Why does the DDR memory speed effect FPS ?

[h4rm0ny]

not sure if done right you have to include cas rating i would believe to .

as im with 2200 at cas 9 faster then 6700k with 2400 or 2600 at cas 14 or 16 , or even more can't remember need to check benchmarks

They did do some rudimentary testing of different CAS values. They found that latency took a fairly distant backseat to the overall speed with the latter compensating several times over for latency.

20% performance gain from memory doesnt seem real in my brain
surprising!

was it sponsored by kingston or corsair or something...not that im jaded.....

Apparently the spur for this article was Asrock emailing them to say their new motherboard was the first to support 4333 (G.Skill has just produced some). The Techspot author didn't think it would make much practical difference but said what the Hell, let's try it. Turns out for the first time in ages, RAM speed is actually starting to make a significant difference in some cases. They also thought that maybe the memory controller on Skylake was a lot better maybe.

You're still better off spending your money on powerful GPU and CPU until you've pretty much exhausted those options, but it's nice to see that RAM is becoming relevant. Occasionally bottlenecks do shift around! But if you have the set up for it (esp. SLi, I would guess), looks like a good buy.

 

[VincentHanna]

Never could get my PC to run using XMP RAM profile/ timings.

I bought 3000mhz DDR4 when I got my X99 system but 2666mhz is a out the fastest I've gotten it to run, stable

 

[chaparral]

Turns out for the first time in ages, RAM speed is actually starting to make a significant difference in some cases. .

Wonder if it anything to do with games now days are using more & more memory ...

And it also be interesting to see what the results would be at different Resolutions like 1080p vs 4k to see if the DDR speeds also effect Vram usage amount

 

[Mei]

They did do some rudimentary testing of different CAS values. They found that latency took a fairly distant backseat to the overall speed with the latter compensating several times over for latency.



Apparently the spur for this article was Asrock emailing them to say their new motherboard was the first to support 4333 (G.Skill has just produced some). The Techspot author didn't think it would make much practical difference but said what the Hell, let's try it. Turns out for the first time in ages, RAM speed is actually starting to make a significant difference in some cases. They also thought that maybe the memory controller on Skylake was a lot better maybe.

You're still better off spending your money on powerful GPU and CPU until you've pretty much exhausted those options, but it's nice to see that RAM is becoming relevant. Occasionally bottlenecks do shift around! But if you have the set up for it (esp. SLi, I would guess), looks like a good buy.

why is sli mentioned so much, is it something it needs or it just the fps count? like would one card with details lower produce the same?
hmhm

 

[h4rm0ny]

Wonder if it anything to do with games now days are using more & more memory ...

More likely it is to do with games using the CPU more and more. This isn't to do with RAM capacity because games seldom if ever need 8GB of system RAM and certainly don't need 16GB (there maybe be random and badly coded exceptions out there). It's to do with RAM speed - which is a different thing altogether. A CPU needs to be kept fed with RAM and the more the CPU is doing, then typically the greater its appetite for fast memory. Think of it as a diner in a restaurant and the RAM speed as the waiters.

DX12 is going to make this situation even more noticeable as it removes (or at least allows to remove) a big bottleneck between the CPU and the GPU. Think of the GPU as the person clearing away the finished plates in the restaurant. Historically they've piled up but now that they're actually being removed faster, the diner (the CPU) can get through more plates faster so wants them brought to the table faster.

And it also be interesting to see what the results would be at different Resolutions like 1080p vs 4k to see if the DDR speeds also effect Vram usage amount

Yep. It would.

why is sli mentioned so much, is it something it needs or it just the fps count? like would one card with details lower produce the same?
hmhm

Presumably it must mean that the CPU needs to do more work for SLi setups than none. This has usually been the case. For example, an AMD FX- chip is usually fine for single GPU set-ups, but starts to struggle with high numbers of cards in SLi / CF because of the increased CPU load.

 

[Mei]

Presumably it must mean that the CPU needs to do more work for SLi setups than none. This has usually been the case. For example, an AMD FX- chip is usually fine for single GPU set-ups, but starts to struggle with high numbers of cards in SLi / CF because of the increased CPU load.

but cpu load is dependant on the amount of frames its pushing not the sparkly effects in them frames, 100fps at whatever res/detail level should be the same cpu usage?
i was wondering if sli relies on system memory speed more than a single card would *shrug* ...or they are just trying to confuse things with sli...

 

[Holst1981]

Is this also true for X99 with its quad chanel memory?
Or only important for skylake?

 

[setter]

I remember a Corsair promotion for its 2400mhz Vengeance pro ram when Bf4 released where it showed there 2400 ram providing a good 5+ frames in Bf4 over 1600mhz ram.

I imagine it's also part of why the Skylake i5 & i7 has such a good fps advantage over the Devil's canyon i5 & i7 in some games.

I remember that article, and tbh I found no difference whatsoever with ram at default of 1333mhz vs its XMP profile of 2400mhz. Sli 780 setup with a 4790k.

 

[flopper]

One reason we want Dx12 to be more common.
you want a balanced load between the cpu/gpu
and note, that test shows with sli, single cards dont show such difference.

 

[Brazo]

How does the cl (timing?) affect things? I have ddr 3 2400 with a cl of 10. How does that say compare to faster ddr4 ram with a higher cl number?

 

[psychas]

got same bandwidth in sisoft as ddr4 3000 in this article and have latency probably lower then ddr4 3000 .

edit:

How does the cl (timing?) affect things? I have ddr 3 2400 with a cl of 10. How does that say compare to faster ddr4 ram with a higher cl number?

your probably somewhere as me around ddr4 3000 mhz speeds , check benchmarks to be more precise

 

[kylew_is_back]

from my understanding:

faster ram helps the cpu to execute stuff faster, essentially the cpu becomes more speedy in some tasks. therefore in games that are cpu bound a faster cpu will result more fps.

 

[varkanoid]

Digital Foundry's testing suggests otherwise. Their video shows a notable performance boost going from 2133MHz to 3200MHz with a single overclocked Titan X.

I tested games with 1600mhz, 2133mhz and 2400mhz DDR3 and got no difference. Benchmarking there was an advantage. So clarification needs to be done its just not a sweeping statement but it can occur in different situations with different scenarios. ie the video uses DDR4.

 

[Aretak]

I tested games with 1600mhz, 2133mhz and 2400mhz DDR3 and got no difference. Benchmarking there was an advantage. So clarification needs to be done its just not a sweeping statement but it can occur in different situations with different scenarios. ie the video uses DDR4.

8Pack has said that Skylake, specifically, enjoys higher memory bandwidth in many workloads, so it might explain why you didn't see much difference. Even Haswell-E apparently doesn't see the same kind of gains that Skylake does with faster memory, despite also using DDR4. Whether that's just a peculiarity of the Skylake architecture or down to something else (like the fact it has a better memory controller than previous chips) is up for debate I guess. Broadwell-E also looks like it'll have a much stronger IMC than Haswell-E, so testing how that benefits from faster memory when it arrives should shed some light on the subject.