Haswell Memory: Cas vs Frequency?

[wasabiman123]

cool so on my rig

1600 mhz at cas 7 = 4.375

Nice, but then again it is at 1600mhz, 2133mhz is a fair jump I would think it bandwitch/access times? This is interesting, really cool ratio almost of frequency/real latency, which is what that one guy was saying but now I get it :sneaky:

 

[Edrick]

example:
1600 MHz at CAS 9 = 5.625
2400 MHz at CAS 12 = 5.000

so in this case CAS 12 is lower latency than CAS 9.

No. 1600mhz is the double rate. The actual rate is 800mhz. So its 9 / 800 x 1000 = 11.25ns

 

[wasabiman123]

No. 1600mhz is the double rate. The actual rate is 800mhz. So its 9 / 800 x 1000 = 11.25ns

Ah...well either way you can tell there is a difference xD

 

[Edrick]

cool so on my rig

1600 mhz at cas 7 = 4.375

Sorry, its actually 8.75ns latency. The real rate is 800mhz (1600 is the double in DDR)

 

[piasabird]

So what happens when you increase the processor bus speed? Is that possible? So does that affect the RAM as well or is that partitioned off? i.e. Memory controller part of Intel CPU. This is something you might need a good benchmark to investigate. Intel built the memory controller maybe they have software to overclock it. i.e. use aftermarket cooler and adjust the Turbo boost variables? So what happens in the processor when it goes into turbo boost?

Just some ideas, dont bother answering, cause I dont plan to ever overclock.

Then there is the low voltage DDR3L AND HD Video and IRIS(Integrated Video)? I think Intel is producing parts with too many variables.

 

[Edrick]

So almost double for 2ns shaved off roughly. Thoughts anyone?

I get the fastest speed with the lowest latency I can afford.

For example, when I bought my new Haswell system, I could have got 2400mhz @ 11 CAS or 10 CAS. The 10 CAS set cost like $10 more. And the real world number are 9.16ns latency for the 11 and 8.33ns for the 10. Easy decision for me.

 

[Makaveli]

Sorry, its actually 8.75ns latency. The real rate is 800mhz (1600 is the double in DDR)

Ahh thank you

 

[wasabiman123]

I get the fastest speed with the lowest latency I can afford.

For example, when I bought my new Haswell system, I could have got 2400mhz @ 11 CAS or 10 CAS. The 10 CAS set cost like $10 more. And the real world number are 9.16ns latency for the 11 and 8.33ns for the 10. Easy decision for me.

Gah, well I guess I gotta go for the best lol, oh well not happy about adding 65 bucks to my planned build, but that's just how the dice rolls. At least it'll be better than me going from 3x1 set to a 3x2 set, to finally my current 3x4.
Never again :'(

 

[Concillian]

No. 1600mhz is the double rate. The actual rate is 800mhz. So its 9 / 800 x 1000 = 11.25ns

Actually YES.

Note I left off the units and never claimed they were ns.
This is because the units don't really matter except that they are both directly porportional to time and of the same order. The time ratio is the same whether you use the "correct" speed or the "doubled" speed.

So both what you posted and what I posted are, in fact, true and correct. I didn't need corrected.

People are more familiar with the advertised "double" speed and determining whether one is faster than another doesn't matter whether you use the actual or advertised speed. I use the advertised rate just because it's easier. And I maintain that it is in no way wrong to do so when all you're trying to do is determine if one is faster than the other.

 

[Edrick]

Gah, well I guess I gotta go for the best lol, oh well not happy about adding 65 bucks to my planned build, but that's just how the dice rolls. At least it'll be better than me going from 3x1 set to a 3x2 set, to finally my current 3x4.
Never again :'(

As the more budget conscious people on this forum will tell you, that the difference in memory speed will not affect your experience with many applications. Only those which really tax the memory subsystem will see a benefit.

 

[Edrick]

I use the advertised rate just because it's easier. And I maintain that it is in no way wrong to do so when all you're trying to do is determine if one is faster than the other.

Agreed, both methods will clearly define which is faster or slower. But if you want the actual nanosecond measurement, you need to take the pre-doubled speed.

 

[wasabiman123]

As the more budget conscious people on this forum will tell you, that the difference in memory speed will not affect your experience with many applications. Only those which really tax the memory subsystem will see a benefit.

I can afford the better RAM, and would like to have peace of mind knowing parts that are being used are carefully chosen for specific reasons, being an enthusiast to a certain degree.

 

[Carfax83]

The same thing happens in the Just Cause games.

I think the engine designer forgot to allow large system memory caching in the Dunia engine since it was designed almost exclusively for the extremely memory capacity limited current gen consoles.

I know that Just Cause 2 only uses about 256 mB of system memory and has massive pop-in issues.

The Unreal 3 Engine games also have this problem when the developers don't set the system memory and video memory cache high enough even when there is available resources on the target system.

Well, whatever causes it, having faster memory doesn't alleviate it; at least in Far Cry 3. I tested it and I was still seeing glaring texture pop in..

Crysis 3, although it's smaller in scope, has ZERO detectable texture pop in on my rig and Crysis 3 makes Far Cry 3 looks like a Super Nintendo game.

The engine is just crap. It was obviously designed for consoles primarily, hence craptastic draw distance. As a matter of fact, some guy posted a video comparing the retail version of Far Cry 3, with the demo version in E3, and the differences are striking:

Far Cry 3 retail vs E3 demo

That's just sad.