help me
May i just ask you something?
So i have a b350 tomahawk and 1600x for about 4.5 years now, i wanted to upgrade to ryzen 5600x or the 5800x, but obviously it wasnt supported, so the only real choices were 3600 and 3700x. Until a few weeks ago when there have apparently been some bios updates that unnoficially add support for ryzen 5000 cpus. Could you verify and make a video about it? Its kind of a big deal but don't see anyone covering it. Cheers!
PSA! Alder Lake non K CPUs have locked voltage control (VCCSA) for Memory Controller (IMC) to 0.89-98v.
If you are going for anything above 3200Mhz Gear 1 or 5400Mhz Gear 2 on non K processor, you may be unable to run at those speeds, it's a lottery!
This should had been mentioned in the review, my 12700F which I returned was not able to run 3600Mhz CL18, best value RAM.
I got 12700K instead and even on b660 the VCCSA is unlocked, so the CPU are locked!
Hmm it would be really good to know if the PCIe bus/RAM interface is the limitation for the CPU in some of these cases. DDR5 offering a lot more bandwidth might be allowing the CPU to run code better while DMA is transferring textures to/from graphics cards.
GPUs don't access RAM directly, they have to go through the CPU.
Help me please . I just buy a onn monitor for my son is a 1080p resolution . The problem is when i connect with a display port the monitor showing 1080p but when i plus my hdmi cable showing me 4k resolution any ideas please?
Would be great to see integrated GPU performance using DDR4 vs DDR5.
A friend bought the 13900k recently. DDR5, 1/3th of the performance of the RX 570, at least in Valheim.
Hi, are you planning to do a B660 mainboard test like you did with the B560 series boards? I‘m kind of afraid to buy anything because B560 was such a mess. Thanks!
3600 CL16 is the best option rn
This is a dumb question but my google fu failed me. My asus mobo didn't have a gear mode choice that I could see or notice. Z690 tuf d4. Where in the bios is the gear mode option? I have seen gear mode clearly labeled in a friend's ryzen uefi so I was wondering if it will be called something else on an intel mobo?
So now the question is what DDR4 board to buy? What would you pair with a Core i7 12700k that was bought on sale for $299.99?
Anyone know the aio name at 13:18 ?
I'm going to wait to upgrade from my current system until DDR5 is not so ridiculously over-priced relative to the performance.
I will be tweaking my current DDR4 installation for more memory and better latency.
Man, so many questions. If you had some 3200/14 memory do you think it would have been the best? Have you ever lowered a ram's speed to get better timings? Can you do a similar test across all Ryzen generations? 😅
"Have you ever lowered a ram's speed to get better timings?"... Uh, that doesn't give you better timings. The big problem here is that timings are given in clock ticks and not nanoseconds like they should be. A 6000CL38 kit has the exact same command latency as a 3000CL14 when measured in nanoseconds.
I could not afford a DDR5 motherboard and DDR5 RAM. I have 16GB x 4 with timings 16, 19, 19, 39. Best I could afford with my Z390 motherboard. Thank you for this review.
I'd also wonder, how some DDR4-4000 CL14 Kits would compare against the DDR5-6200 CL36 Kits. Because in my country, the first costs around 530 USD compared to only 540 USD for the DDR5 stuff.
Is DDR5 worth it just for future proofing? Getting some lower tier sticks now, and then once you feel like you need to upgrade in the future when it’s gotten cheaper and better, to get some better sticks without having to change the motherboard as well?
Hardware Unboxed can make an infinite number of video's about Alder Lake and i'll still be waiting a longer for the Ryzen 7 5800X3D
I feel like budget DDR4 used to advertise / run as 3000 rather than the 2933, at least on 16 GB kits.
You guys are awesome. so many info and the way you present it, fantanstic!
I suppose all that data refer also to 16 ( 2x8 ) kits? because I dont see a massive reason to upgrade to 32 for now, and I own a 2x8 3600 cl16 ballistix set.
@@jamesdoe7605 single rank?
Something looks wonky with the Hitman 3 charts. The 1440p chart is visually what I expected based on the other games, but the FPS measurements are higher than 1080p.