GA-EP45-UD3P and 8GB RAM recommendation

Toggle sidebar Toggle sidebar
A

abelhern

Distinguished
Jul 31, 2009
6
0
18,510
Hello,
I've been having the most frustrating time with a GA-EP45-UD3P(rev 1.1) and RAM when I fill all four banks with 2GB modules (8GB total). I've tried Corsair 8500 and G. Skill 8500. I've spoken to Gigabyte tech support
for a total of about 12 hours now during the past 2 months. For reason when all banks are filled with 2GB sticks, no matter what values in the Bios I change/tweak, the modules run at PC6400(800).

If I wanted 800 speeds, I would have purchased PC6400 modules. Apparently, if all banks are filled, the RAM speed needs, and the system does it automatically to be underclocked.

I'm wondering if anyone has had success in running 8GB of PC8500 at stock speed. And if the reality is that in order to use all four banks the RAM needs to be underclocked, is it recommended to get much faster RAM so at least I'll be able to run it at a speed higher than 800? If so, does anybody have recommendations?

Any comments, input, direction to other resources, advice, etc. will be EXTREMELY appreciated.
 
yea i think i read something a while back stateing that these boards needed a bios upgrade to use memory over 800mhz

But it could always be that the ram you bought was never tested and true on that specific motherboard..it happens. Some boards are picky.. Also depnds on if your mixing them with various other speeds. The slowest speed will be run. Mislabeled ram also happens :heink:
 



Thank you so much for your suggestion. I actually have the exact same G. Skill product. I had purchased 2 kits of 2x2GB of Corsair because of the heat dissipation design of the modules. I had the G. Skill from a previous build and thought they ran a bit too hot for my liking even at 800 (I did not measure the temp. I went by feel). The Corsair turn out to be useless so I installed the G. Skill and the system was running stable at 800. Of course, I purchased 1066 RAM to run it at 1066. If I wanted to run RAM at 800 I would have purchased PC-6400 modules.

Anyway, the bottom line is that no matter what I did based on suggestions from Gigabyte tech support and G. Skill email support the system persistantly under clocked the 8GB of RAM. I also tested the modules in EVERY possible combination with all 4 memory banks on the motherboard.

To summarize, I'd like to inform you that I followed the recommended steps outlined in the post you refferred me to and although my CPU is the Quad Core 9550S, The system is running smoothly, and very stable and cool with 8GB of RAM at 1066 and the CPU's FSB at 1600 with a speed of 3.4 as opposed to stock speed of 2.83(?) with FSB of 1333.

I am extremely appreciative for responding to my post and directing me to one of your posts relevant to my situation. It was extremely helpful, informative and enlightening. Due to that information I finally accomplished my goal without any problems. Thank you.

Regards,

Abel

 
Always welcome!

I think that posting's been a 'three for' - took care of three people's problems with this MOBO (so far); as they used to say on the 'A-Team' - "I love it when a plan comes together!" :lol:
 
Ummm - if you're interested, you probably have quite a bit of 'headroom' left in that processor; you could bump the host clock up a few or five at a time - see what she does. As you pass 410 or so (which will put your memory at 1090 - I've tested six sticks, and all were capable of 1100, at or below rated 2.1V), you will want to lower the memory multiplier; you can try the 2.5 multiplier on the 266 latch (actually, 'strap'), but it may not work at that high a frequency - the 2.4 multiplier on the 333 latch will work, and lower your memory to 985 at a 410 clock; on that multiplier, you should be good, memorywise, to about a 450 host clock; most people who've OC'd that chip seem to be good 'till the general neighborhood of 425 or so (3.6 GHz) with CPU voltage held down to 1.3V

Always loved the A-Team - only place you can see six thousand rounds of automatic weapons fire in forty minutes, without one drop of blood!
 
Not directly related to your hardware, but this reminds me of two similar cases in the past.

Intel's i815ep has 3 Ram slots, offers CL2 and CL3 timings, but the data sheet tells that the Bios should switch to CL3 if it detects 3 Ram modules, as the i815ep's buffers can't move heavy loads that fast. I don't know which mobo manufacturer followed the recommendation then. You could have a look at the P45's data sheet.

Asus did a nasty trick on the falsely praised TUSL2-C (with said i815ep). As soon as the user put the FSB over ~145MHz in the Bios (instead of 133MHz) the Bios switched the Ram to 3-3-3-7 and the AGP to 2x, without any remedy within the Bios' settings. As these slow settings allowed a higher Fsb, users believed Asus had achieved a faster Mobo.

And by the way, Intels gives the P45 for 800MHz only, even with a single module. Gigabyte is out of specs at 1066MHz - though I firmly believe the P45 does it easily. I can easily imagine that Gigabyte's Bios throttles the Ram if using 4 modules.
 



Ok, now I'm a little lost. I vaguely got the part about cranking up the CPU, but I'm a bit confused about the RAM. My main objective was to run the RAM at 1066. You mentioned that I can go even higher (1100) with the RAM. Then you commented on cranking up the CPU but I'd have to run the RAM at a slower speed.

Please pardon my ignorance, but I always thought that faster RAM made a big difference on system performance. You seem to suggest that CPU speed even with RAM running a bit slower is preferable. Can you please enlighten me if you can? I would greatly appreciate it.
 
Please pardon my ignorance, but I always thought that faster RAM made a big difference on system performance.
Click to expand...
People pretty consistently overestimate the actual 'real-world' effect of faster RAM - I have a little (well, not so little, but instructive) 'canned' piece:

To further complicate matters, people often misunderstand the actual quantitative speed improvements inherent in faster ram... Here's the mistake: 1066 is 33% higher than 800 ([1066-800]/800 = 266/800 = .33), so 1066 RAM must be a third faster than 800, right? Not so! You have to figure in latencies. Most 800 will run at 4-4-4-12, while most 1066 is rated at 5-5-5-15, or, even worse, 5-5-5-18. Here's how to appraise the situation in reality: at 800 MHz, a RAM bus cycle is 1.25 μSec long (1000/800); at 1066 (1000/1066), it is roughly .938 nSec long - so, with an 800 stick at a 4 average latency, a RAM bus transaction takes 1.25 (cycle time) times 4 (latency), or 5nSec, while at 1066 it is .938 (cycle time) times 5 (latency), for a transaction time of (roughly) 4.7nSec - so you see, by going to nominally 33% faster RAM, you actually gain three tenths of a nSec per transaction - .3 (transaction gain) over 5(transaction total) = .06, for a real-world improvement of 6%

Now, you have control over the basic system clock (I'm going to cal it B_CLK), once you start manually timing the MOBO through the BIOS. B_CLK times four is your FSB (once again, Front System Bus); B_CLK times your memory multiplier is your DRAM rate; B_CLK times your CPU's multiplier is your CPU frequency.

Examples: if you set your system clock to 333, you will need a 2.4 memory multiplier (333 x 2.4 = 799.blahblahblah) to run your RAM at 800, and if the CPU multiplier is, say, 8.5, you will get a CPU clock of 2.83GHz; at that same B_CLK you would need a memory multiplier of 3.2 (3.2 x 333 = 1065.6) to take advantage of 1066 RAM. Now, lots of CPUs that are rated at a 1333 nominal FSB will run a lot faster, sometimes with a little more 'oomph' from a voltage increase; for example, I run a Q9550 that is rated at 1333 FSB (333 B_CLK) times an eight point five multiplier, for a 2.83GHz speed. It will comfortably run with the B_CLK well over 450 - and here's where faster RAM comes in. The smallest RAM multiplier available from a MCH (Memory Control Hub - or 'NorthBridge') is 2.0, but, with a 2.0 multiplier, that means at a 450 clock, your RAM will need to run at 900 (again, 450 B_CLK x 2 = 900), which most 800 RAM just won't do! This is referred to as a 'RAM limited bus', meaning the CPU can't run a B_CLK any higher than (roughly) half the RAM's available speed - and thus, the need for faster RAM. Mind you, this only applies if you both can, and intend to, run your FSB above 1600 (once again, a B_CLK of 400+ times 4 gives you a 1600+ FSB)...

When overclocking, things work in 'steps'; the memory multipliers are not actually decimal numbers, as they are displayed in the BIOS - they are integer ratios: the 2.66 we are using is actually a 4:3 ratio; 2.5 on the 266 'strap' is actually 5:4, the 2.4 on the 333 'strap' is 6:5 - so you see, we cant just pick pick from a continuous 'range' of multipliers; if the 2.5 on the 266 strap will not work (and my guess is it won't), there is nothing else between 2.66 and 2.4 - we can't just 'dial-in', say, 2.60, because there's no such ratio available. So, as you bring the FSB up, you have to take the memory multiplier down in discrete 'jumps', which means that, at the lower end of the FSB range, the memory, with the lower multiplier will actually run a trifle slower, but the advantages of having the CPU faster outweigh this big time!





 
jebus.. most fast computers these days run the ram faster then the cpu... great for video / games that load lots of frames per second / textures at once.. it was once prefered to run it 1:1 or as close to it as possible.. w/e that may be i don't get how you get a 1:1 ratio from 3ghz cpu vs 1ghz ram.. other then it's estamated bandwidth.. the cpu would then be 40-400GB/s and ram at todays speeds 40-80gb/s to me thats 3:1 without multipliers applied.. simulated clock speeds of ram i have now..(basically 400mhz x2..x4 and some other multiplyer) 3.4ghz and a 3.2ghz cpu.. so pretty close to 1:1 in theory..

"decreaseing" the timings could present new instabilities and certain large data pools won't be able to flow like they should.. often just cause the data doesnt stay there long enough or as expected by the software running.. and cause more heat, becuase the speed of the flow is increased..

"increaseing" the timings to a higher number tends to open up the ram, slow down the speeds or latency and run cooler but more stable.

In the end you are still stuck with what works... I've had ram that normally ran at 4-4-4-10 and used 2-3-3-10.. the first number is the clock speed / number of cpu cycles the rams uses to communicate to the cpu and the system.. but the lower it is the Less time it takes that info to cycle the system.. at under 90ns getting ram under 100ns is a feat.. everything ran great, but hot.. untill i started playing MMO games.. that used 1gb or more of my ram.. then i got blue screens stateing halts and memory overflows etc.. exceptions etc

getting a program like pc wizard 08 helped me understand data flow in sizes for ram
visually.. which then helped me dial in settings that worked.. and stayed below 100ns.. 3-4-4-10

It also benchmarks the ram itself in the process.. letting you know what settings would be best for several diffrent sizes of information..

I hope this is less confusing and more informative.. But i definately recommend that program, before you go frying your new ram..
 
I put together a little chart here to illustrate:
0048n.jpg


The top shows 800 RAM, the multipliers for each Intel standard FSB, and you can see that, as 2.0 is the lowest multiplier available, you can't take the FSB much past 1600 (400 system clock) as, for each each 'tick' of the system clock, you are going 2 more past 800 - which the RAM won't like...

The bottom shows 1066 RAM, the multipliers for each Intel standard FSB, and goes into detail for raising the FSB past 1600 (400+ system clock); I made the chart to show what happens when you have 1066 RAM that you know 'tops out' around 1080 or 1090; with the base 2.66 multiplier, you're going to 'run out of headroom' somewhere between 406-410 system clock; if you can get the 2.5 multiplier (which may not work, as it's on the 266 'strap' [more about that later...]) to operate, it will, at the same point where your RAM 'peters out' at 2.66, only be hitting 1025 - and the 2.5 will have you 'covered' till the system clock hits 436 - where, once again, your RAM will be at 1090, its upper limit... Then we have the 2.4 multiplier (which will work, as its on a high strap), which has you covered up to a 454 system clock. So, you can see how there is a 'step-wise' operation here, ratcheting your RAM speed up and down somewhat as you 'walk up' the system clock to crank your processor...

Now, I'll yack a bit about the 'straps'; much like memory, which requires more bus cycles of latency as the cycles themselves become shorter (faster), the northbridge's connection to the CPU has an inherent need to adjust its latencies as the FSB increases. These are arranged in four sets, to correspond to the four Intel supported FSBs. On the P43/P45 boards, you select these 'straps' explicitly, by setting the "(G)MCH Frequency Latch" in the BIOS - once this is set, you will only 'see' that strap's available multipliers in the next item, "System Memory Multiplier (SPD)"; in the older boards, based on the X38/X48, the strap that applies to each multiplier is expressed as a letter or symbol after the multiplier's decimal expression. If you use AMD, you're an aberrant who deserves whatever particular hell Murphy and his immutable Law assign to you!

So, for the 200 system clock (800 FSB) strap, which is shown by a 'C' or a '+' after the multiplier, you get 2.66 (4:3), 3.33 (5:3), and 4.00 (2:1); for the 266 system clock (1066 FSB) strap, which is shown by an 'A' or a '~' after the multiplier, you get 2.50 (5:4), 3.00 (5:3), and, once again 4.0 (2:1); for the 333 system clock (1333 FSB) strap, which is shown by a 'B' or a '#' after the multiplier, you get 2.00 (1:1), 2.40 (6:5), 3.20 (8:5), and, yet again 4.0 (2:1); and for the 400 system clock (1600 FSB) strap, which is shown by a 'D' after the multiplier,you get 2.00 (1:1), 2.66 (4:3), and, 3.33 (5:3).

People impute a certain 'magical' capability to the fabled 1:1 ratio, which happens only with the 2.0 multipliers on the 333 and 400 straps... For the most part, this is pretty much hogwash. The 'magic' is this: certain memory bus transactions (and they are a limited few) can only occur when the memory and processor busses are 'synch'ed; in other words, with say, an 8:5 ratio (3.2 mult), the memory and CPU insert 'wait states' till one finishes counting cycles to eight, and the other one five (doesn't matter which way you look at it, the time elapsed is, by definition, the same). That means that the average wait for these transactions is half the worst case, or 7/2, three and a half cycles. Obviously, at a 1:1 ratio, these transactions are always available on the 'next count', so for these transactions, it is faster. However (and it's a big however), you are far ahead by running your RAM at a higher speed - and damned the transaction waits! On my system, my system clock is at 450, and I'm running the 2.4 multiplier, for a 1080 RAM speed. To get 1:1 (2.0 mult), I'd have to lower my RAM speed to 900, and the loss of throughput ([1080-900]/900=.2, or 20%) far outweighs the 'immediate transaction' gain (a couple percent, at best)...

Hope this has made these factors in ram speed and overclocking a bit clearer!

Bill
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom