RAM freq actually matters?! a DigitalFoundry test

[zMeul]

source:

 

 

---

 

and here is the interesting bit:

 

 

there is quite a noticeable difference between 2133Mhz RAM and 2666Mhz

they also ran tests with the older i3:

 

 

maybe Linus / Luke would take another crack at RAM frequency influence in CPU bound scenarios

[Lays]

Why wouldn't it matter in CPU bound scenarios? This has been a thing for quite some time now.

Anything that can make the CPU faster at doing whatever it's doing helps in CPU bound scenarios.

 

 

Also IIRC this is a repost, somebody already made a big thread about this stuff like last week or so, will see if I can find it

[PlayStation 2]

In CPU bound scenarios, it can matter and with things with HUGE CPU usage (video/3D rendering specifically) it makes a big ass difference

[zMeul]

Also IIRC this is a repost, somebody already made a big thread about this stuff like last week or so, will see if I can find it

last week!? this was just posted by Eurogamer

[Lays]

last week!? this was just posted by Eurogamer

 

 

 

http://linustechtips.com/main/topic/474420-so-skylake-i3-is-the-new-budget-gamer-king-mem-speed/

[zMeul]

http://linustechtips.com/main/topic/474420-so-skylake-i3-is-the-new-budget-gamer-king-mem-speed/

it's a different video, but basically on the same theme ... I missed it 

[deviant88]

no this is bs ram doesnt give such insane performance benefits from 1600 to 2133 or 2133 to 2666, the i3 probably has glitches or worse memory controllers that makes it seem it performs better with higher freq ram, any i5 i7 or extreme cpu show no improvement in gaming with better ram and dedicated gpus only the onboard graphics benefits as always

 

in the video 3:16 with ryse son of rome from 40 fps 2133 to 100+fps on 2666 THAT IS BULLCRAP stupid reviewer thats a clear bug/glitch in hardware with i3 since it shows signs in the other games too, if its true that it gets such benefits its because of the glitch and not because of 533 extra ram mhz

or its because of the imbalance of an i3 with titanx 

if it werent an i3 glitch then why i5 and i7 skylake get no benefits at all from 2133 to 2666

[MageTank]

no this is bs ram doesnt give such insane performance benefits from 1600 to 2133 or 2133 to 2666, the i3 probably has glitches or worse memory controllers that makes it seem it performs better with higher freq ram, any i5 i7 or extreme cpu show no improvement in gaming with better ram and dedicated gpus only the onboard graphics benefits as always

 

in the video 3:16 with ryse son of rome from 40 fps 2133 to 100+fps on 2666 THAT IS BULLCRAP stupid reviewer thats a clear bug/glitch in hardware with i3 since it shows signs in the other games too, if its true that it gets such benefits its because of the glitch and not because of 533 extra ram mhz

or its because of the imbalance of an i3 with titanx 

if it werent an i3 glitch then why i5 and i7 skylake get no benefits at all from 2133 to 2666

This is only in situations with very high CPU overhead. You will not notice this on fast CPU's, or heavily optimized games.

 

This is not some recent development. This has been common knowledge for a few years now. http://www.overclock.net/t/1487162/an-independent-study-does-the-speed-of-ram-directly-affect-fps-during-high-cpu-overhead-scenarios (Thread is over a year old, and its based on claims that were made long before its existence too)

 

While this might be news for some people, a lot of us that use slower CPU's (Pentiums, i3's, etc) we have known this fact for quite some time.

[SteveGrabowski0]

last week!? this was just posted by Eurogamer

 

They posted the video with the results last week too. I was surprised to see them reprise it, but I guess I can understand since it goes so strongly against what most gamers have believed for years.

[ICantThinkOfAnyGoodName]

no this is bs ram doesnt give such insane performance benefits from 1600 to 2133 or 2133 to 2666, the i3 probably has glitches or worse memory controllers that makes it seem it performs better with higher freq ram, any i5 i7 or extreme cpu show no improvement in gaming with better ram and dedicated gpus only the onboard graphics benefits as always

in the video 3:16 with ryse son of rome from 40 fps 2133 to 100+fps on 2666 THAT IS BULLCRAP stupid reviewer thats a clear bug/glitch in hardware with i3 since it shows signs in the other games too, if its true that it gets such benefits its because of the glitch and not because of 533 extra ram mhz

or its because of the imbalance of an i3 with titanx

if it werent an i3 glitch then why i5 and i7 skylake get no benefits at all from 2133 to 2666

Iirc every intel processor of the same architecture and socket is built exactly the same, just that all of them except for the i7s have some cache and either hyperthreading or cores disabled, so the memory controller is the same on all of them.

[Sauron]

the money spend for the faster ram could have bought you an i5. consider that.

[MageTank]

the money spend for the faster ram could have bought you an i5. consider that.

Faster memory is pretty cheap though. The difference between 8gb of 2133mhz and 2800mhz is roughly $10-$15 on average. The issue would be getting a board that can run them at those speeds. The difference between a Z and H board can be hefty depending on features and form factor. That being said, i would rather someone invest in a better board + ram, with a cheaper CPU, than the other way around.

 

That, and one can still just overclock any kit of memory they get. Plenty of headroom going from 1.2V to 1.35V.

[Sauron]

That being said, i would rather someone invest in a better board + ram, with a cheaper CPU, than the other way around.

 

as long as the motherboard is decent it won't give you any problems, the only downsides are less features - if you're on a budget a better cpu is definitely a better idea. Also consider ram latencies - the difference for a higher frequency kit may not be high, but if the latencies are the same the prices climb quickly. I'd also like to point out that crysis 3 and especially ac unity are not good samples to represent the majority of games.

[crystal6tak]

Yea I've seen that video a while ago, I'm still unsure, is this just a thing for skylake i3's?

 

Techbuyersguru made a test a year back and found miniscule difference jumping from 1600->2400

Heck Linus himself did a video 2 years ago and found absolutely no difference. At least for average fps.

 

This is only in situations with very high CPU overhead. You will not notice this on fast CPU's, or heavily optimized games.

 

This is not some recent development. This has been common knowledge for a few years now. http://www.overclock.net/t/1487162/an-independent-study-does-the-speed-of-ram-directly-affect-fps-during-high-cpu-overhead-scenarios (Thread is over a year old, and its based on claims that were made long before its existence too)

 

While this might be news for some people, a lot of us that use slower CPU's (Pentiums, i3's, etc) we have known this fact for quite some time.

What CPU was the BradleyW guy using? I can't seem to find it.

[MageTank]

as long as the motherboard is decent it won't give you any problems, the only downsides are less features - if you're on a budget a better cpu is definitely a better idea. Also consider ram latencies - the difference for a higher frequency kit may not be high, but if the latencies are the same the prices climb quickly. I'd also like to point out that crysis 3 and especially ac unity are not good samples to represent the majority of games.

From what I (and others at OCN) have tested, Cas Latency seems to have little to no impact on FPS. So far, clock has been king in these tests. That's not to say one should ignore Cas latency, as we all know its best to have a good balance between the two. That being said, The cheapest 8gb kit of DDR4 i could find, was $46 ($36 after rebate) and the cheapest 8gb 3200mhz kit i could find, was $50. 

 

http://pcpartpicker.com/part/patriot-memory-psd48g21332h

 

http://pcpartpicker.com/part/mushkin-memory-992202t

 

When on a budget, one is likely to choose the cheapest options everywhere, including board quality. Those locked H boards will not support the higher clocked memory. Investing in a more expensive board to enable memory overclocking would be ideal, assuming someone understands the headache and math of memory overclocking.

 

 

Yea I've seen that video a while ago, I'm still unsure, is this just a thing for skylake i3's?

 

Techbuyersguru made a test a year back and found miniscule difference jumping from 1600->2400

Heck Linus himself did a video 2 years ago and found absolutely no difference. At least for average fps.

 

What CPU was the BradleyW guy using? I can't seem to find it.

He was using a 3930k clocked to 4.5ghz. 

[SteveGrabowski0]

From what I (and others at OCN) have tested, Cas Latency seems to have little to no impact on FPS. So far, clock has been king in these tests. That's not to say one should ignore Cas latency, as we all know its best to have a good balance between the two. That being said, The cheapest 8gb kit of DDR4 i could find, was $46 ($36 after rebate) and the cheapest 8gb 3200mhz kit i could find, was $50. 

 

http://pcpartpicker.com/part/patriot-memory-psd48g21332h

 

http://pcpartpicker.com/part/mushkin-memory-992202t

 

When on a budget, one is likely to choose the cheapest options everywhere, including board quality. Those locked H boards will not support the higher clocked memory. Investing in a more expensive board to enable memory overclocking would be ideal, assuming someone understands the headache and math of memory overclocking.

 

 

He was using a 3930k clocked to 4.5ghz. 

 

That's possibly also because the actual latency (in terms of time) associated with the column strobe is the CL number of clock cycles divided by the clockspeed, so RAM with faster clockspeeds usually has similar or even lower latency than slower kits with lower CL numbers.

 

E.g., the latency for the column strobe associated with a CL 9 DDR3-1600 kit would be  (9 cycle)/(800 MHz) = 11.25 nanosecond  (1 MHz = 10^6 cycle/second), while for a CAS 13 DDR3-2400 kit it would be (13 cycle)/(1200 MHz) = 10.83 nanosecond.

[Tomsen]

Iirc every intel processor of the same architecture and socket is built exactly the same, just that all of them except for the i7s have some cache and either hyperthreading or cores disabled, so the memory controller is the same on all of them.

Yes, most Intel consumer processors comes from the same wafer.

Doesn't mean they can't disable certain parts or "nerf" it for segmentation.

[MageTank]

That's also because the actual latency (in terms of time) associated with the column strobe is the CL number of clock cycles divided by the clockspeed, so RAM with faster clockspeeds usually has similar or even lower latency than slower kits with lower CL numbers.

 

E.g., the latency for the column strobe associated with a CL 9 DDR3-1600 kit would be  (9 cycle)/(800 MHz) = 11.25 nanosecond  (1 MHz = 10^6 cycle/second), while for a CAS 13 DDR3-2400 kit it would be (13 cycle)/(1200 MHz) = 10.83 nanosecond.

Exactly. People often complain about the DDR4 latency, but they forget what it is rated for. It can support up to 4266mhz. We already have 4266 CL19 kits available to us. http://pcpartpicker.com/part/gskill-memory-f44266c19d8gtz I would NEVER pay that price for such a kit, but just saying. We have reached that number already.

[Guest Kloaked]

http://linustechtips.com/main/topic/474420-so-skylake-i3-is-the-new-budget-gamer-king-mem-speed/

 

Right, so can this thread be rewritten to match this information, @zMeul? Don't want confusion or misinformation going around. You should have did your research on this to begin with.

[Doobeedoo]

I've seen this, wasn't surprised much though.

[Prysin]

snip

If luke and linus is gonna include FX, i hope they use a proper board that supports 2133 DDR3 and doesnt just run off sprouting bullshit about lower end AMD boards not supporting more then 1600...

 

 

Exactly. People often complain about the DDR4 latency, but they forget what it is rated for. It can support up to 4266mhz. We already have 4266 CL19 kits available to us. http://pcpartpicker.com/part/gskill-memory-f44266c19d8gtz I would NEVER pay that price for such a kit, but just saying. We have reached that number already.

well, i will not argue with you on DDR4 being rated for higher speeds. and Patrick may be right that internal latencies has been reduced (as usual i couldnt find any sources even after an hour of googling)

 

however, around the 2400MHz mark, DDR3 with way lower latency is still cheaper, AND FASTER, because its like CL10 vs CL 15 or at best CL 13

[MageTank]

If luke and linus is gonna include FX, i hope they use a proper board that supports 2133 DDR3 and doesnt just run off sprouting bullshit about lower end AMD boards not supporting more then 1600...

 

 

well, i will not argue with you on DDR4 being rated for higher speeds. and Patrick may be right that internal latencies has been reduced (as usual i couldnt find any sources even after an hour of googling)

 

however, around the 2400MHz mark, DDR3 with way lower latency is still cheaper, AND FASTER, because its like CL10 vs CL 15 or at best CL 13

Yup, and density plays a large role too. I have a buddy that has a 2x4gb kit of Samsung memory that is 2400mhz CL9. Granted, it does not come this way out of the box (It's normally 1600mhz CL9) but it overclocks so well. 

 

DDR4 is getting cheaper thanks to Skylake. I also believe Zen will help push the prices down even further once the AMD fans get their champion platform to rally behind. 

 

That being said, at least with the topic at hand, latency does not mean that much for gaming. At least from what i could tell with my current tests on DDR3 (My Z170 board and DDR4 memory will arrive tomorrow, so i will conduct further tests once those arrive, using a G4400). It just seems that games only care about high memory clock speeds. So even a 2800 C15 kit might still out perform a 2400 C12 kit in gaming. I won't say for absolute certainty yet, until i can prove it with my own testing. Just something to keep in mind from what we are seeing so far with that old OCN thread, and that new i3 video.

[Prysin]

Yup, and density plays a large role too. I have a buddy that has a 2x4gb kit of Samsung memory that is 2400mhz CL9. Granted, it does not come this way out of the box (It's normally 1600mhz CL9) but it overclocks so well. 

 

DDR4 is getting cheaper thanks to Skylake. I also believe Zen will help push the prices down even further once the AMD fans get their champion platform to rally behind. 

 

That being said, at least with the topic at hand, latency does not mean that much for gaming. At least from what i could tell with my current tests on DDR3 (My Z170 board and DDR4 memory will arrive tomorrow, so i will conduct further tests once those arrive, using a G4400). It just seems that games only care about high memory clock speeds. So even a 2800 C15 kit might still out perform a 2400 C12 kit in gaming. I won't say for absolute certainty yet, until i can prove it with my own testing. Just something to keep in mind from what we are seeing so far with that old OCN thread, and that new i3 video.

yes, the benchmarks in the OP does show little improvement. BUT!

 

here is a great example... (i will stick to AMD for this, because they got CHEAP boards supporting the speed)

 

PCPartPicker part list / Price breakdown by merchant

CPU: AMD FX-6300 3.5GHz 6-Core Processor  ($97.88 @ OutletPC)

Motherboard: ASRock 970M PRO3 Micro ATX AM3+/AM3 Motherboard  ($63.89 @ OutletPC)

Memory: Silicon Power Xpower 8GB (2 x 4GB) DDR3-2400 Memory  ($41.17 @ Amazon)

Memory: PNY Anarchy 8GB (2 x 4GB) DDR3-1600 Memory  ($33.99 @ Newegg)

Total: $236.93

Prices include shipping, taxes, and discounts when available

Generated by PCPartPicker 2015-11-05 17:07 EST-0500

 

here you see the cheapest 1600 RAM and cheapest 2400 RAM i could find...

CL 9 and CL 11 respectively... while you can get a 2400MHz CL 10 kit for 5 bucks more hten the CL 11 one, i decided to go "cheap ass vs cheap ass" for a better comparison.

 

we see that going from 1600 to 2133 gives roughly 5 FPS... so going from 1600 to 2400 should net you AT LEAST 6 FPS... maybe 7... for 8 USD more...

 

imagine how much you pay pr FPS when it comes to GPUs and CPUs.... paying 8 bucks more for faster RAM, if the motherboard supports it, may have some merits.

 

i also concur about DDR4 prices. However at the lower end, DDR3 will likely remain king for a while, as they got way way way lower latencies... so until DDR4 hits around CL 12-13 as the "average" latency, it will be "too slow" to be really worth it over DDR3 options.

[Dabombinable]

This is only in situations with very high CPU overhead. You will not notice this on fast CPU's, or heavily optimized games.

 

This is not some recent development. This has been common knowledge for a few years now. http://www.overclock.net/t/1487162/an-independent-study-does-the-speed-of-ram-directly-affect-fps-during-high-cpu-overhead-scenarios (Thread is over a year old, and its based on claims that were made long before its existence too)

 

While this might be news for some people, a lot of us that use slower CPU's (Pentiums, i3's, etc) we have known this fact for quite some time.

I know for a fact that increasing the frequency of the SDRAM that was being used with my Celeron 500 (increased independent of the CPU that is since no overclocks were stable with that CPU) leads to massive performance boosts (it was going from 66MHz to 150MHz). And that was before the turn of the millennium.

[Toddwjp]

im surprised digital foundry stuff isnt on here more, they do some great analysis