Welcome Guest ( Log In | Register )

Outline · [ Standard ] · Linear+

 [Article] Why you dont need more then ddr2 667mhz, If you dont do 450+ fsb overclocks :)

views
     
TSbryanyeo87
post Jan 27 2008, 02:43 AM, updated 18y ago

Below the Belt
*******
Senior Member
3,175 posts

Joined: May 2006
QUOTE
In personal computers, the front side bus (FSB) or system bus is the physical bi-directional bus that carries all electronic signal information between the central processing unit (CPU) and the northbridge.

Some computers also have a back side bus which connects the CPU to a memory cache. This bus and the cache memory connected to it are faster than accessing the system RAM via the front side bus.

The maximum theoretical bandwidth of the front side bus is determined by the product of its width, its clock frequency and the number of data transfers it performs per clock tick. For example, a 32-bit (4-byte) wide FSB with a frequency of 100 MHz that performs 4 transfers/tick has a maximum bandwidth of 1600 megabytes per second (MB/s). The number of transfers per tick is dependent on the technology used, with (for example) GTL+ offering 2 transfers/tick, EV6 4 transfers/tick, and AGTL+ 8 transfers/tick.

Notice that many manufacturers give the speed of the FSB in megatransfers per second (MT/s), and not in megahertz (MHz). The MT/s is affected by how many ticks are performed for each MHz, so if a motherboard has a 266 MHz FSB and performs 4 transfers per clock tick, it has a total data transfer rate of 1066 MT/s. That is what the manufacturers give as the speed of the FSB. Intel calls this technique which has 4 ticks per cycle Quad Pumping.





So, the FSB is actually a very inefficient bus system for the CPU to communicate with the rams and other parts of the PC. So lets take a look how much a bottleneck that the fsb does to the rams.


Test System:
Intel C2D E6750
Patriot Value rams 1gb x2
P35 DS3

Benchmark Used:
SuperPi XS 1.5, 2M calculations
Lavalys Everest cache and memory benchmark v4.10.1120 beta
Sisoft Sandra Engineer 2007.6.11.42
Windows XP 32bit SP3

This post has been edited by bryanyeo87: Jan 27 2008, 03:33 AM
TSbryanyeo87
post Jan 27 2008, 02:44 AM

Below the Belt
*******
Senior Member
3,175 posts

Joined: May 2006
E6750(333x8) 1:1 rams at 667mhz
user posted image



Results:

Sandra Test:
Ram Bandwidth Integer Buffered iSSE2: 5824MB/s
Ram Bandwidth Float Buffered iSSE2: 5826MB/s

Lavalys Memory:
Read: 6581MB/s
Write: 6063MB/s
Copy: 5791MB/s
Latency: 74.8ns





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






E6750 (333x8) 12:10 rams at 800mhz
user posted image



Results:

Sandra Test:
Ram Bandwidth Integer Buffered iSSE2 5991MB/s
Ram Bandwidth Float Buffered iSSE2 6154MB/s

Lavalys Memory:
Read: 6983MB/s
Write: 6063MB/s <=== results the same as the 667mhz
Copy: 6201MB/s
Latency: 73.1ns





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


E6750 (333x8) 16:10 rams at 1066mhz
user posted image


Results:
Sandra Test:
Ram Bandwidth Integer Buffered iSSE2 6370MB/s
Ram Bandwidth Float Buffered iSSE2 6406MB/s

Lavalys Memory:
Read: 7357MB/s
Write: 6062MB/s <=== results the same
Copy: 6488MB/s
Latency: 79.5ns


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

This post has been edited by bryanyeo87: Jan 27 2008, 12:26 PM
TSbryanyeo87
post Jan 27 2008, 02:44 AM

Below the Belt
*******
Senior Member
3,175 posts

Joined: May 2006
Now let see what happens when we raise the fsb to 400, and how much does it affect the Lavalys read and overall performance smile.gif

E6750 (400x8) rams at 1:1 for 800mhz
user posted image



Results
Sandra Test:
Ram Bandwidth Integer Buffered iSSE2 7106MB/s
Ram Bandwidth Float Buffered iSSE2 7087MB/s

Lavalys Memory:
Read: 7984MB/s
Write: 7275MB/s
Copy: 7080MB/s
Latency: 61.9ns

Wow, even at ddr2 800 the speeds was way faster then at 1066 is it was used on dividers



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






E6750 (426x8) rams at 1:1 for 800mhz
user posted image


[Results]
Sandra Test:
Ram Bandwidth Integer Buffered iSSE2 6395MB/s
Ram Bandwidth Float Buffered iSSE2 6425MB/s

Lavalys Memory:
Read: 7086MB/s
Write: 7745MB/s
Copy: 6723MB/s
Latency: 74.1ns

omg, the results suxxors, what happend??
The next person who can answer this gets a prize wink.gif

This post has been edited by bryanyeo87: Jan 27 2008, 12:40 PM
TSbryanyeo87
post Jan 27 2008, 02:45 AM

Below the Belt
*******
Senior Member
3,175 posts

Joined: May 2006
As you can see, at stock speeds of 333fsb and putting in higher speed rams only changes everything but of one value only. that value is Lavalys write speed, and that is the bottleneck that affects gaming and real life usage, the rest are all synthetic tests and do not reflect real life usage. Test it out, 1:1 and with dividers.

That is why if you do not do more then 450 fsb overclocks at 1:1 ratio, ddr2 667 rams are enough to justify. And anyway, not many people do more then 450fsb smile.gif

This system only affects Intel systems. AMD on the other hand, having the memory controller built into the cpu, has a higher bandwidth rate and a lower latency.



Conclusion:

Running the cpu not synchronously is not beneficial and yields little in performance gains. So best for Intel systems is 1:1.

This post has been edited by bryanyeo87: Jan 27 2008, 11:48 PM
TSbryanyeo87
post Jan 27 2008, 02:45 AM

Below the Belt
*******
Senior Member
3,175 posts

Joined: May 2006
-reserved-

uploading laugh.gif
TSbryanyeo87
post Jan 27 2008, 01:19 PM

Below the Belt
*******
Senior Member
3,175 posts

Joined: May 2006
yeah thats the one which affects everyday usage and gaming smile.gif


TSbryanyeo87
post Jan 27 2008, 04:56 PM

Below the Belt
*******
Senior Member
3,175 posts

Joined: May 2006
QUOTE(Dreamworks @ Jan 27 2008, 03:35 PM)
Now I'm having doubts about upgrading my RAM to performance ones since my proc can't do high enough clocks anyway laugh.gif  sweat.gif

I'm still curious though, what's the answer to your question? hmm.gif
» Click to show Spoiler - click again to hide... «

Is it due to insufficient voltage? Not enough juice to keep the RAM working at optimal status which causes the bandwidth to drop blush.gif  rclxub.gif
*
Not that i know of, but rams dont drop speed just because of insufficient volts, afaik it BSOD's tongue.gif
No, see closer ;p


QUOTE(a1098113 @ Jan 27 2008, 04:16 PM)
tips for smart computing brows.gif dont get parts u dont need and only get what u need laugh.gif for intel users. good point stated Bryan.
*
haha, thanks, belum finish it la, was quite late last nite haha
But putting in higher rated rams into a lower fsb system and using dividers does have a slight performance gains in 1 or 2 benchmarks, which is rarely used la haha


QUOTE(Hoodlums @ Jan 27 2008, 04:45 PM)
Hmm daily use mine is 350Mhz*9..
So for best performance RAM must be 700Mhz? hmm.gif
*
no, best is 1:1, meaning even ddr2 667mhz rams can be oced to that level which is 700mhz.
TSbryanyeo87
post Jan 27 2008, 05:34 PM

Below the Belt
*******
Senior Member
3,175 posts

Joined: May 2006
QUOTE(Dreamworks @ Jan 27 2008, 05:31 PM)
Man... This is like a game of "spot the difference" laugh.gif

It's the Performance Level! One is at 7 while the other at 14 tongue.gif What does Performance Level do anyway? Mine is 6 at the moment hmm.gif
*
yeah thats the killah on performance, normally people overclock with them set to auto in bios, its also known as tRD in bios

at 1600fsb or less it can be done with 7, or above 8 or 9, this affects a large part of the memory bandwidth, as well as stability la tongue.gif


Added on January 27, 2008, 5:39 pm
QUOTE(uzairi @ Jan 27 2008, 05:30 PM)
No point getting 10th Anniversary if you max out at 460mhz lets say and even Value rams can do it. Mem timings doesnt affect much unless you got very deep pockets and dont know what to do with the money.  laugh.gif
*
not really, lower latency and at 1:1 still the same then, say a higher ram speed at the same latency but of course higher speeds = more voltage

but averagely, for intel, cas latency tighter then 5 or even 4 in fact, does little to get better performance:voltage sweetspot
unless of course having deeper pockets then ok la hahaha






Footnote:
-the above is true for Intel systems as of the time of post.

-AMD x2 and s939 with onboard mem controller benefits alot more with tight timings and higher bandwidth due to the architecture.

This post has been edited by bryanyeo87: Jan 27 2008, 05:39 PM
TSbryanyeo87
post Jan 27 2008, 06:15 PM

Below the Belt
*******
Senior Member
3,175 posts

Joined: May 2006
QUOTE(Dreamworks @ Jan 27 2008, 06:02 PM)
For Performance Level, is higher better or lower better? rclxub.gif
*
See results tongue.gif

compare the higher performance level vs the lower tongue.gif

This post has been edited by bryanyeo87: Jan 27 2008, 06:16 PM
TSbryanyeo87
post Jan 27 2008, 11:36 PM

Below the Belt
*******
Senior Member
3,175 posts

Joined: May 2006
QUOTE(wodenus @ Jan 27 2008, 09:19 PM)
That is assuming you can push your 667 RAM to 800. If not you'll still have to use 800 RAM.
*
true also la, but ask u la, how many people in LYN actually push it above 400, or even 400 .__.

quite few mar hehe tongue.gif

and if 667 rams cant do even 800mhz, i would suggest to snap it into 2 and toss it away lmao, no im serious LoL
TSbryanyeo87
post Jan 30 2008, 03:00 PM

Below the Belt
*******
Senior Member
3,175 posts

Joined: May 2006
QUOTE(Radeon @ Jan 30 2008, 09:01 AM)
actually he posted the picture wrongly,
terbalik di
*
no la, its because performance level was set at a higher level, its also known as tRD in bios, its the NB FSB latency timing

and no ur not the first to answer it, so no prize for u tongue.gif hahaha
TSbryanyeo87
post Feb 1 2008, 10:42 PM

Below the Belt
*******
Senior Member
3,175 posts

Joined: May 2006
lol.... XMS2.... doh.gif

no comment, you bought paliah punya rams. yes its paliah. lol...

for the same price you could try getting balistix or 10th annies or hell, even the elite rams


Added on February 1, 2008, 10:46 pmoh yeah, btw, there is a intel 45nm thread, you may ask there for better response laugh.gif and happy ocing thumbup.gif

This post has been edited by bryanyeo87: Feb 1 2008, 10:46 PM
TSbryanyeo87
post Feb 10 2008, 02:46 PM

Below the Belt
*******
Senior Member
3,175 posts

Joined: May 2006
QUOTE(skymyxe @ Feb 4 2008, 11:17 PM)
I'd buy the Value RAM's and save some cash if ain't goin' more than 450 FSB.  brows.gif
*
another thing to moot is that, high fsb = high Northbridge temps and the possibility of raising the NB voltage.

cstkl1 has brought up something interesting is that tRD vs fsb vs timings, so sometimes having uber high fsb is bad too...


QUOTE(Radeon @ Feb 5 2008, 02:00 PM)
it's it hynix fpy-5 chipset?
450 is consider quite low
*
nah, fpy5 kinda suxxors nowadays, those before nov 2007 is ok la haha
TSbryanyeo87
post Feb 14 2008, 02:47 PM

Below the Belt
*******
Senior Member
3,175 posts

Joined: May 2006
yala thats why i said it is bad to have extremely high fsb la sweat.gif


i wouldnt say the thread is misleading, because the point is to find what and where exactly is the bottleneck in the NB to the rams, which was the Memory Read, which does not increase unless FSB is pumped

 

Change to:
| Lo-Fi Version
0.0319sec    0.82    6 queries    GZIP Disabled
Time is now: 3rd December 2025 - 04:25 AM