so here it is, my OC on stock voltage with water (albeit warm water :D)
http://dillusion.net/pics/x1605.jpg
if you cant see, thats 658/819(1638). I would like to know how to raise the core voltage thru ati tool, i dont see the option anywhere.
Printable View
so here it is, my OC on stock voltage with water (albeit warm water :D)
http://dillusion.net/pics/x1605.jpg
if you cant see, thats 658/819(1638). I would like to know how to raise the core voltage thru ati tool, i dont see the option anywhere.
Nvidia just got pwned. The days of the 9800 are BACK!Quote:
Originally Posted by Lightman
The X1k series were just dumies to see how well the new architecture works out. Preforms well with little pipelines and now it's releasing so much pipelines WHEW
not bad, though I expected 6k+ you got 5.11?Quote:
Originally Posted by Dillusion
5.13Quote:
Originally Posted by sabrewolf732
hmmmm now lets speculate a bit about r580 and x1700 then :DQuote:
Originally Posted by sabrewolf732
x1300pro = 4x1, x1600xt = 4x3 right?
and a x1800xt is 16x1 and a 1300pro is 4x1 right?
we know that (more or less):
1800xt/r580=1300pro/1600XT
and can now guess that (more or less):
r580=?x1800xt performence wise
but since the 1600 vs the 1300 compare is complicated as they have the same amount of max threads, and have a similar design in some aspects but a different in others, like 2vs 5 texture units and 12gb/s vs 22gb/s bandwidth, lets not focus on this now as its complicated to compare the two and get a reliable result from it.
we also know other things wich make it way easier to speculate about r580 and 1700 performence :D
1300 vs 1800
core clock
1300pro is at 600
1800xt is at 625
memory bandwidth
1300 has 12gb/s
1800xt has 48gb/s
vertex processors
1300 has 2
1800 has 8
texture units
1300 has 4
1800 has 16
render back ends
1300 has 4
1800 has 16
zcompare
1300 has 4
1800 has 16
max threads
1300 has 128
1800 has 512
so the core clock is almost identical, the bandwidth/quads or pixel processors and all other aspects of the 1800 is exactly 4 times a 1300pro.
so a 1800xt is 400% a 1300 hardware and bandwidth wise.
now lets look at the performence:
2003
5853 vs 16359 1800xt=280%1300pro
2005
2786 vs 8929 1800xt=320%1300pro
riddick 10x7
33.2 vs 98.1 1800xt=295%1300pro
so a 1800 performs around 3x as good as a 1300 when it comes to raw performence.
we know that (more or less):
1. 1800=4x1300pro hardware wise
2. 1800=3x1300pro performence wise
3. 1700xt=2x1600xt hardware wise
4. r580=4x1600xt hardware wise/r580=2x1700xt hardware wise
and can now guess that (more or less):
1. 1700xt=1.5x1600xt performence wise
2. r580=3x1600xt performence wise/r580=2x1700xt performence wise
this only applies IF the 1700xt and r580 have the same relative bandwidth/quad and pixel pipeline and clockspeeds as the 1600xt, and the new way of organizing the pixel and vertex processors etc in the vpu scales the same from increasing the quads as it does from the 1300 to the 1800. the first one is unlikely but more about that later, the latter is very likely, and even if it scales differently it should scale similar so our guesstimations wouldnt be that far off.
1600xt vs 1700xt (speculation)
2003
9112 vs 13668
2005
5209 vs 7813
riddick 10x7
63.4 vs 95.1
1600xt vs r580 (speculation)
2003
9112 vs 27336
2005
5209 vs 15626
riddick 10x7
63.4 vs 190.2
to perform like this the 1700xt would need the following specs:
8x3 vpu design
24 pixel processors
10 vertex processors
8 texture units
8 render back ends
16 z compare units
256 max threads
590mhz core
44gb/s bandwidth =~1400mhz memory clocks
to perform like this the r580 would need the following specs:
16x3 vpu design
48 pixel processors
20 vertex processors
16 texture units
16 render back ends
32 z compare units
512 max threads
625 mhz core
88gb/s bandwidth =~2800mhz memory clocks
i doubt a mid end card will have 1400mhz memory on it, but its possible without making the card too expensive to be competitive, but 2800mhz memory wont be possible even with gddr4... maybe oced, but i highly doubt there will be 2800mhz memory. lets say they manage to get memory at around 2000mhz, wich is realistic if you ask me. that means they only get around 60gb/s bandwidth. i havent seen any results of how the x1000s perform with less bandwidth and how much it kills performence, but i dont think it will have a big impact. lets not forget that it will need 2x as much cpu power as well, so id say the 1700xt should be good for 12K in 2k3 and 7k in 2k5 wich is a little pessimistic, but i dont wanna create any false hopes, and hardware does, unfortunately :D, not scale linerily, so its realistic i think.
r580 should be good for 23k in 2k3 and 13k in 2k5
then again, the 1600 and 1300 and 1800 results i used are from a 2.4ghz a64, by the time the r580/1700xt are out they will be tested with a64 x2 and intel dual cores wich should offer more cpu power, plus ati might tweak the design some more and up the clockspeeds, so r580 might do 25K in 2k3 and 15K in 2k5... maybe thats even what ati set as a goal for themselves... who knows.
this is all speculation based on the assumption that r580 is 4x a 1600xt hardware wise and 1700xt is 2x a 1600xt hardware wise, wich is what everybody seems to say/believe.
very nice analysis. According to you, r580 is gonna pwn. :woot:
lol, your kidding right?Quote:
Originally Posted by bassv2
nvidia didnt get pwned, ati got pwned when nvidia released their 7800gtx 512mb
now lets see what atis next cards perform like and how nvidias answer to those will perform befor we jump to any conclusions ;)
and the 1k series were just dummies to see how the architecture works?
i dont think so. :D
sabrewolf732, what cpu speeds for your scores?
I think you mean dillusion ;) And the x1800xt is much closer to the 7800gtx with the new driver series. Also, seems to be about even in crossfire as well.Quote:
Originally Posted by saaya
of course, the more vpu power you add, aka sli/crossfire, the more the cpu becomes the limiting factor, and the difference between two cards that perform differently grows smaller and smaller ;)Quote:
Originally Posted by sabrewolf732
and yes, the 1800xt is on par with the 7800gtx, but the 7800gtx 512mb beats it silly :D its barely available :rolleyes: but its faster than the 1800xt. check out the review marcus made, comparing the 7800gtx 512mb and 1800xt voltmodded and with dry ice and even ln2 i think.
the 7800gts had a cold bug somehow and couldnt run below -20°C and he didnt have time to do a vmem mod, and still the 7800gtx beat the xt!
it got 12K and wasnt maxed out and the 1800xt only 11k and it was moxed out vmod and cooling wise...
so what cpu speed did you run that 1600 with?
dillusion what was the cpu speed
thats a pretty slow x1800xt for that kind of cooling :stick:Quote:
Originally Posted by saaya
Me. Maybe i'm just unlucky and have to deal with a lot of crappy sapphire stuff? :(Quote:
Originally Posted by Saaya
You guys might try comparing fillrate scores between the x1300 and x1600 series.. x1800 too if you can. There is a program called rendertree or something (forget exaclty, dont have it on hand) that tests some stuff like fillrate very well.. might be worth looking into.
actually, the x1800xt beats the gtx pretty easily and comes very close to the 512. Look at new benchies.Quote:
Originally Posted by saaya
my bad, the 1800xt was on air??? :eek:Quote:
Originally Posted by STEvil
hope so :SQuote:
Originally Posted by STEvil
hmmm why should it be interesting?Quote:
Originally Posted by STEvil
everyone's spouting this 4x3, 16x3 stuff like they forgot architectures still work in quads, not triads.
thats what i wondered about as well... so its wrong?Quote:
Originally Posted by Cybercat
gotta read the beyond3d stuff again :D
what it is is the four in 4x3 is the four fragment pipes in each quad. There are three of them.Quote:
Originally Posted by saaya
I think everyone took the claims from ATI that each pipe is three times as efficient, in tandem with the confusion about TMUs versus pipelines, to mean that each fragment is somehow really three fragments, or some messed up explaination like that. It's a big mess, and not a lot of people seem to really get it straight. You can't have 16 fragments in a single quad, since obviously a quad implies only four.
Let me try to clear this up right now. It's 12 shader fragments, or three quads, four texture mapping units, four ROPs. The texture units in the RV530 are sort of enhanced from those in the R520. Each texture mapping unit can serve three fragments at once. So it only takes four to handle all shader units in the RV530. Beyond3D talks about this as well.
For the R580, there are 48 shader fragments, or 12 quads, 16 texture mapping units, and 16 ROPs. Again, each texture mapping unit can address up to three fragments at once, so 16 covers all 48 shader units.
For the rumored RV560's configuration, there 24 shader fragments, which means there must be 8 TMUs to cover them all. The number of ROPs doesn't matter, since there's traditionally a crossbar between them all, but assumably they also work in quads, just like the shader units.
Your core clock should start improving with higher voltages...which should make 6k runs much easier.Quote:
Originally Posted by Dillusion
Perkam
then the voodoo 2 never existed, because it was 1x2 architecture!! omgz!!!Quote:
Originally Posted by Cybercat
:stick: :slap:
;)
They work however the manu builds them.
very funny, comparing a voodoo 2 to current architectures. While we're at it, let's take the original Radeon and Radeon 7000 series, with their 2x3 architecture!Quote:
Originally Posted by STEvil
Good call! :toast:Quote:
Originally Posted by Cybercat
I still think it is more accurate to say: 12 x 1/3 (only 4 texture units). Since when do the texture units get top billing :D , with a set trio of pixel pipes assigned to each?
That fits with the diagrams I was talking about earlier in this thread...shows 3 pipeline quads. Just because they are fragment pipes doesn't mean there aren't 12.Quote:
Originally Posted by Cybercat
270x9 = 2430mhzQuote:
Originally Posted by althes
anyone know how to raise the core voltage?
Anyone know which software to use for voltage adjustments?
hey saaya where you at?
Quote:
Originally Posted by Dillusion
yeap!
look into this thread
http://www.xtremesystems.org/forums/...ad.php?t=83562
enjoy!
and brake some 6k :D