I assumed you realised that a I7 3820 & the I7 2600K are essentially the same chip except for the memory channel count (and PCIe lanes, which we'll agree it won't make a difference in single GPU setups) So it's a perfectly valid comparison.

 

It would be valid, but only in case of computational capabilities comparison. Due different I/O structure and possibilities it isn't valid in overall tests. Intel i7 3820 has completly different die. Only the same component of i7 3820 (SB-E) comparing to i7 2600 (SB) is microstructure of 4 Core units. Rest are different - LLC (2.5MB per core/shared vs 2MB per core/shared), System agent* and absence of Integrated graphics unit. Especially important are differences in System Agent.

*consist IMC, DMI, PCI-e controllers

 

So Aida64, just like MaxxMem2 and any other memory benchmarks are wrong, along with just about any real world performance tests out there, while the only synthetic bench that we should consider is Sandra just because it exposes the benefit from that theoretical extra throughput?

 

I don't judge which one of these testing applications (AIDA64 vs SiSoft Sandra) give us more true answer. Just point significant differences showed between two memory tests published in the one article.

 

There's a reason why no one (with enough knowledge and common sense) considered Bloomfield anymore for FSX

 

I don't understand why you call old Bloomfield here if we are talking about SB-E CPU. It was Gulftown between and finally SB-E, which has new microarchitecture and I/O unit. Ability to use more than dual channel architecture isn't good reason to set SB-E in one line with Bloomfield.

It would be valid, but only in case of computational capabilities comparison. Due different I/O structure and possibilities it isn't valid in overall tests. Intel i7 3820 has completly different die. Only the same component of i7 3820 (SB-E) comparing to i7 2600 (SB) is microstructure of 4 Core units. Rest are different - LLC (2.5MB per core/shared vs 2MB per core/shared), System agent* and absence of Integrated graphics unit. Especially important are differences in System Agent.

*consist IMC, DMI, PCI-e controllers

 

Sorry but no, they are the same thing except for the amount of L3 cache, memory channel count, PCIe lanes and the IGP (lack thereof in SB-E). It's the same architecture, the same system agent, the same LLC (again only difference is the amount of L3) same IMC, same cores. Of course none of those benchmarks involve testing the IGP since there's no IGP in SB-E.

There's a myriad of benchmark results showing both chips doing exactly the same and that's good proof of...

 

a) they're the same thing for the most part (do you honestly think that both get exactly the same scores clock for clock and core for core in everything by sheer chance?)

B) the few things that are different are non factors when it comes to performance, including the memory channel #

 

I don't judge which one of these testing applications (AIDA64 vs SiSoft Sandra) give us more true answer. Just point significant differences showed between two memory tests published in the one article.

 

Just one benchmark. The SysSoft Sandra memory BW benchmark. That's the only one I know and it's not even a real world application

 

I don't understand why you call old Bloomfield here if we are talking about SB-E CPU. It was Gulftown between and finally SB-E, which has new microarchitecture and I/O unit. Ability to use more than dual channel architecture isn't good reason to set SB-E in one line with Bloomfield

 

Just because Bloomfield sports tripple channel. If a newer platform like Sandy Bridge yields better results across the board (including memory benchmarks with the only exception of Sandra) on just two channels, that should give you a clue of what's going on. This is nothing new. Back in the day when Nehalem was the cream of the crop, you had Lynnfield and Bloomfield looking like this

 

http://www.anandtech...oduct/107?vs=46

 

looks familiar huh?

Guys

 

Sorry if I've started a bit of a war here... After much reading of your info and other stuff online, today I ordered G-Skill 16GB (4x4GB) DDR3 2133MHz RipjawsZ X79 Memory Kit CL9 1.65V to replace the 1600 memory currently in use. This should allow me to punch the overclock on my motherboard a bit higher too, as I "feel" the 1600 ram was holding it back...

 

Its currently running at 840mhx because I have the 3.6gHz CPU running at 3.8, so a slight overclock...

I tried the ASUS extreme overclock setting (Auto overclock) but it only went up to 4gHz before becoming unstable, and from my findings, I kind of feel its the ram holding it back, so hopefully, when the new ram arrives, I'll slap it in and see if that allows the board to get a better overclock.

 

That, in turn, for me, makes everything faster, but yeah, I can see how faster memory would give quicker and more responsive loading of graphics/textures, but I "thought" my ATI6990 took care of all that?

 

All this gaming and overclocking is still fairly new to me, as I'm more a nuts/bolts/networks/servers person...

 

Heck I was overjoyed recently to find a £6k server I bought came with 32mb of onboard VGA graphics. Happy days!

You didn't really need to get new RAM. It's true that it was most likely limiting your auto-OC because apparently your board decided to use a combination of BCLK, CPU strap, CPU multiplier & RAM multiplier that doesn't make much sense at all. This is how crappy auto overclock can be. God only knows how much Vcore it's pushing. So you should seriously consider reseting your BIOS to defaults ASAP and use a manual overclocking approach instead. It's really not that hard.

I know this jibber jabber will probably sound intimidating, but the basic principle in OC is very simple. You have a base clock (BCLK) and a number of multipliers applied to that base clock that effect different components. The BCLK should always remain at 100MHz in all Sandy Bridge variants including Sandy Bridge E and Ivy Bridge. You can find the relevant setting in that BIOS at 13:31 in the youtube video: "BCLK frequency"

then you have 3 multis,

1.- CPU multi for your CPU frequency. It's set to 35 at stock settings, so 35 x BCLK = 35 x 100MHz = 3500MHz = 3.5GHz. To set the CPU multi, see video @ 14:27; set Turbo Ratio to "By all cores" and then "By all cores (Can adjust to OSD" to your value of choice. The thing is that your 3820 has a locked CPU multi, well, not quite, it can't be increased past 43 (that would be 4.3GHz, so your auto OC could have just set this to 38 or 40 and not mess with the BCLK or the CPU strap at all and achieve the same overclock). Other Intel chips can go much higher than 43, but that doesn't mean your 3820 overclocks any worse, see CPU strap below in case you aim to overclock higher than 4.3GHz

2.- CPU Strap. This is simply a multiplier that effects both you CPU and your RAM speeds. See vid @ 13:30 "CPU Strap" to locate the relevant BIOS setting. The CPU Strap is set to 1 by default, and the only setting you need to care about here is the 1.25 strap if you want to get past 4.3GHz. Why? this is the formula to calculate your final CPU freq:

BCLK x CPU strap x CPU Multi

so with a 1.25 CPU Strap

100MHz x 1.25 x CPU Multi

since the default CPU Multi is 35

100 x 1.25 x 35 = 4250MHz

and since the max CPU multi is 43, you can (theoretically) go as high as

100 x 1.25 x 43 = 5375MHz

Now the CPU Strap also applies to RAM speed. So let's address the RAM multi first and then go back to the CPU strap and how it effects your final RAM frequency

3.- The RAM Multi, just like the CPU Multi and every other multi is based on the BCLK

RAM multi x BCLK = RAM speed

you have a number of RAM multis that are going to allow you set your RAM at 1066MHz, 1333MHz, 1600MHz 1833MHx or 2133MHz. That's provided that the BCLK is set to 100MHz as it should be and the CPU strap remains at 1x

If you increase the CPU strap to 1.25 those RAM ratios become:

1066 x 1.25 = 1333MHz
1333 x 1.25 = 1666MHz
1600 x 1.25 = 2000Mhz
1833 x 1.25 = 2291MHz
2133 x 1.25 = 2666MHz

So with your new 2133 RAM set you should probably use the 1600 x 1.25 = 2000Mhz ratio, but if you can get it stable at 2291MHz that will boost performance by a wopping 2-3%, no seriously, probably not worth the hassel, but if you want to try it, well that's your call. The thing is that overcking RAM makes a minute difference and can easily make a system unstable.

See vid @ 15:53 for "Memory Frequency" adjustment. That's where you set your RAM multi

RAM speed is also determined by RAM timings (aka latencies) see vid @ 16:00. Basically you should just run your RAM as close to manufacturer's specs as you can (XMP profile)
The lower the timings, the faster your RAM works, but if RAM freq makes little difference, well moreso timings. (I know some will disagree, I won't bother with RAM speed arguments... for now haha)

So this covers the basics on multies. The other crucial part are voltages (Vcore, Vdimm, Vccio...)
I'll cover that later

Awesome, thanks mate..

Gives me plenty to mess about with for now...

I'll sit down and have a good watch over that all later, as I'm building 2 identical systems almost, 1 with P9X79 Pro and the other with the Deluxe..

Gives me plenty to mess about with for now...

 

It can be written a bit simpler way:

 

1. First is bus clock - BCLK, it affects CPU and memory. Default value is 100 MHz. For Sandy Bridge, SB-E and IB it can be raised in range from 100 up to 108 MHz (max) depending mobo model, but it is recomended to leave it at default value.

 

2. Next is BCLK multiplier avaliable on X79 mobo. Possible values are 1.00, 1.25, 1.66, 2.55. Default value is 1.00. Formula for calculation is very simple - BCLK * BCLK multiplier. Good values for use are 1.00 (100MHz) and 1.25 (125Mhz). Second two are too high for normal using.

 

3. Next one is CPU multiplier. Formula for calculation also is very simple - BCLK * n, where n is multiplier from range 17 to 43* in case of 3820. Default value for 3820 is 36.

* it may be 44 due to differences in avaliable documentations.

 

Examples of using:

1) Default values, BCLK multiplier stays on default (1.00), CPU multiplier stays on default (36), BCLK stays on default (100MHz)

BCLK multiplier * CPU multiplier * BCLK = CPU speed

1.00 * 36 * 100 MHz = 3.6 GHz

 

2) only BCLK multiplier is changed to 1.25, CPU multiplier stays on default (36), BCLK stays on default (100MHz).

BCLK multiplier * CPU multiplier * BCLK = CPU speed,

1.25 * 36 * 100 MHz = 4.5 GHz

 

3) BCLK multiplier stays on default (1.00), CPU multiplier is changed to 43, BCLK stays on default (100MHz)

BCLK multiplier * CPU multiplier * BCLK = CPU speed,

1.00 * 43 * 100 MHz = 4.3 GHz

 

4) BCLK multiplier is changed do 1.25, CPU multiplier is changed to 38, BCLK stays on default (100MHz)

BCLK multiplier * CPU multiplier * BCLK = CPU speed,

1.25 * 38 * 100 MHz = 4.75 GHz

In case of such high frequency you need really efficient cooling and have to raise voltage.

 

Recommended way is to change BCLK multiplier to 1.25 and CPU multiplier to reasonable value (for example - 37).

Check your memory modules for avaliable XMP profiles. You may try one of such settings if avaliable:

- memory frequency 1600MHz * 1.25 BCLK multiplier will giving you finally 2000MHz for your DDR3 modules.

- memory frequency 1866MHz * 1.25 BCLK multiplier will giving you finally 2333MHz for your DDR3 modules.

Don't forget about latencies change if you won't be using ready XMP profiles.

 

 

Sorry but no, they are the same thing except for the amount of L3 cache, memory channel count, PCIe lanes and the IGP (lack thereof in SB-E). It's the same architecture, the same system agent, the same LLC (again only difference is the amount of L3) same IMC, same cores.

 

I repeat - Intel i7 3820 has completly different die than i7 2600. Only the same component of i7 3820 (SB-EP-4) comparing to i7 2600 (SB) is microstructure of 4 Core units. Rest are different. Sandy Bridge-EP-4 die is "cut off" version of Sandy Bridge-EP-8 die. Here is comparative statement - http://pc.watch.impr.../523/674/04.jpg - only the same parts for i7 3820 and i7 2600 are light blue area (core units).

 

Sandy Bridge-EP-8 die - http://images.anandt...GA_2011_Die.jpg

Sandy Bridge die - http://images.bit-te...dge-die-map.jpg

I repeat - Intel i7 3820 has completly different die than i7 2600. Only the same component of i7 3820 (SB-EP-4) comparing to i7 2600 (SB) is microstructure of 4 Core units. Rest are different. Sandy Bridge-EP-4 die is "cut off" version of Sandy Bridge-EP-8 die. Here is comparative statement - http://pc.watch.impr.../523/674/04.jpg - only the same parts for i7 3820 and i7 2600 are light blue area (core units).

 

Sandy Bridge-EP-8 die - http://images.anandt...GA_2011_Die.jpg

Sandy Bridge die - http://images.bit-te...dge-die-map.jpg

 

The layout is obviously going to be different, but the technology is the same. Remember when Lynnfield introduced the on-die PCIe controller while it remained in the North Bridge in Bloomfield? One in the mobo, the other one in the die, one 16 lanes the other one 32... but the same controller.

You can choose to believe whet you want man, but this is a very well known thing and once again there you have the benchmark resuts to prove it