Dear forum members,
We are delighted to inform you that our forums are back online. Every single topic and post are now on their places and everything works just as before, only better. Welcome back!


Discussion on Article:
What is the best memory for Haswell? From DDR3-1333 to DDR3-2933: performance scaling test

Started by: beenthere | Date 03/28/14 10:20:17 PM
Comments: 150 | Last Comment:  11/29/16 07:04:27 AM

Expand all threads | Collapse all threads


So the bottom line is exactly what it has been for the past 3+ years: Paying more for RAM with a frequency higher than ~2133 MHz. is a BAD decision because it does not produce any tangible improvement in system performance with real applications. For most people and most PC systems 1600 Mhz. or 1866 MHz. delivers all the performance the system can possible provide, period.

This of course is in spite of the marketing efforts of G.Skill and others who constantly try to convince people that they need the highly over-priced, huge profit DDR3 RAM, which they do NOT. Just because some Intel CPUs can run the RAM at high frequency does not make it practical or necessary.

In addition DDR4 will NOT be cheaper than DDR3 and it will NOT provide any tangible performance gains for desktop PCs over DDR3. DDR4 is designed primarily for servers where it can show some performance improvement but at a high cost. Unlike with DDR3 where you can add RAM as you desire with DDR4 you will be required to replace ALL of the RAM if you want to increase the system capacity.

A technically educated PC enthusiasts doesn't get duped by the marketing hype from RAM makers or purveyors of crap PC hardware.
1 1 [Posted by: beenthere  | Date: 03/28/14 10:20:17 PM]

"By changing your DDR3 SDRAM parameters, you can actually ensure a performance boost of 20-30%!"

^That statement really needs an asterik!

My take on the results:

- The isolated memory benchmarks aren't relevant to an end user.
- One game at low resolution showed significant improvement.
- The rest of the results showed insignificant improvement.

Don't waste your money on fast memory. This would agree with what most guru's in forums have been saying for nearly a year!
1 0 [Posted by: sanity  | Date: 03/28/14 10:25:23 PM]

Do I see prices correctly, is this 8GB kit more expensive than 32GB of DDR3 2133?

IMHO would be quite stupid to build a fast PC with low 8GB memory. Could it be upgraded at all? Shouldn't you at least check if 4 DIMM configuration work at that speed?
0 0 [Posted by: popej  | Date: 03/29/14 07:14:45 AM]

8GB is all that most people need for a typical desktop PC. If you have programs that can actually use more RAM like some video editing software, etc. and you're using an O/S that can use more than 8GB of RAM then there can be a modest improvement in performance but nothing major.

No one is going to see a 20-30% improvement in performance because most people use 1600 MHz. or slightly higher DDR3. The whole article like previous high frequency RAM articles is intended to dupe the technically challenged Intel fanbois into buying hyper expensive RAM that shows no tangible system performance gains with real applications. The technically challenge won't understand this even though it's documented in the test results. They will cling to meritless claims of 20-30% gains and not even know they are being deceived.
1 1 [Posted by: beenthere  | Date: 03/29/14 01:18:15 PM]

Seems to me that 128MB L4 cache of Intel would solve this? When is AMD coming out with something similar?
0 0 [Posted by: Ferdinand82  | Date: 03/29/14 02:19:54 PM]

This article is lacking the benchmarks for the integrated GPU.

There the memory speed should have bigger impact.
0 0 [Posted by: hkultala  | Date: 03/30/14 02:45:47 PM]

Can you add opencl and dx11 benchmarks using hd4000 with different memory clocks?
0 0 [Posted by: Dilinger  | Date: 03/30/14 06:37:18 PM]

IMO, the most important factor when choosing parts for your computer is BALANCE.

Fast RAM with an average CPU and VGA card won't change a thing.

I've done some test with memory speed using BF4. A fast CPU with powerfull graphic system and fast memories shows noticeable improvement. Remove one of these from the equation and it's over. Here's the test and results:

Here's the framerates I got in BF4 on my system using DDR3-1333 ( vs 1600 ( vs 2133mhz (

Test system:
Intel i7 3770k (stock speed)
Asus P8Z68-V Pro
2x4gb G.Skill F3-2133C10-4GXM
SSD-Muskin Chronos Deluxe 480gb
Sapphire 7970 Dual X as main and a Gygabite 7970 OC (clock 1000 gpu, 1475 ram)
Window 7 64bit (2yo installation)
AMD Catalyst 13.11 BETA

Recording with FRAPS in the Suez mission from chopper 1 appearance to chopper 2 down at Ultra preset; 1080p:

7970 single Gpu 1080p:
BF4 64 bit Framerates at 1333: 62.156 avg
BF4 64 bit Framerates at 1600: 62.53 avg
BF4 64 bit Framerates at 2133: 62.199 avg

7970 Crossfire 1080p:
BF4 64 bit Framerates at 1333: 93.425 avg
BF4 64 bit Framerates at 1600: 105.600 avg
BF4 64 bit Framerates at 2133: 111.012 avg
0 0 [Posted by: MHudon  | Date: 03/31/14 12:35:58 PM]

This supports what I've been saying on forums for over a year; faster memory = faster performance on modern platforms, and 2400 MHz is the sweet spot for RAM right now for price/performance gains. I don't care if it's only a few percent, the price increase is only a few percent, of RAM cost, not system cost, but the entire system enjoys the slight increase. The difference on NewEgg between my Adata 16GB 2400 MHz kit and the cheapest 16GB 1600 kit is only $20. I will always pay an extra $20 for 2% to 5% performance increase system wide. Others might not care, but I want my system to run at maximum potential.
1 0 [Posted by: Hood  | Date: 04/07/14 02:11:46 AM]
- collapse thread

Heeee nope, you've misinterpreted the results methink.

This supports the fact that (in gaming at least) systems equipped with fast CPU AND VERY FAST graphic solutions will take benefit from the speed increase. Others wont feel a difference.
1 0 [Posted by: MHudon  | Date: 04/09/14 03:32:45 PM]

Whilst the 5% of ppl with top of the range GPU's or dual GPU may see a SLIGHT improvement in FPS using overpriced top of the range DDR3 (along with the extreme $$$) the vast majority (95%) are wasting their time/$ using anything over 1866MHz

The Dramm manufacturers marketing dept. will tell you otherwise
0 0 [Posted by: alpha0ne  | Date: 04/18/14 04:32:47 AM]

"We can’t say anymore that memory subsystem parameters don’t matter."

Yes, we can...

"Judging by our test results, you may want to choose high-speed DDR3 SDRAM in two cases: when you build a gaming computer or a home PC for image and HD video processing."

Nope, it has no tangible effect on gaming at actual *PLAYING* resolution. The only 2 scenario's are:

1- Compressing stuff, which is usually done in the background while doing something else and only once in a while (who compresses stuff 24/7 non stop for a living???), so nothing critical if it take 20 seconds more, certainly not justifying 100% RAM price increase.


2- HD video processing, yes, *BUT* more data would be needed as GPGPU processing, independent from main RAM, hasn't been tested here... I wonder why?

The actual conclusion is:

There are other subsystem bottlenecks CLEARLY preventing 100% more RAM bandwidth to have a DIRECT CORRELATION with actual performance. For example, twice the CPU clock or twice the GPU shaders or having a RAID0 setup doubling storage speed, are all, more often than not, resulting in LINEAR performance increase, a progression curve that RAM will NEVER be able to achieve in the platform technology currently available.

In other words, spend your money elsewhere in your system if you want tangible and measurably better performance and buy the cheapest 32GB 1866 CL9 1.5V kit you can get. When it comes to RAM, more is STILL far better than faster.

How the author of this article could have reached any other conclusion is beyond me.
0 1 [Posted by: Ramon Zarat  | Date: 06/12/14 03:45:31 PM]
- collapse thread

Of course, we know that increasing the RAM bandwidth by 100% does not increase performance by 100%. This is not what the editor was trying to say, doofus.
0 0 [Posted by: Bo_Fox  | Date: 11/06/14 08:51:33 PM]


Back to the Article

Add your Comment