Confused CPU/RAM i7

ged325

Distinguished
Dec 3, 2007
32
0
18,530
I've been out of the HW game for a little bit, so please forgive me if the question is simplish.


With haswell coming out I'm noticing that the ram supported is DDR-3 1600mhz. I thought one of the major architecture changes to the i chips was that they removed the FSB (and please correct if I'm wrong.) How does the memory / CPU clock come into play in the newer architectures?

If I were to put DDR3-2400 would I just be throwing away money?

Is there a good article out there that explains the newer architecutre and the relationships between Motherboard, CPU, RAM, voltage and speed?

I'm beginning my research for a new build so any help greatly appreciated.
 
Solution

Maxx_Power

Distinguished
The CPU memory controller has a few different ratios, so if you run the RAM at a different ratio, say 2:3 to the memory controller clock, you might get 3/2*1600 = 2400 Mhz. The modern Intel CPUs offer a bunch of different ratios to support, effectively, 1333, 1600, 1666, 1800, 1866, etc...

If you were going to spend some money, there is a diminishing point of returns in how fast the ram is and what the net benefit there is to be had. If you are not using the onboard graphics, you can get away with just 1600 Mhz with full CPU performance. For reasons like OCing and flexible memory speeds, you can invest in a set of 1866 or 2133 Mhz RAM, so as to have room to play with ratios, and speeds. The price on the 1866 stuff and some 2133 Mhz RAM are nearly the same as for the 1600Mhz, so you can cover your future expansions nicely too, suppose you might upgrade the CPU+Board in the future.

I'll try to find you an article on this.
 
since the processor only supports up to 1600 in order to reach anywhere near those speeds you will need to overclock quite a bit. But as shown in certain tests on this site the improvement over 1600 mhz to 1866,2100, etc is very small. the memory bus clocks are where you see the real improvement in performance with games etc. there is no fsb anymore in these chips like amd. the now have a bclk. Its basically the same thing as fsb but it seems less stable when trying to up the bclk of intel. also the bclk i believe is on die vs on mobo but correct me if im wrong
 

Maxx_Power

Distinguished


The memory doesn't necessarily run 1:1 to the memory controller's default speed (usually 1333 or 1600). There are what used to be called memory dividers back in the day of S939 Athlons. Now they are more frequently referred to as just memory ratios. Ivy Bridge, for example has a final memory frequency determined this way

03.png


F(DDR) = BCLK x (1.33 or 1.00) x DDR Ratio

So you see you can change the DDR Ratio as allowed by the memory controller to get a bunch of supported frequencies. These are the FULL supported frequencies on a Ivy Bridge platform:

frequencies.png


When you choose a different speed, the motherboard requests the memory controller for a different ratio (strap).

Taken from this article (although common knowledge anyway) and highly recommended for the OP who was asking for these kinds of articles:

http://www.xbitlabs.com/articles/memory/display/ivy-bridge-ddr3.html

As for the FSB, there is no such thing any more. The Front Side Bus (FSB) was a term for pre-integrated memory controller/hubs days, where the CPU "talks" to the North Bridge, which had the memory controller inside, at a rate, called the FSB frequency. It is so called "Front side" because there used to be back side buses where the CPU connected with cache memory, like in Pentium IIs. Nowadays, the AMD platforms use HyperTransport (which doesn't negotiate any more CPU-memory traffic) and QuickPath Interconnect on Intel systems, which does the same thing, with variations.

 

ged325

Distinguished
Dec 3, 2007
32
0
18,530


@ Maxx

Thanks very much I think I got it. Increasing the BLCK speed will still require you to up the voltage to the chip correct (for stability)?

From that article it looks like the 1.33 or 1.0 doesn't make that much of a difference, but the RAM speed definitely does.

Please correct me if I'm wrong or confirm the accuracy about the facts but based on this I will most likely be running the 2666mhz as the 2800 is just way overpriced.

 

Maxx_Power

Distinguished


As far as the voltage is concerned, there are 2 main voltage sets you need to deal with in an modern Intel platform. There is the CPU voltage (set) and the memory voltage (singular).

All memories are rated for some specific voltage, with a JEDEC standard of 1.5V for nominal voltage, and 1.35V for low voltage (also DDR3L), there are also lower voltages like 1.25V and even lower, but I'm not sure if those are standard voltages for desktop use (although you CAN get 1.25V sticks for desktop). For Ivy Bridge and Sandy Bridge CPUs, the absolute max WAS stated to be 1.65V, because beyond this, the transistors degrade rapidly in the memory controller, which is onboard the CPU. Since a lot of motherboards have a voltage tolerance that isn't tight, getting 1.65V memory for an Ivy Bridge is risky, because although you set the voltage at 1.65V in the BIOS, the actual voltage could be, say 1.70V when measured with a meter. I would recommend (and you can gather by looking at memory specifications) that unless you have a specific need for super-fast RAM, stick to what the fastest is, at 1.5V. If you don't use the onboard GPU on a modern CPU (Ivy B or AMD's APUs), you don't need to necessarily go above 1600 Mhz to get all the performance benefits. BUT, for the ease of future upgrades and some OC room, I would recommend 2133 Mhz or 1866 Mhz RAM (the fastest ones you can get while keeping RAM voltage at 1.5V). Higher speed RAM also have higher compatibility issues, for example, DDR3 at 3000 Mhz (VERY FEW of these exist) are reputed to work on 1 in 10 Ivy B systems.

The CPU voltage actually consists of a few different voltages (or voltage planes) on modern CPUs, like the memory controller/PCI-E hub/etc which might use its own voltage, the CPU cores (which might even have voltage per core), the cache, etc, etc. On an Intel (or AMD) platform, you can increase the memory controller voltage (small adjustments are safe) to support higher speed memory, to allow the memory controller to "pump"/"charge" the memory in a smaller unit of time (since the clocks are ticking faster on a faster set of RAM). From Xbitlabs:

"Our experiments suggest that high-speed memory modes are perfectly functional and do not even require any tricks like fine-tuning secondary voltages. For example, it only took a small (by a mere 50 millivolts) increase in memory controller voltage for our Ivy Bridge CPU to work faultlessly with DDR3-2667 SDRAM."

These are the different voltage planes on an Ivy Bridge CPU:

02.png


-----------------------------------------------------------------------------------------------

NOW, as for the BCLK clock, it is a Master clock of sorts on an Intel platform. This means that nearly all other functions on the board that requires a clock to strap to (sync to, latch on to, or phase-lock on to), will be strapped to the BCLK. Usually as a rule of thumb, if you are overclocking the BCLK, plan on being able to get about nominally 5% before you start seeing weird issues. Although the CPU might be perfectly stable at higher BCLKs, the rest of the system (USB controllers, SATA controllers, etc) may start to have issues. A higher BCLK typically causes other onboard components to have problems, since those components are not usually overclocked/overclockable and are made with rather small clock margins. This is about how much (maximal) you'll be able to get from BCLK, although for stable day to day use (and Win8 is particularly picky about the BCLK), I'd recommend less than this OC:

image015.png


Taken from Tom's roundup of boards:

http://www.tomshardware.com/reviews/z77-express-ivy-bridge-benchmark,3254-34.html
 
Solution