People still wasting money, being fooled by ram speeds

[UsandThem]

I tried using DDR2 in my Z270 motherboard to test the OP's hypothesis, but it wouldn't fit. I think I will just need to use more force to get it in there.

 

[BonzaiDuck]

I personally prefer to stay within official specs when running expensive hardware or even not so expensive hardware. As I may not be able to replace it if it dies early.

There are "official specs" and there are board and RAM specs. I can only speak about a chipset of a generation and a half ago -- the Z170. But IIRC, the "official" speed of RAM for that generation of boards was 2133 Mhz of DDR4. The board was capable in its published specs for RAM running at 3200 Mhz or higher.

You then had the choice of a wide range of speeds and varying timings. I did my perfunctory research and decided that I wanted DDR4-3200 -- not 2133 and not 3600 or 4000. It seems to me that the last several generations of G.SKILL RAM I could find provided the tightest spec timings at the rated, spec speed, although there might be other makes of RAM meeting those specs. So I bought DDR4-3200 CAS 14, when G.SKILL offers the same model-class or "model line" in kits with CAS 16.

For 16 GB in 2x8GB kits, I didn't feel "cheated." At the time, such a kit cost me $130. Then, I think the price climbed as high as $185 when I decided to fill four slots rather than upgrade to 2x16GB. The price factor was significant in my choice there.

The only difference was the likelihood of running command rate = 1 versus 2. These days, that may be the limit of tweaking that I will pursue, since you can pretty much buy the spec you want and speed you want for a few dollars more or less.

And if you're going to build a new system to replace your DDR2 or DDR3 rig, you're only going to get DDR4 because that's all the boards support.

 

[whm1974]

There are "official specs" and there are board and RAM specs. I can only speak about a chipset of a generation and a half ago -- the Z170. But IIRC, the "official" speed of RAM for that generation of boards was 2133 Mhz of DDR4. The board was capable in its published specs for RAM running at 3200 Mhz or higher.

You then had the choice of a wide range of speeds and varying timings. I did my perfunctory research and decided that I wanted DDR4-3200 -- not 2133 and not 3600 or 4000. It seems to me that the last several generations of G.SKILL RAM I could find provided the tightest spec timings at the rated, spec speed, although there might be other makes of RAM meeting those specs. So I bought DDR4-3200 CAS 14, when G.SKILL offers the same model-class or "model line" in kits with CAS 16.

For 16 GB in 2x8GB kits, I didn't feel "cheated." At the time, such a kit cost me $130. Then, I think the price climbed as high as $185 when I decided to fill four slots rather than upgrade to 2x16GB. The price factor was significant in my choice there.

The only difference was the likelihood of running command rate = 1 versus 2. These days, that may be the limit of tweaking that I will pursue, since you can pretty much buy the spec you want and speed you want for a few dollars more or less.

And if you're going to build a new system to replace your DDR2 or DDR3 rig, you're only going to get DDR4 because that's all the boards support.

With the Ryzen APUs/CPUS I would get the ddr4-3200 as the price is still reasonable compared to higher speeds. Of course with the new APUs, they really benefit from it.

 

[SteveGrabowski]

Fallout 4 likes it some fast ram. No matter the platform.

https://www.youtube.com/watch?v=4cL_2rELqng

Fallout 4 really does. I have a system with a Xeon E3-1231v3 and a GTX 970, and going from 2x4GB CAS 11 DDR3-1600 to 2x8GB CAS 11 DDR3-2400 was the difference between reliably getting 48 fps at my worst spot (inside the top floor of the Concord speakeasy where there is a lot of dust floating around) and only dropping to 57 fps in that spot and 60 fps everywhere else except in intense firefights where it could again drop into the mid 50s. This was at 1080p with settings maxed other than godrays and draw distance (I think I had both at medium) and on the version before there were a bunch of physx effects added to shooting walls and such. It was an immediately noticeable difference in the stability of my framerate. I can't imagine it had anything to do with the increased capacity from 8GB to 16GB since I was never anywhere close to 8GB total usage from system + Fallout 4. I don't even think I had broken 6GB usage according to RTSS.

 

[SteveGrabowski]

The strategy of many gamers has been to build a platform that will last for years, and require nothing beyond a vid card upgrade later. A good number of members here are still rocking gen 2-4 Intel or AMD FX based systems. Budgeting for high performance or excellent overclocking memory makes sense, as it will be anchoring the system for years to come.

I suggested back when ram was cheap, that everyone asking how much to put in their new build, move to 16GB as the standard, and there were some arguing against it. "You don't need it. 8GB is fine for your new build." Some of us that have been around long enough even intimated that cheap ram was not a norm, and there was no reason not to buy more. The gamers that took the advice to just buy 8GB done goofed.

And before someone chimes in with how 8 is still good; it is mostly still servicable, but 16 is better. Some games already make use of the extra memory, and various tasks can benefit from the high performance kits.

If you are on a tight budget, it is definitely one of the places you can trim the cost down. But otherwise, pony up for better stuff, and thank yourself later.

I remember when I built my current gaming system in 2014 there were three pieces of PC Master Race religion that were hyped relentlessly at places like Linus Tech Tips and Tom's Hardware:
(1) You only need 8GB for gaming
(2) RAM speed doesn't matter
(3) Anything beyond an i5-4690k was a waste for gaming because hyperthreading isn't used in gaming

Now I find it funny when I see people who bought i5 complaining their cpu stops them from hitting 60 fps in games like Battlefield 1 while people who bought 3770, 3770k, 4770, 4770k, E3-1231v3, 4790, 4790k, etc still report great performance. I knew (3) was BS based on how badly the hyperthreaded dual core i3-4130 at 3.4 GHz stock killed the non hyperthreaded dual core 4.5 GHz Pentium G3258 in Digital Foundry's gaming videos. They'd have a framerate graph and a frametime graph overlaid where they'd show how badly framerate oscillated on the G3258 in games where it was stable on the 4130. So since HT actually was shown to make a difference I figured it made sense to go for a hyperthreaded quadcore E3-1231v3, seeing that the XB1 and PS4 that drive AAA game development absolutely require decent parallelization when they're running eight Jaguar cores clocked at 1.6 GHz or so for cpu.

(2) I actually believed until Digital Foundry shot down that piece of PCMR religion with some tests they published in 2015. I never believed (1) and thankfully when Digital Foundry showed unequivocally that RAM speed does matter in gaming a nice 2x8GB DDR3-2400 kit was only $75 so was a no-brainer, especially when I could move my slow 8GB kit (that I paid around $100 for in 2014) to an HTPC.

 

[DrMrLordX]

DDR3 prices were low in 2015 anyway, so getting DDR3-2400 CAS10 was a no-brainer. If only DDR4 prices were like that today . . .

 

[rdelaporte]

Hey there,

I'm noticing more and more, that when it comes to RAM speeds, people are still spending HUNDREDS and HUNDREDS of dollars per build for a board that supports the latest frequency of ram, sometimes an extra 100 for a DDR x board if the new ram stepping is out..
Can i ask... why? Why do people still seem to class RAM in their System build as JUST as important to have the newest and fastest and most overclockable like it's just as big of a benchmark as the CPU's ability to boost, keep cool and overclock? Or the graphics card? I mean, they've gone as fat as COVERING the little things with massive heatsinks and even fans on them, sometimes spending like 500 dollars on the laest 128gb set of DDR4 10k overclocked to it's maximum stable of 10240MHz etc...

Again, why? I've been following the speeds, ram timings, ram types on and off for years, and i think everyone has been had!

Again to reiterate, people seem to be upgrading their ram, and overclocking it and treating it like it's just as important to keep this stuff up to date and overclocked just as much as the CPU and GPU, which is absolute nonsense.

The biggest difference to usage and gaming, was the dump from SDR to DDR. Double data rate. That's it.
Like DDR literally doubled the throughput of data between itself and the CPU.
And because of this, alot of people naturally think that DDR2 triples the rate, and DDR3 Quadruples it etc, when this simply isn't the case.

I was doing alot of testing back when the switch went from DDR2 to DDR3. I tested a couple of games on my 4gb DDR2 @800Mhz vs my new DDR3 @ 1600 MHz, so naturally you'd think "Well this is gonna give me double the data rate". But it didn't. It barely made any difference at all. Like in some cases i was getting a couple of more frames per second from the DDR3, but then in the same benchmark in the same game, i was getting more FPS on the DDR2!

And the same is happening again today. There's LITERALLY no difference to the end user in speeds or noticible FPS between DDR3 at 1600Mhz, and DDR4 at 4800Mhz overclocked within an inch of it's life.
you MIGHT see a 2-3% difference in SOME small parts of benchmarks, but to the player, himself, playing the game, rendering the model or scene etc, it's ABSOLUTELY 100% and categorically identical.

There's no difference in real world applications between DDR3 1600Mhz and DDR4 4800MHz overclocked, and there CERTAINLY isn't any difference whatsoever between Ram running on the same stepping with just different timings. (like 2 lots of the same DDR 4 and 1 is running at an overclocked rate). All you're doing is putting stress on your components, costing yourself more money in electricity etc, and wasting hundreds of dollars per year, and thousands over your whole career as a system builder.

I know it sounds like i'm shouting lol, i'm not, i love overclocking and stuff, i started off back in the early 2000's when i had an AMD Athlon Thunderbird AXIA 1.3Ghz which i had overclocked on air at 1.71GHz (that was massive back then) and 256MB of SDRam, and a Geforce 2 GTS overclocked on the memory and core up to Grforce Ultra type speeds.

We're being ripped off with the ram.
The fact that everyone has switched to a new motherboard, new CPU's etc in alot of cases to accomodate this magical DDR4 4800MHz that can be overclocked like this is gonna give them some extra magic boost in their gaming etc, is just nonsense.

Has anyone else been following any of this? Like there's barely any tests ANYWHERE of someone testing the real world speeds od DDR vs DDR 2 vs DDR 3 vs DDR 4 and CERTAINLY none doing all types at all common and overclocked speeds and putting all the data out. All we're seeing is like Sisoft or whatever telling us what should happen theoretically, when in reality, there's is hardly any difference between the generations and I'd say NO difference between ram timings in ns or frequency,

if anyone would like to add onto this, or to correct me I'd be delighted to chat.
Thanks!

Isn't it a fact that ram speeds have a significant impact on the ryzen CPUs where the transport layer runs at the speed of the memory?

Sent from my SM-G950U using Tapatalk

 

[Hail The Brain Slug]

https://www.techspot.com/article/1171-ddr4-4000-mhz-performance/

This comparison shows a fairly conclusive performance increase for Skylake in a variety of things from DDR4 2133 to DDR4 4000.

Saying that memory speed is irrelevant is an ignorant opinion. It has always mattered to an extent, but more on the latest microarchitectures than ever.

 

[Insert_Nickname]

Isn't it a fact that ram speeds have a significant impact on the ryzen CPUs where the transport layer runs at the speed of the memory?

That is correct. If you use a Ryzen system primarily for gaming, you'll want the highest speed @ lowest timings you can afford. DDR4-3200 is the sweet spot currently.

 

[Topweasel]

I remember when I built my current gaming system in 2014 there were three pieces of PC Master Race religion that were hyped relentlessly at places like Linus Tech Tips and Tom's Hardware:
(1) You only need 8GB for gaming
(2) RAM speed doesn't matter
(3) Anything beyond an i5-4690k was a waste for gaming because hyperthreading isn't used in gaming

Now I find it funny when I see people who bought i5 complaining their cpu stops them from hitting 60 fps in games like Battlefield 1 while people who bought 3770, 3770k, 4770, 4770k, E3-1231v3, 4790, 4790k, etc still report great performance. I knew (3) was BS based on how badly the hyperthreaded dual core i3-4130 at 3.4 GHz stock killed the non hyperthreaded dual core 4.5 GHz Pentium G3258 in Digital Foundry's gaming videos. They'd have a framerate graph and a frametime graph overlaid where they'd show how badly framerate oscillated on the G3258 in games where it was stable on the 4130. So since HT actually was shown to make a difference I figured it made sense to go for a hyperthreaded quadcore E3-1231v3, seeing that the XB1 and PS4 that drive AAA game development absolutely require decent parallelization when they're running eight Jaguar cores clocked at 1.6 GHz or so for cpu.

(2) I actually believed until Digital Foundry shot down that piece of PCMR religion with some tests they published in 2015. I never believed (1) and thankfully when Digital Foundry showed unequivocally that RAM speed does matter in gaming a nice 2x8GB DDR3-2400 kit was only $75 so was a no-brainer, especially when I could move my slow 8GB kit (that I paid around $100 for in 2014) to an HTPC.

PCMR has had this issue and core count discussions are another one of those that is going on today. It is really hard to get PC people as a whole to look forward. That's not to say everyone needs to get a 8c machine with 32GB of Ram a 1080ti and an NVME drive. But people need to separate what a good machine for right now is from someone looking to build a new machine for the next 3-5 years. But that doesn't happen instead its i5 with no HT and geting a 7600k or 7700k (some even still saying those now) and 8 or 16GB of ram and a cheap SATA III SSD. Now all of them are kinda good enough for right now. But how long will they last and how long will they last in comparison to other options that have nearly the same price.

We really need to get to a point where we as a whole can look at something and say X is a great value for right now, but Y might last you longer for this reason. Without it becoming some flame war.

 

[Excessi0n]

Fallout 4 likes it some fast ram. No matter the platform.

https://www.youtube.com/watch?v=4cL_2rELqng

I actually got better than linear scaling with memory speed on my heavily modded Skyrim install. Too bad my memory can't really go past 3200 MHz, because I'd love to see how far it could scale...

 

[whm1974]

PCMR has had this issue and core count discussions are another one of those that is going on today. It is really hard to get PC people as a whole to look forward. That's not to say everyone needs to get a 8c machine with 32GB of Ram a 1080ti and an NVME drive. But people need to separate what a good machine for right now is from someone looking to build a new machine for the next 3-5 years. But that doesn't happen instead its i5 with no HT and geting a 7600k or 7700k (some even still saying those now) and 8 or 16GB of ram and a cheap SATA III SSD. Now all of them are kinda good enough for right now. But how long will they last and how long will they last in comparison to other options that have nearly the same price.

We really need to get to a point where we as a whole can look at something and say X is a great value for right now, but Y might last you longer for this reason. Without it becoming some flame war.

I always thought that you should get as much hardware for your money as you can. You may end keeping your computer(s) longer then you planned on.

 

[Dasa2]

Overclocking ram on my systems makes more difference than overclocking the cpu in most games and putting both together make for a very nice increase in cpu performance
Is there a point where it becomes bad value? sure but it does help and it is possible to get high speed ram that's good value (much better value for gaming than a i7 over a i5)
The kits below were samsung oem ddr3 1600c11 and g.skill 3200c14 which were purchased due to there overclocking capability and didn't cost that much more than slow kits of ram that couldn't oc

As for going over 16G I agree its a complete waste if all you do is game although there is some limited tests that suggests there may be a benefit from running dual rank ram which would mean running 32g with most of today's chips and its unclear if the gains from dual rank out way the loss of frequency

For arma 3 i used a stratis time demo average fps are nice and consistent but the minimum were very random so i left them out

Average fps in fallout 4 are not perfectly accurate im afraid due to it being a short run through the park with low fps at the start and high fps at the end if i was a second out starting\stopping the test it could throw the results a bit but the minimum seem very consistent

R6 your typical gpu bottlnecked game most ram reviews use here you can see the 3770k at stock with the gtx1070 oc is faster than the 6700k oc with stock gtx1070
R6 had the gpu pegged at 100% load while arma stayed well under being cpu bottlnecked
There is minimal improvement but the ram oc still makes a bigger difference than the cpu oc

Test System
3770k -- 6700k
Asus Maximus V Gene Z77-- Asus Maximus VIII Gene Z170
Samsung DDR3-1600c11 2x4GB 1.35v dual sided -- G.Skill DDR4-3200c14 2x8GB single sided
EVGA GTX1070 SC (OC=+75 core +750 mem +112 powertune use on all tests)
64g Samsung 830, 1tb Sandisk ultra II, 3tb WD green
Seasonic XP 1050 < tad overkill for the current system which has topped out at 306w from the wall
Silverstone TJ08-EW
Swiftech Apogee drive II
200mm Phobya rad + Silverstone Air Penetrator AP181
EK VGA Supreme HF
Win10

lightroom also sees over 20% improvement from higher ram speeds when exporting but only ~10% when rendering previews

time taken to export 150 24mp .nef to jpeg
2133c15 3:20
3200c14 2:50
3866c16 2:39

time taken to build 150 24mp 1:1 previews
2133c15 6:03
3200c14 5:33
3866c16 5:23

I also tested with aida64 and showed all sub timings used at each setting if you want a look
http://www.overclock.net/forum/1805...0c11-2133c9-ddr4-2133c15-3000c12-4000c17.html

 

[IntelUser2000]

RAM speeds, does matter. I don't know where people got the idea it doesn't.

The idea was always to get the 200MHz equivalent, and at least 166MHz. That translates into:

DDR: 333/400MHz
DDR2: 667/800MHz
DDR3: 1333/1600MHz
DDR4: 2666/3200MHz

When comparisons were done between gen x 200MHz versus gen x+1 133MHz, they would often perform worse despite 33% higher theoretical bandwidth. It's due to the overhead caused by the new memory standard.

I never considered getting DDR2-533 or DDR3-1066, because you needed the 166MHz equivalent to start outperform the top end of the previous generation. For DDR4, that would mean you'd want to discard DDR4-2133, and get DDR4-2666 at least. DDR4-3200 becomes the ideal point to shoot for.

No wonder Anandtech showed minimal improvements between Skylake and Haswell. Skylake used DDR4-2133, which is the suboptimal 133MHz equivalent. And some cases worse than DDR3-1600.

Nowadays though it seems 200MHz equivalent isn't the standard anymore like it used to back in the DDR2 and DDR3 days. DDR2-800 and DDR3-1600 were common among computers. Are DDR3-3200 modules common in modern systems? No!

When DDR5 comes out, shoot for DDR5-5333/6400, not DDR4-4266.