Age old debate: speed vs latency

[alexisvx]

I have read about it but still find it so confusing, can you please advise? Which one will perform better for gaming:
All the below are $73-$80. My Mobo supports O/C

1. Hyper X - DDR3 1600 (PC3 12800) - Cas Latency 10 - Voltage 1.5V
2. Hyper X - DDR3 1333 (PC3 10600) - Cas Latency 9 - Voltage 1.5V
3. Team Vulcan - DDR3 2400 (PC3 19200) - Timing 11-13-13-35 - Cas Latency 11 - Voltage 1.65V
4. G.Skill - DDR3 1600 (PC3 12800) - Timing 9-9-9-24-2N - Cas Latency 9 - Voltage 1.5V
http://www.newegg.com/Product/Product.aspx?Item=N82E16820104442

1.
http://www.newegg.com/Product/Product.aspx?Item=N82E16820104442

2 http://www.newegg.com/Product/Product.aspx?Item=N82E16820104430

3.
http://www.newegg.com/Product/Produ...-na-_-na&cm_sp=&AID=10446076&PID=3938566&SID=

4.
http://www.newegg.com/Product/Produ...-na-_-na&cm_sp=&AID=10446076&PID=3938566&SID=


My system:
Motherboard:
Asrock Z97 Extreme 4

CPU:
i5 4690k

CPU cooler:
Noctua NH-D14

PSU:
EVGA 750W ATX12V

Gcard:
GeForce GTX 650Ti Boost 2GB

RAM:
Corsair XMS3 DHX 4GB (DDR3 1333)

Hard Drives:
1x SSD - 70GB
1x 300 GB

Case:
Fractal Design Define R4

 

[rayraysfunhouse]

For gaming you are really not going to see a huge difference from 1333 to 2133 maybe 2-3 fps, maybe.

 

[EarthDog]

Its going to be a six of one, half dozen of the other really...

 

[ATMINSIDE]

2x4GB? Grab some G.SKILL 2133 CL9. Best of both worlds.

 

[alexisvx]

Ok, thank you for the info. I'll probably wait a bit and see if any better deals turn up for thanksgiving week.

 

[BugFreak]

I'm a little confused here. After reading the overclock guide sticky, I was under the assumption that timings do matter with these chips? I understand that CPU clock is most important but the article talks about tighter timings helping. Is it just so marginal it doesn't matter?

Thus, you know all these high speed kits that were selling well with Ivy Bridge? Get ready to see how tight they can tighten their timings at lower speeds. Timings are back folks. No longer can IC manufacturers count on increasing memory clock, timings be damned. If you can get your hands on an older PSC or BBSE based kit with tighter timings...

OC Guide Sticky

 

[Woomack]

For gaming you are really not going to see a huge difference from 1333 to 2133 maybe 2-3 fps, maybe.

1333 memory is actually causing system to slow down. Personally I see the difference in general usage when memory is running at 1333/1600 auto settings and 2133+ at tigher timings like CL9-10.

On haswell memory performance ( I mean regular settings from memory which you can buy ) is not scalling good above 2133 so 2133 clock at tighter timings seem optimal. Other option is something like 2400 CL10 which usually cost not much more.

On AMD FX memory is not really scalling above 1866 clock and 1866 is actually default memory clock. In AMD APU series memory is not scalling above ~1866 but it helps in IGP performance a lot.

On SB/IB were different memory controllers and difference between 1600 and 1866+ was lower than in haswell. Haswell has faster cache and new CPUs have generally large cache so what you need more is high memory bandwidth while on older series access time was usually more important.

Manufacturers are going for larger and faster cache + higher memory bandwidth what is clearly visible in DDR4. Even though timings look really bad then general performance is quite good and depends more from cache and memory controller than memory itself.

In all cases most important is balance between high clock and tight timings what gives you optimal performance. Too high clock makes you set more relaxed timings so you lose on access time while too low clock but tight timings will give you good access time but low bandwidth.
Every platform is acting different so there is no one clear rule how to set memory.

 

[EarthDog]

Here is a good read ALex...http://www.anandtech.com/show/7364/memory-scaling-on-haswell

 

[Woomack]

Here is a good read ALex...http://www.anandtech.com/show/7364/memory-scaling-on-haswell

There is only one problem with this article. It's not covering single and dual rank matter when difference in memory performance can be up to 20%. It doesn't change much in daily usage as it's stil +/- 3% general performance but this comparison isn't really good.
Example can be performance of double sided 2133 CL9 kit which is about as high as 2666 CL11 single sided while 2666 CL11 double sided is beating everything lower clocked.
It's just not one straight line from 1333 to 3000 as they presented.

 

[EarthDog]

Its a good high level read (what the OP wants)...

 

[ATMINSIDE]

There is only one problem with this article. It's not covering single and dual rank matter when difference in memory performance can be up to 20%. It doesn't change much in daily usage as it's stil +/- 3% general performance but this comparison isn't really good.
Example can be performance of double sided 2133 CL9 kit which is about as high as 2666 CL11 single sided while 2666 CL11 double sided is beating everything lower clocked.
It's just not one straight line from 1333 to 3000 as they presented.

I'm sorry, but if you really think he (the OP) cares about single/dual sided and benchmarking performance then you've gone silly.
All he was asking is if timings or speed should be given more attention.

 

[Woomack]

I'm sorry, but if you really think he (the OP) cares about single/dual sided and benchmarking performance then you've gone silly.
All he was asking is if timings or speed should be given more attention.

In my last post I only commented article which EarthDog posted. I said what I think about high clock or tight timings in other post.

 

[BugFreak]

1333 memory is actually causing system to slow down. Personally I see the difference in general usage when memory is running at 1333/1600 auto settings and 2133+ at tigher timings like CL9-10.

On haswell memory performance ( I mean regular settings from memory which you can buy ) is not scalling good above 2133 so 2133 clock at tighter timings seem optimal. Other option is something like 2400 CL10 which usually cost not much more.

On AMD FX memory is not really scalling above 1866 clock and 1866 is actually default memory clock. In AMD APU series memory is not scalling above ~1866 but it helps in IGP performance a lot.

On SB/IB were different memory controllers and difference between 1600 and 1866+ was lower than in haswell. Haswell has faster cache and new CPUs have generally large cache so what you need more is high memory bandwidth while on older series access time was usually more important.

Manufacturers are going for larger and faster cache + higher memory bandwidth what is clearly visible in DDR4. Even though timings look really bad then general performance is quite good and depends more from cache and memory controller than memory itself.

In all cases most important is balance between high clock and tight timings what gives you optimal performance. Too high clock makes you set more relaxed timings so you lose on access time while too low clock but tight timings will give you good access time but low bandwidth.
Every platform is acting different so there is no one clear rule how to set memory.

What would you consider tight timings at the 2133 speeds you mentioned?

 

[ATMINSIDE]

CL9 is what I would look for in 2133

 

[Silver_Pharaoh]

What was it again?

CAS / Speed * 1000 = latency

Sooooo:

1. Hyper X - DDR3 1600 (PC3 12800) - Cas Latency 10 - Voltage 1.5V
10 / 1600 * 1000 = 6.25ns

2. Hyper X - DDR3 1333 (PC3 10600) - Cas Latency 9 - Voltage 1.5V
9 / 1333 * 1000 = 6.75ns

3. Team Vulcan - DDR3 2400 (PC3 19200) - Timing 11-13-13-35 - Cas Latency 11 - Voltage 1.65V
11 / 2400 * 1000 = 4.58ns

4. G.Skill - DDR3 1600 (PC3 12800) - Timing 9-9-9-24-2N - Cas Latency 9 - Voltage 1.5V
9 / 1600 * 1000 = 5.625 ns

Purely from that standpoint, the Team Vulcan "is the fastest".

 

[Woomack]

Best option seems TridentX 2400 CL10. It cost like most 2133 CL9 kits and can usually run at 2133 9-10-10 / 9-11-11.
Team Vulcan is slow memory. 2400 CL11 2x4GB kits are single sided Hynix = perfomance like double sided 1866 CL10. Still better than any other kit on the list from 1st post.

 

[ATMINSIDE]

What was it again?

CAS / Speed * 1000 = latency

Sooooo:

1. Hyper X - DDR3 1600 (PC3 12800) - Cas Latency 10 - Voltage 1.5V
10 / 1600 * 1000 = 6.25ns

2. Hyper X - DDR3 1333 (PC3 10600) - Cas Latency 9 - Voltage 1.5V
9 / 1333 * 1000 = 6.75ns

3. Team Vulcan - DDR3 2400 (PC3 19200) - Timing 11-13-13-35 - Cas Latency 11 - Voltage 1.65V
11 / 2400 * 1000 = 4.58ns

4. G.Skill - DDR3 1600 (PC3 12800) - Timing 9-9-9-24-2N - Cas Latency 9 - Voltage 1.5V
9 / 1600 * 1000 = 5.625 ns

Purely from that standpoint, the Team Vulcan "is the fastest".

And this is why I like my Ares 2133 CL9.
4.219ns

 

[Woomack]

And this is why I like my Ares 2133 CL9.
4.219ns

Your ARES 2133 CL9 are double sided Hynix CFR modules. Nearly all new 2x4GB kits are single sided = read, up to 20% lower read/copy bandwidth. You can compare your kit to new 2x4GB 2666 CL11.
I was also happy about ARES 2133 CL9 kit but sadly new kits are not so good. I could run it 24/7 @2666 11-13-12 1.65V on IB or HW.

From new 2x4GB kits only part of TridentX 2133 CL9 or 2400 CL9/10 are double sided. Price of 2133 CL9 and 2400 CL10 is about the same while 2400 CL9 cost much more and are hardly available.

If PC is for gaming then it won't change much so anything 2133+ is good option even at more relaxed timings.

 

[BugFreak]

So based on that, would these Mushkin be good for a new Haswell build or should I find something better?

 

[ATMINSIDE]

So based on that, would these Mushkin be good for a new Haswell build or should I find something better?

Those will be good.

I typically recommend these:
http://www.newegg.com/Product/Produ...=g.skill_ares_2133_8gb-_-20-231-554-_-Product