TerraRaptor
-
Posts
1929 -
Joined
-
Last visited
-
Days Won
44
Content Type
Profiles
Forums
Events
Blogs
Posts posted by TerraRaptor
-
-
Good score Digg. Did you test it on REX, what was your max 32m fsb with it?
-
Hey bro, no subtest window here
-
-
Quite efficient I must say
-
I was testing 1GB version (samsung GDDR3 HC12), memory won't do specs no matter what (just 700mhz stable lol). I also have a ddr3 1GB card that clocks 1150mhz with 1.5v (but core is cut down version with just 8sp/4tmu). So gddr3 is not a thing, one needs a lucky combo of chip, memory, pcb and bios.
-
Impressive. Chips are samsung?
-
Use standard vga driver and disable hw acceleration in display properties?
-
With earlier generations of nvidia, bugs like this heavily depend on vram manufacturer. With 7300/7600gs the most buggy were samsung chips while hynix tend to freeze system completely.
-
Oh no:) Very nice, my friend:)
-
Usually, gt tests scale quite the same percentage. So, as an example, if all people get gt1 as 1fps and gt2 as 2fps (so gt 2 is x2 faster) and you get your gt1 as 1fps (like all people do) and gt2 as 3fps (so x1.5 times faster than others and x3 times faster than your gt1) that simple math will expose a bugged run with gt2. Overclocking usually gains same or similar performance gains - for above example, 1.1fps and 2.2fps (+10%) is expected score, and if you get 1.1fps and 2.6fps (so +10% for gt1 and +30% for gt2) then you should revalidate your score.
Don't think you've explored something very new. So called black screen runs are known for decades. You should use common sense first of all. If image is deadly corrupted to bring you extra fps then rebench. Artefacts itself are still okay and may happen even with stock clocks sometimes but not when it breaks the scene completely.
-
40 minutes ago, CL3P20 said:
for each core
No way. GTL0/1 are adress and data reference levels for die0, gtl2/3 are address and data reference levels for die1.
Also, wprime thread synchronization has very small relation to GTL tweaking (just in case of severe instability one of the threads may be delayed to fix errors), wprime likes to desynchronize in general if OS is bad. If you want to track down GTL tweaking results, better way is using hyperpi 32m - once a thread stops with error just go to task manager and check the affinity of superpi thread that stopped working.
-
Exploiting artefacts/bugs is not allowed. Score should be removed.
-
Great, great, great
-
Quite efficient, respect!
-
Doesn't look ghetto-style:) Should be enough for cpus like i3 3220 etc.
Will wait for performance numbers.
-
-
Was waiting for your score:) I will need LN2 to improve (but this likely not gonna happen so I'm out of competition:)).
Great AGP WR!
-
1
-
-
So, the limit is definitely on cpu side?
-
Better now
-
-
8 hours ago, Zamin said:
It not related my Mobo Which i can't do pl5?
https://www.anandtech.com/show/2427
-
1
-
-
You need cl6 elpida hypers to run pl5 with strap200
-
you likely need to adjust dram skews for dual channel. Try changing CLK skews - advance +350ps for chA (a2) and normal/-25/-50/-75/-100ps delay for chB (b2) (needs pricise adjustment). Also, may be low voltage issue for NB with 2 channels active - what vnb you use?
-
Set dynamic write enable first.
Digg_de - Core 2 Quad Q9650 @ 4950MHz - 9min 18sec 109ms SuperPi - 32M
in Result Discussions
Posted
Just to compare. I also think it's not a good idea to skip attempts to teach REX to OC quads:)