DDR4 Vs DDR3

[nicklmg]

ADATA DDR4 XPG Z1 16GB Kit

NCIX: http://bit.ly/1r0p0nX

Amazon: http://geni.us/3qVE

 

ADATA DDR3 XPG V1 16GB Kit

NCIX: http://bit.ly/1oZse7V

Amazon: http://geni.us/3Hqt

 

DDR4 is a pretty hot discussion topic right now, and we became curious... can you expect to see a tangible performance benefit from running DDR4 compared to DDR3, keeping all other variables as close to the same as possible?

 

[Jvagle875]

I didn't see any apples in this comparison. Linus was paid off.

 

Also, why didn't you use an x79 platform that way you could use quad channel in both?

[Deletive]

I really want now x99

Edit: WOAH 5820k is huge compared to a 4790k.

[Jobbe03]

No way near worth it a 2x the price

[-TesseracT-]

Why isn't this video in my sub box?

[alby800]

Perfect timing! Just got off "work". Thanks Nick

Why isn't this video in my sub box?

I think it might be unlisted.

[-TesseracT-]

Perfect timing! Just got off "work". Thanks Nick

I think it might be unlisted.

[Deletive]

I don't understand the ending of this video... He says improvements from broadwell, which is nothing much more really then just a die shrink and igpu improvement with ~5% ipc. At the end I can't tell if he was implying that broadwell was coming in ddr4 or not; I think he might be mistaken with skylake which would really be the first ddr4 consumer platform. (depending on carrizo)

[Solkia]

Huh. JayzTwoCents made the comparisons sound more impressive.

[Geekazoid]

@nicklmg, hello to Linus' official YouTube Video Forum Poster? Has Linus given you this job?

[Blarg]

Plz never so many benchmarks on one page plz god plz k tank~~my brain

[XTankSlayerX]

Why didn't you test Battlefield 4?

 

You said it would be very interesting when DDR4 would come along.

[Lord Byoss]

DDR4 is just no worth the price at the moment, but will a price drop, it will be well worth it.

[GaryC]

Plz never so many benchmarks on one page plz god plz k tank~~my brain

You can just pause the video, you know.

[Narussistic]

Hey, Linus. Why did you go for 4790k instead off something like 4930k?

[SKULLCRU5H3R]

I expected those results.... it's always like that when a new memory spec begins to be implemented, it will not be until at least 18 months that we will see some gains in something that is not benchmarking, what will really benefit of DDR4 memory at high frequencies will be an APU, since the IGP scales better as memory frequency goes up.

[Stefano]

Was this recorded at 60 fps and then played back at 24? Because Linus was talking extremely slowly...

[Faa]

I couldn't create a situation where gaming performance was limited in any meaningful way etc

The graphs were showing the GPU's performance variation, how can we expect a difference from "better ram" when the GPU is the bottleneck? Turn the logic around; when you're limited by the CPU you're not going to push more out of a better GPU if you can't even push your current GPU to half of its maximum load. There's no way to get more performance from RAM when the gpu is flacking at its maximum load other than clocking the GPU or getting a new GPU. It needs to be tested when the CPU is the bottleneck, not the GPU. Like changing the resolution to 720p, or add a 2nd card, or even a 3rd in there.


I've done some testings a while ago, although with the same timings for each clock just to see how much performance you can get from a bump in frequency. I kept SLI disabled because there's no gain as the CPU bottleneck was very obvious. http://imgur.com/a/7NYNK (800MHz-1600MHz 25% gain)

Anyways the gains in games aren't usually there, I've seen here and there some 10-15% benchies but that's all about it.

I dont get why Linus didn't include a test between DDR4 & DDR3 at the same timings & frequency after calling the video apples to apples and equaling the corecount & speed.

 

[paulsterio]

I think most of us would have expected these results, given the overwhelming evidence that faster RAM speeds do not really yield higher performance in most real-world applications including compute style calculations, gaming and video rendering. Yes, I know there were benchmarks from Corsair showing BF4 loving high speed RAM, but would you really trust a RAM manufacturer's benchmarks?

 

I personally think that memory has stagnated for the past few years. I remember when I built my i7 860 all the way back in 2009, I had 8 GB of RAM, which was pretty much the new standard at the time. Most people on Core 2 Duos and Core 2 Quads were still on 4GB RAM, so 8GB for the new platform made sense. I soon upgraded to a 3570K, retaining 8GB of RAM and I'm now on a 3930K with 16GB of RAM and to be honest, that's because I just got another 8GB kit for quad channel. I don't use that RAM at all, even in video editing. Yes, it does have a slight performance benefit, but I regularly render quite large files and I would say that the difference, whilst not negligible, is definitely not worth spending whatever the cost is to double your RAM and that you would get much better performance upgrading your CPU or your GPU. 

 

Hence, what I really see RAM as is an enthusiast's playground. I've always built systems with the cheapest RAM I could find, any 1600 MHz DDR3 kit would do and I have never had any issues whatsoever. Sure, it doesn't look the best, but if you're after performance and bang-for-buck, are you really going to spend $50 on getting Corsair Vengeance RAM for no performance boost? I certainly wouldn't and I'd rather spend that $50 upgrading to a larger SSD or getting a slightly better GPU, e.g. 760 -> 770 is around a $60-$70 difference. Even RAM capacity beyond 16GB has been pretty overkill for a while now.

 

Whilst most people are willing to spend up for a 16GB kit, how many of you actually use all your 16GB? I would say that the mainstream amount these days is still 8GB. In 2009, it was also 8GB. That was 5 years ago. Guess what the average RAM capacity was in 2004, 5 years prior to 2009? My pretty state-of-the-art Pentium 4 had a whopping 512 MB or RAM. Yeah, from 2004 to 2009, the standard amount of RAM increased x16 folds (0.5GB -> 8GB). In the five years since, it's stayed the same, even if you wanted to say the average is now 16GB, still a lackluster x2 fold. 

 

Seems to reflect the state of PC technology at the moment anyway. 3 years ago, Sandy Bridge was released. Take Haswell vs. Sandy Bridge compared to Sandy Bridge vs. Conroe, the architecture 6 years ago, so 3 year period before Sandy Bridge. Yeah, tech moved quickly back then. 

[TheYonderGod]

Linus/team, you should use Battlefield 4 for your gaming benchmarks about memory. I used to be a competitive battlefield player (switched to CS now), so I was looking for any performance boost possible. I went from 8gb 1600 cl9 to 8gb 2400 cl11, and I got a noticeable performance improvement. (I was playing on lowest settings at 1080p and you can see the rest of my specs in my signature)

[MZ Farhad]

I wasn't expecting much from DDR4, in terms of perfprmance. cos RAM was never realy a bottle neck.

quick question, was there a huge difference between DDR2 and DDR3 at launch.

[SwedishHaxMafia]

So once again, memory does not matter and no one cares

[psychostudent]

How about using the extra RAM as a ramdisk?  If you get a 16 or even better a 32gb kit, put 6-8gb or something as system memory and the rest as a ramdisk, whould games load notacibly faster?

 

I've seen other ppl use ramdisk applications to load entire games into RAM and run it from there. from what they say, loading times are so fast that load screens just blink and you're in the game. Sure you have to copy over all the game files into RAM before you can start playing but with an SSD copying a couple of gigs shouldn't take more than a few seconds.

[SweatyWhiteJedi]

I really want to see a side by side comparison of the new highest i7 and the new xeon 18 core.  I know the xeon cores are not for gaming but I still really want to see what would happen with these cores and DDR4. 

 

 

[Peter Yacono]

I agree. JayzTwoCents said that he saw considerable bandwidth improvement over DDR3?

 

Would have been nice to see a 6 core X79 vs 6 core X99 - especially considering dual vs quad memory channels.