• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Super Pi - 417mhz 3-3-3-8 vs 500mhz 4-4-4-8 who wins ?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
eva2000

eva2000

Member
​
Joined
Aug 5, 2002
Location
Brisbane, Australia
Decided to see how well tight 3-3-3-8 timings compare to high clocked 4-4-4-8 timings. Only difference is 4-4-4-8 timings were done on 2GB Gskill PC6400PHU2 HZ Micron D9GMH 0629 batch @500mhz 4-4-4-8 and tight 3-3-3-8 timings were done with 2GB OCZ PC2-8000 Titanium Alpha VX2 Micron D9GKX @417mhz 3-3-3-8 at 2.39v

System:
  • Intel Core 2 Quad Q6600 Kentsfield QUPT B3 Step 7
  • Corsair Hydrocool 200EX
  • Intel D975XBX Badaxe rev304 1441 bios
  • 128MB Gainward FX5200 PCI card
  • 3x320GB Seagate 7200.10 SATAII Matrix Raid 0
  • 1x74GB Western Digital Raptor 8MB
  • Pioneer DVD-RW
  • FDD
  • 1KW PCP&C SLI PSU


6400phu2_1.jpg


vs

titanium_alphavx2.jpg

Intel Core 2 Quad Q6600 Kentsfield quad core cpu was clocked on both tests at 9x333FSB = 3001Mhz.

2:3 divider 500Mhz 4-4-4-8 vs 4:5 divider 417mhz 3-3-3-8

cpuz_validated_tn.png


vs

cpuz_validated_tn.png


Results​

500mhz 4-4-4-8 (left) vs 417mhz 3-3-3-8 (right)

1M
16.907s vs 16.812s

2M
40.750s vs 40.750s

4M
1m 34.454s vs 1m 34.000s

8M
3m 31.156s vs 3m 31.157s

16M
7m 44.281s vs 7m 44.141s

32M
16m 48.438s vs 16m 47.562s

Amazing that 417mhz 3-3-3-8 timings not only kept up with but also beat 500mhz 4-4-4-8 timings in 4 out of 6 of the Super Pi times. Especially in 32M Pi test! I suspect it's the tighter timings but also due to tighter default tRD value. According to memset v3.0, 417mhz 3-3-3-8 had by default tRD = 5 whereas 500mhz 4-4-4-8 had tRD = 7 by default.

Probably explains why PC6400 rated DDR2 memory in 3-3-3-8 default timings usually cost as much if not more than PC8000 rated memory due to it's performance as well as the time and cost it takes for the hand testing and binning process! :)
 
Nice analysis george, it is interesting indeed. Seems like top mhz is the best for benching except if you use 3-3-3. Guess that's why all the PI WRs are w/ 3-3-3 :)
 
I honestly think its virtually tRD all by itself that's able to push CL3 forward.

Btw George how come you couldn't run tRD 5/CL 3 in the past, but can now??
 
I do remember reading a prediction somewhere that memory timings were going to be much more important with C2D's architecture than they were with the old netburst. This seems good confirmation of that conjecture.
 
thanks guys

for intel 975x chipset 1:1 is optimal for cpu:mem any extra bandwidth only helps in bandwidth intensive apps like super pi

with most recent games they're more GPU and/or CPU intensive so probably best to have tight timings 1:1 high clocks anyway

Gautam said:
I honestly think its virtually tRD all by itself that's able to push CL3 forward.

Btw George how come you couldn't run tRD 5/CL 3 in the past, but can now??
Click to expand...
only OCZ was capble of tRD 5, Gskill all 3 sets of mine can't handle tRD 5 or cas 3 higher than 400mhz on P5W DH or D975XBX

gskill i have are high mhz capable of 563mhz 4-4-4-8 at 2.54v just not good for tight timings, tight tRD heh
 
eva2000 said:
thanks guys

for intel 975x chipset 1:1 is optimal for cpu:mem any extra bandwidth only helps in bandwidth intensive apps like super pi

with most recent games they're more GPU and/or CPU intensive so probably best to have tight timings 1:1 high clocks anyway


only OCZ was capble of tRD 5, Gskill all 3 sets of mine can't handle tRD 5 or cas 3 higher than 400mhz on P5W DH or D975XBX

gskill i have are high mhz capable of 563mhz 4-4-4-8 at 2.54v just not good for tight timings, tight tRD heh
Click to expand...
tRD 7 at CL4? Can you try out 6?
 
well these are meant to be untweaked oc from bios to windows tests first... memset tweaks come later heh

the OCZ PC8000 Titanium Alpha VX2 can handle 500mhz 4-4-3-4 at tRD 6 but Gskill are only handle 500mhz 4-4-4-4 at tRD 7 for badaxe board with 2:3 divider

i.e OCZ @500mhz 4-4-3-4 but memset tweaked







AC3421 said:
how does he have a quad?
Click to expand...

bought it off a special cpu fairy :)
 
Last edited:
Eva, what happens when you take the 3-3-3-x down by say 5 MHz, in other words, at what settings are both kits dead even? Could you also re-run the 2M test? Any special reason why it is an equal score?

BTW, congrats to you Aussies on winning the Champions Trophy. :)
 
thanks Super Nade...

no special reason for 2M time to be same.. just happen to end up that way...

To be more thorough, I also tested OCZ PC2-8000 Titanium Alpha VX2 at same 500mhz 4-4-4-8 timings and default bios boot up as the Gskill PC6400PHU2 HZ and compared them to the existing results.

cpuz_validated_tn.png


Gskill PC6400PHU2 HZ D9GMH @500mhz 4-4-4-8 (left)
vs
OCZ PC2-8000 Titanium Alpha VX2 @417mhz 3-3-3-8 (middle)
vs
OCZ PC2-8000 Titanium Alpha VX @500mhz 4-4-4-8 (right)

1M
16.907s vs 16.812s vs 16.766s

2M
40.750s vs 40.750s vs 40.578s

4M
1m 34.454s vs 1m 34.000s vs 1m 33.609s

8M
3m 31.156s vs 3m 31.157s vs 3m 30.218s

16M
7m 44.281s vs 7m 44.141s vs 7m 42.282s

32M
16m 48.438s vs 16m 47.562s vs 16m 43.828s

Plotting the graph two different ways shows:

d9gkx_d9gmh.png


d9gkx_d9gmh_2.png


Looks like the Micron D9GKX based OCZ PC2-8000 Titanium Alpha VX2 is just all round faster than the Micron D9GMH based Gskill PC6400PHU2 HZ 0629 batch memory for both 4:5 417mhz 3-3-3-8 and 2:3 500mhz 4-4-4-8. The difference is small, but for those Super Pi junkies like me, it looks like interesting times a head for my Super Pi adventures with this Micron D9GKX based OCZ memory :)

But as folks will ultimately also factor in price, the Gskill PC6400PHU2 HZ Micron D9GMH are probably a better bang for your buck being on average 35-45% cheaper in Australia :)
 
eva, what does memset show with regards to the minor timings? What I'm trying to get at is, if ALL things being equal, will the GMX beat out the GMH? :)
 
How does this relate to other brands of memory? Should we expect then, that a tighter timing setup will be faster (although minutely) than a higher frequency setup.


On a personal anecdote, I did have an epxerience with this. I recently got my new mobo in and went about tweaking to get the most effective overclock. I found that my memory (with a divider of course) performed better at the 1M superpi at 230mhz with 2-3-3-8 rather than 3-3-3-8 at maybe 240 or 250mhz.
 
Yep superpi likes tight timings, although I would want to look at the subtimings also becuase GKX is binned higher which could make a difference comapared to GMH.

If I get a chance i'll try this on AM2 and see what happens since it almost seems more FSB dependent.
 
Very nice Eva. You just don't see people doing low latency comparisions with DDR2 stuff like they did back in the day with TCCD's and BH/UTT IC's.

Thanx for posting :thup:
 
good stuff! i know you said you're not really into the whole gaming benchmark thing... but any chance of seeing some kind of "real-world" app... maybe some A/V encoding? thanks for posting this... interesting stuff!

edit: just read your other post... lol, sorry didn't mean to be whinging about benchmarks, it's cool... hope you can break into the top 10 pi's!

Super Nade said:
BTW, congrats to you Aussies on winning the Champions Trophy. :)
Click to expand...

:cool: thanks mate! :thup:
 
Last edited:
You must log in or register to reply here.

Similar threads

QuadFusion
Prime95 Stable, but Super Pi Fails 32M test in Vista [Solution inside]
Replies
19
Views
27K
AndreaACME
AndreaACME
G
Memory bandwidth tests... any real differences (PC4300 vs. PC7100)
2
Replies
39
Views
7K
deeppow
deeppow
afireinside
AM2 and memory timings: 3-3-3-8 vs 4-4-4-12 1T and 2T in super pi 32M
Replies
7
Views
1K
afireinside
afireinside
eva2000
Asus P5K Deluxe - Important tip = Black dimm slots > Yellow dimm slots
Replies
9
Views
4K
johan851
johan851
Back