Memory bandwidth tests... any real differences (PC5300 vs. PC8888)

[harpoon84]

Btw, to further this topic, I'm currently compiling results for a FSB scaling experiment.

I have decided on 2.25GHz, and testing will be done as follows:
6 x 375 / DDR2-750 3-3-3-9 1:1 ratio
10 x 225 / DDR2-750 3-3-3-9 3:5 ratio
10 x 225 / DDR2-450 3-3-3-9 1:1 ratio

This should provide ample analysis on how much extra performance pure FSB scaling brings to the table, as well as seeing how a 'slow FSB/fast RAM' combination compares to a 'fast FSB/fast RAM' combination.

Stay tuned!

 

[harpoon84]

OK, here are the FSB scaling tests:

Timings as per previous post, 3-3-3-9 for all tests. CPU is @ 2.25GHz for all tests.

wPrime 32M: (lower is better)
6 x 375 / DDR2-750 - 41.672 secs
10 x 225 / DDR2-750 - 41.702 secs
10 x 225 / DDR2-450 - 41.687 secs

Aquamark3 1024x768 0xAA/4xAF Max Details:
6 x 375 / DDR2-750 - 126.90 fps
10 x 225 / DDR2-750 - 121.09 fps
10 x 225 / DDR2-450 - 115.79 fps

SuperPi 1M: (lower is better)
6 x 375 / DDR2-750 - 24.797 secs
10 x 225 / DDR2-750 - 25.156 secs
10 x 225 / DDR2-450 - 25.625 secs

X3 Reunion Rolling Demo 1024x768 0xAA/0xAF Low Details:
6 x 375 / DDR2-750 - 75.57 fps
10 x 225 / DDR2-750 - 71.296 fps
10 x 225 / DDR2-450 - 69.179 fps

X3 Reunion Rolling Demo 1680x1050 8xAA/16xAF High Details:
6 x 375 / DDR2-750 - 62.933 fps
10 x 225 / DDR2-750 - 61.30 fps
10 x 225 / DDR2-450 - 60.701 fps

WinRAR benchmark:
6 x 375 / DDR2-750 - 1023 KB/s
10 x 225 / DDR2-750 - 964 KB/s
10 x 225 / DDR2-450 - 870 KB/s

Cinebench R10:
6 x 375 / DDR2-750 - 4306 CB (3m:25s)
10 x 225 / DDR2-750 - 4298 CB (3m:25s)
10 x 225 / DDR2-450 - 4261 CB (3m:27s)

x264 Benchmark:
6 x 375 / DDR2-750 - 65.762 fps P1, 15.90 fps P2
10 x 225 / DDR2-750 - 64.094 fps P1, 15.76 fps P2
10 x 225 / DDR2-450 - 63.442 fps P1, 15.626 fps P2

Nothing surprising from these results, it again confirms that C2D only scales marginally from higher FSBs and RAM speeds. This is good news for budget overclockers as they don't need to buy top end RAM and/or high end overclocking mobos to attain 'near bleeding edge' performance.

 

[graysky]

Yeah dude... good data set. Thanks for taking the time to run/post it. I did a similar analysis in a thread entitled, Which is faster, 9x333 or 8x375 on a q6600? and drew a similar conclusion.

 

[Conjugal Visit]

This is just a great thread. This, along with graysky's O/C thread for Intel CPU's, has been so helpful. The comments and links folks have posted in response to graysky's efforts have been just as helpful. I've read and re-read them many times and used graysky's basic formula to O/C my own rig. But as much as I try to understand all this (and google and google) I still have some questions that I can't find answers for - at least that I can make sense of. For instance:

Is it a fact, that in real-world applications (and gaming) it is preferrable to to sync memory and FSB at 1:1 on an Intel C2D platform?

If the answer is yes (and my understanding is that it is yes), why would I plug anything faster than good performance ram spec'd at DDR2- 800 (4-4-4-12) into my board?

Example:
E6750 - x8 multiplier - 333mhz x 8 = 2.66ghz stock speed.
Air-cooled
1:1 ratio
Increase DDR2 from 400 to 450 (12.5% O/C - 4-4-4-12 maintained and doable on good Ram)
FSB increases - 400 x 8 (multiplier) = 3600 or 3.6GHZ O/C (35% increase in FSB)

3.6GHz is a pretty healthy O/C on air (mandating an aftermarket HSF)

Why would I want to plug PC-1066 memory into an Intel board if PC-800, running stock @ 400 will yield an immediate O/C to 3.2GHz, and will permit an O/C to 3.6Ghz with a reasonable 12.5% O/C on the memory?

Asked another way, wouldn't the use of PC-1066 necessitate running the memory underclocked, at 450 x 8 = 3.6GHz, versus stock value of 533? I mean, 533 x 8 = 4.26GHz, and that just isn't within the realm of possibility with these CPUs. Or is it?

So again, if syncing memory with FSB is desirable - real world - with Intel's CPUs/chipsets, why is there a market for PC-1066 and beyond? Latencies are better at PC-800 anyway, aren't they?

Someone help me to understand this, would you?

Thanks in advance and regards,

 

[JustaGeek]

IMO, there is no real gain in running the memory 1:1 with the FSB.

If there was a problem, why would Intel and MB manufacturers even suggest using the memory dividers...?

The only advantage might be the increased stability. It might be easier to achieve a stable overclock with the memory bus running in sync with the FSB, but if you can speed up your memory by using a divider, the increased bandwidth will increase the overall performance of the system.

The gain might not be significant, but in the world of "crunching numbers", the gain is the gain nevertheless.

Just my 2 cents...

 

[graysky]

@Conjugal Visit - the higher dividers are there because some folks wanna run their hardware balls-to-the-wall and if company ABC's 'enthusiast' board didn't have those options, and company XYZ's board did... that would represent a competitive disadvantage. To put it another way, the speedometer on some cars goes up to 180 MPH. Do you think most owners drive it near that level

I think your question about PC-1066 is probably right.... 6x533 = 3.2 GHz which I could see someone running, or 7x533, etc. Again, not everyone knows that simply by adding/running your memory at high speeds translates into scant speed gains. Some people just buy hardware on specs and think they're getting their money's worth out of the hardware in my opinion.

 

[VirtualLarry]

This thread shows how powerful Intel's marketing has become, even though there is little to no gains from faster FSBs and faster RAM. Core clock speed is still king, for the C2D.
It also makes me wonder what Intel's engineers were smoking, when they designed the chipset memory multipliers, since they are actually multipliers, and not divisors, like AMD implemented in their IMC. Since the multipliers aren't practical (or useful), but a divisor would allow a higher FSB (and thus core clock speed!), while still using slower RAM. AMD got these things right, IMHO, while Intel still doesn't have that much of a clue.

 

[theprodigalrebel]

I just wanted to bump this thread for graysky and harpoon84's excellent test results. Mighty helpful.

 

[Emission]

Originally posted by: JAG87
Graysky, there is no sense in having memory bandwith when you have no interconnect bandwith. You are pretty much maxing out a 333 mhz FSB with 333 mhz memory. If you want to see the 555 mhz memory shine you need to increase the FSB, otherwise the ram is being bottlenecked by it.

But if you increase the FSB, the chipset strap will increase, and the latencies will increase, and its all back to square one. So deep down you have the correct idea, but your attempt to prove it is very biased.

Yes, the methodology is technically incorrect, because you're running your memory faster than your system bus, so there will be minimal if any gains in this configuration.

Nice try though, the problem here is that these chips don't need more bandwidth than they we're given to begin with, unless of course you overclock the chip. A 2.25 GHz quad is not very memory hungry, if we we're to push past 3 GHz, you'd see some pretty big gains.

How about 10x300 vs 6x500 ? If it's possible with your hardware that is.

 

[tjcinnamon]

Extreme noob:

So what is the point of getting pc2-8500 over pc2-6400 or even pc2-3200. If FSB does not play a significant roll in performance gain, why doesn't everybody try to get ram that has the lowest latencies instead? I'm like a monkey with a machine gun. I have 4GB of corsair dominator running at stock 5-5-5-15 and probably 533MHZ. So is what your saying all I should do is lower the latency.

I'm a bit lost so be gentle.

Thanks,
JOe K.

 

[graysky]

@Emission - Can't hit 10x300 since my multiplier is locked @ 9 and 6x500 doesn't POST.
@tjcinnamon - People get faster mem usually because they wanna run a higher FSB. You can try to lower the latencies, but you should test the results; I have dropped mine from 4-4-4-12 to 3-3-3-9 and found that superpi gets slower.

 

[tjcinnamon]

Isn't the whole point of what you guys were talking about is that raising the FSB has nominal performace gain and does not out weigh the added heat in most cases?

 

[tenax]

the whole point is that the performance variations you see by most benchmarks mean nothing..if the computer doesn't feel faster, if you don't get a notable faster burn, or transfer of data or better gameplay (and by better i'm saying you go from some stutters to smooth play..or you can up the visual appeal of a game or multitask without affecting other tasks..i.e. record a show without stutters while burning a dvd while surfing like you can since dual core cpus arrived) then it's all nothing but fun. even 25% increases in memory bandwidth are not going to be noticeable in tasks..and i can affirm that my computer runs "faster" in feel running stock fsb and memory at the lowest cpu temp possible, then it does if i push it to a 50% overclock and push my voltage from .9 to 1.4 on my 6400. the 3.2 gig versus 2.13 looks great on paper..but in the real world application, means diddly squat.

 

[graysky]

tenax hit the nail on the head. I always recommend that people do their own benchmarking using applications they use all the time on the pc. It really doesn't matter if your overclocked 8800GTS can score 12,500 in 3dmark if you don't play games. Or if your Q6600 can run @ 3.6 GHz when all you do with the machine is browse the web.

The point is that overclocking can be fun to do, but unless you're extracting some real benefit from doing it, all you're really doing is pushing your hardware, using more electricity, and creating more heat

That said, if you're playing Crysis and you can push your video board from 650 MHz to 800 MHz and your frame rate goes up from say 23 FPS to 31 fps, and that increase makes your gaming experience more enjoyable, that is what I would call a real gain. Or, say that you do video editing and have queued up 60 hours of mpeg-2 encodes in your queue. Running your machine @ an o/c of say 25 % (3.0 GHz vs. 2.4 GHz for example) reduces that 60 hour queue to 45 hours then again we're talking useful work here.

 

[tenax]

well said graysky..exactly

 

[graysky]

..the classic push-back to everything you and I said is that, the user owns the hardware and if he/she wants to push it for no reason other than "because it's mine and I can," who are we to say otherwise

 

[graysky]

double-post

 

[tenax]

agreed, graysky..when i write comments like i did, it's not to educate the experienced..it's to educate those new to overclocking who walk into it thinking that their system is going to feel 2x faster than it did because of overclocking..now, if i wasn't an enthusiast or didn't get into htpc, i would have likely stopped with my a64 3400 which served me quite fine. my decision to switch to my cd 6400 was based on research..and a big need to have a system that i could still do other stuff with while i was recording shows. with my 3400, recordings would come out jittery, audio out of sync on playback, due to the tasking i was trying to do otherwise. and i record a lot of tv for the family at large, burn movies and shows to dvd..so there's a case where switching hardware was driven by necessity now, switching from my 6400 to a 9450 or 9550 is simply the enthusiast in me, jealousy and the money burning a hole in my pocket

 

[polecatt]

Let me make a statement about ram usage then you guys tell me if i'm understanding it correctly.
Also I have included a question about my motherboard.

Q6600 has a FSB 1066 or effective FSB of 266.5mhz (1066/4=266.5) 266.5*9 Multiplyer = 2.4 Ghz

If i want to increase my q6600 from 2.4 to 3.0ghz I can raise the FSB to 333mhz or (1333FSB)
9 multiplyer * 333mhz = 3.0Ghz

So if I have DDR2-1066 (pc2-8500) the FSB is equal to 533mhz.
I don't have to change the Ram Ratio, Ram Timming, Ram Speed
Because my RAM will not run any faster than 333mhz FSB
Realy I could get away with DDR2-667 (pc2-5300) because the FSB=333mhz
So I would just use the manufatures timming, voltage settings, and fsb settings.

Oh and one more thing, if my motherboard supports 1066/1333 FSB

Then if I increase the FSB to 400mhz then 9*400mhz= 3.6ghz
Then my rated FSB = 1600mhz
Isn't this over spec'ed for my motherboard?

Or I would then be overclocking my motherboard and need to increase the voltage on the NB, SB, FSB term. etc
for it to work......?

 

[tenax]

actually your ram will go faster than 333. at timings is 533 (x2) maybe you meant to say that anyway..to get 400/400 (1:1 ratio) you're going to be overclocking your fsb a bit (from 266 to 400) and underclocking the spec on your ram (533 to 400) which should mean no extra volts needed to ram, opportunity to try tighter timings if you like on the ram, and some voltage increases to hit that rise in fsb..(on your northbridge, cpu etc)

sorry..edited..if you read before my edit, i thought it was a 333 fsb part you were talking about, not 266 so i predicted a milder tweak on voltages than you would require to make this jump..

 

[polecatt]

How can my ram run faster than 333mhz @ 3.0Ghz?

If the effective FSB is 333mhz then my DDR2-1066 (PC2-8500) is overspeced. (actual FSB of 533mhz)

I thought my ram can only run as fast as the processors FSB.

 

[tenax]

no..your mb should have options in bios to run in ansynchronous mode meaning you could run for example, 266 fsb/400 on memory speed (2:3 ratio) i prefer 1 to 1 myself..i don't think you get any benefit running asynchronous cause your fsb speed is still a bottleneck..that argument has been going on forever by the way, since i had my first asus a7v build around 1997. look at it it this way:

if you could have an fsb of 533 (4 x 533..wouldn't that be sweet) and your memory running at 533 ( x 2 as it's ddr2) then you would be in 1 to 1 mode and making the most of that memory you have.

i currently have a 266 fsb x 4 processor as well..but my memory is rated for 400 x 2..i can't run my processor at 400 fsb without pumping a lot of juice through it so in fact, i'm in much the same boat as you..but, i have a 333 x 4 capable board, memory is easily capable of handling it, so once i get a 1333 quad core..i'm going to be laughing, if they aren't too tough to overclock.. 400 x 4-1600 fsb, memory to match..and i'll push the board a bit past 1333 but don't expect it'll be a big deal.

 

[polecatt]

OK, so if the motherboard is rated at 1333 and you increase the FSB to 400*4 or 1600
1600-1333 = a 266mhz jump in FSB of the motherboard.

So you think my motherboard can handle the jump, I would have to mess with the voltages. right?


But if I can keep my Q6600 cool enough I should be able to hit 400mhz effective FSB
Hit 3.6Ghz and still not have to mess with my Ram, "DDR2-1066 (PC2-8500)", (keep it 1:1) right?

 

[tenax]

correct! i'm not familiar with all your bios settings, but if i had a quad 6600 in my board with that goal, same memory i would:

raise the system, fsb and mch voltages by .1 volt at a time. (i'm using terms as referenced for my board and bios)
leave the memory voltage as it is.
give .1 more juice to the cpu

and see where that takes me.

 

[polecatt]

Thanks, you where very informative, hope I can get some pointers later.
My parts come in next week, First time I built a computer in 9 years. I'm trying to get
me a nice gaming machine to compete with my friends. They will be amazed when I show up
with a soopped up machine.

Boy it took a lot of research to figure out this overclocking stuff, I just want to be knowledgeable
about what I'm doing, and this will be fun too, I need a new hobby. And some bragging rights......

Besides I don't want do something stupid and fry my nice new components.