CrossFire Woes - need *Help* 2900xt/2900p

[apoppin]

Originally posted by: keysplayr2003

Originally posted by: apoppin

Originally posted by: keysplayr2003


Actually, you should try both the 2900XT and the 2900pro in each individual slot (single card config) to see if there is any actual performance hit going from a 16X slot to a 4X slot.

there is. It has been tested many times ... about 10-15% practically for a 2900xt/GTS class card and up to 25% in absolute worst case scenarios [for a GTX, i think].

BUT ... you have to remember the 2nd card is bandwidth limited by the slot, not the first card also. And my 2nd card will be an OC'd Pro [if i can get it going]. Now my XT will be made to run at whatever core and RAM speeds i can get with my Pro - the performance hit of running a 256-bit Card with a 512MB card is unknown [to me] ... both cards are 512MB

My *goal* is to get the 2nd card up in Core and memory as high as possible - to equal my XT speeds, if possible
--right now i have a very slow stock 2900p crossfire

it *appears* i can just short both of my 2 extra pins on each card and enable OverDrive
... or OD ... whatever comers first
:Q


i am gonna try it as soon as i find something better than 2 paper clips

Bolded above: If the card in the 4x slot is limited, will overclocking it even help?
That's like trying to pour 2 gallons per minute of water down a drain that only allows 1 gallon per minute. And if I am correct, no amount of overclocking the pro will improve your performance? As you said, the XT will have to substantially lower it's performance to a pro that runs in a 4X slot, no matter what.

Am I on the wrong track here?

apparently so



first note that "stock" Pro CrossFire is at least 10-15% faster than my single 2900xt --just look at my stock Pro/XT CrossFire scores ... then note the performance increase of another +10-15% when it was OC'd from 506Mhz on the core to 697 with also a memory jump of 513>843Mhz

what i did not know until JUST before i went to sleep [at 2AM] is that is that the XT is NOT held back by "modern" crossfire at all - it's core and memory clocks are set *independently* of the 2nd card in the slower slot ... all the 4x slot does is "add" to the 2nd card ... and 256bit vs. 512bit appears to be a minimal impact

so i should get a further 2-3% - we'll see - just by using CCC to OC them separately
:thumbsup:

right at this moment, CCC is using Overdrive to find and set the fastest clocks - independent of each other ... the Pro is not "holding back" the 2900xt - at all
[although to be fair, a 2900xt in the 2nd slot would be faster still - but limited by the 4xPCIe slot a bit more than an OC'd pro]

i am beginning to really like CrossFire
-now i want to see how it does with a 2nd monitor

what *boggles* my mind is that nvidia had a year's Head-start over AMD with multi-GPU ... and yet AMD's CrossFire has surpassed nvidia's SLI - in *every* way

i don't think anyone has to worry about future support for X2 GPU from AMD

CrossFire has finished with my Pro at 686/840 and my XT at 743/828 ... and i am quite sure i can go 10% over those conservative settings
- i settled on 700/840 for the Pro and 752/849 for the XT ... the XT is evidently not held back like in earlier editions of Xfire

Edited/updated

 

[zeroburrito]

Originally posted by: apoppin

Originally posted by: zeroburrito
seems a little low. i have a 2180@3ghz and a single 3870 and score almost 11k. btw, that pro core should reach the xt speeds. its the same thing. infact, it should go 800+. what is the fan speed set to on the pro?

not really ... check out Fudzilla's mix 'n match

http://www.fudzilla.com/index....=view&id=5641&Itemid=1


and WoW! ... i can set the clocks INDEPENDENTLY in ccc
- thanks to Sylvanas for finding that link

whats the processor they are using there? 3dmark06 is HEAVILY dependent on your cpu.. my crappy budget system should not get more marks than crossfire anything. i know 3dmark means almost nothing but crossfire should get over 10k unless something is seriously wrong.

 

[apoppin]

if you don't want to look at 3DMark06's improvement with a single 2900xt from 10.5K to 13.1K with 2900xt/pro Xfire then look at LP/FEAR/HL2 and CoJ's FPS increase ... and i seriously doubt it is my +100Mhz OC on my e4300.
--overall, it's pretty *solid* for an el cheapo upgrade ... imho
-i'm keeping it

BtW, Keys, you are probably right .. my extra OC on the Pro/XT did nothing for my 3DMark score ... i am at the point of "diminishing returns" with my 256-bit limit and the restricted bandwidth of the 4x slot
*However* it is like you taking your GTS-640 and spending $150 8 months later to get ultra performance ... i am evidently subscribed to AMD "step-up" program.
--so i have a 'matched system' again ... from the PS to the LCD


and i am using 3DMark to TRACK changes in my system
-that IS what is really for

[not really "really" - it's an 'e-pen thing']

 

[zeroburrito]

its just that 3dmark06 is super sensitive to quad core/sli setups. i guess 10.5 to 13.1 is ok, but 10,300 for a 3870 crossfire? something is seriously wrong there. maybe they were using the old p4's or the athlon xp.

 

[Skunkwourk]

Originally posted by: apoppin

as it turned out, this solution was to "fool" CCC into over clocking with 2 alligator clips and some wire ... unreal

Pics of what you did?

I don't quite understand how all this worked out, but it sounds trippy.

 

[zeroburrito]

its just a safety precaution for overclocking. it wants to make sure you have it all plugged in the way you're supposed to. if it doesn't need it, just fool it and it will work. if it did need it, it wouldn't work. just a precaution.

 

[apoppin]

Originally posted by: zeroburrito
its just that 3dmark06 is super sensitive to quad core/sli setups. i guess 10.5 to 13.1 is ok, but 10,300 for a 3870 crossfire? something is seriously wrong there. maybe they were using the old p4's or the athlon xp.

10.3 for a very mismatched Xfire ... evidently ...
their scores are VERY discouraging ... compared to mine [using Vista64 also]

look at a single 3870 in 3DMark06 - 9478 ... but with Xfire, their best mismatched score was only 10371, so their CPU must be slow-ass

otoh, they did get a decent increase in games ... showing 3DMark06 weighs the "CPU score" a bit too much for a 3D Benchmark

oto[other]h, i went from 10.5 to 31.1 K showing my CPU is a lot faster AND my GPUs are probably a better match - even with the 4x limitation in the 2nd PCIe slot

i do better much "reviews" then they do i am supposed to "guess' what chipset they are using by looking at their silly photos.
:roll:



==============

Originally posted by: m0mentary

Originally posted by: apoppin

as it turned out, this solution was to "fool" CCC into over clocking with 2 alligator clips and some wire ... unreal

Pics of what you did?

I don't quite understand how all this worked out, but it sounds trippy.

look back at the first page ... there are links to good photos - 3 links posted in a row [edited and added:]

http://www.xtremesystems.org/f...howthread.php?t=155646

more ... i struck gold

http://www.tbreak.com/forums/showthread.php?t=39077

or ... for something completely different:

http://www.xtremesystems.org/f...howthread.php?t=155646

i used the 2 clips and just grounded them to a molex red cable
:Q


BLACK !!!!!!!!!!!!

 

[lavaheadache]

excellent work Mark. Excellent thread!

 

[Keysplayr]

Originally posted by: apoppin
if you don't want to look at 3DMark06's improvement with a single 2900xt from 10.5K to 13.1K with 2900xt/pro Xfire then look at LP/FEAR/HL2 and CoJ's FPS increase ... and i seriously doubt it is my +100Mhz OC on my e4300.
--overall, it's pretty *solid* for an el cheapo upgrade ... imho
-i'm keeping it

BtW, Keys, you are probably right .. my extra OC on the Pro/XT did nothing for my 3DMark score ... i am at the point of "diminishing returns" with my 256-bit limit and the restricted bandwidth of the 4x slot
*However* it is like you taking your GTS-640 and spending $150 8 months later to get ultra performance ... i am evidently subscribed to AMD "step-up" program.
--so i have a 'matched system' again ... from the PS to the LCD


and i am using 3DMark to TRACK changes in my system
-that IS what is really for

[not really "really" - it's an 'e-pen thing']

Oh, I wasn't even concerned with what you paid. That makes no real difference in what I was trying to find out. I was just trying to figure out what your performance "should" have been had you not been restricted by the second PCI-e slot, in comparison with your current situation. I was also wondering if you would not have been better off just selling your 2900XT, and the money you spent on the pro added to that, you could have gotten a GTX I would think. Or a GTS G92.

I know you wanted to crossfire, but I am just thinking out loud about how you could have traded a bit of coolness, for a bit more functionality and money's worth.

i know your happy with what you did, so that's all that matters.

 

[apoppin]

Originally posted by: keysplayr2003

Originally posted by: apoppin
if you don't want to look at 3DMark06's improvement with a single 2900xt from 10.5K to 13.1K with 2900xt/pro Xfire then look at LP/FEAR/HL2 and CoJ's FPS increase ... and i seriously doubt it is my +100Mhz OC on my e4300.
--overall, it's pretty *solid* for an el cheapo upgrade ... imho
-i'm keeping it

BtW, Keys, you are probably right .. my extra OC on the Pro/XT did nothing for my 3DMark score ... i am at the point of "diminishing returns" with my 256-bit limit and the restricted bandwidth of the 4x slot
*However* it is like you taking your GTS-640 and spending $150 8 months later to get ultra performance ... i am evidently subscribed to AMD "step-up" program.
--so i have a 'matched system' again ... from the PS to the LCD


and i am using 3DMark to TRACK changes in my system
-that IS what is really for

[not really "really" - it's an 'e-pen thing']

Oh, I wasn't even concerned with what you paid. That makes no real difference in what I was trying to find out. I was just trying to figure out what your performance "should" have been had you not been restricted by the second PCI-e slot, in comparison with your current situation. I was also wondering if you would not have been better off just selling your 2900XT, and the money you spent on the pro added to that, you could have gotten a GTX I would think. Or a GTS G92.

I know you wanted to crossfire, but I am just thinking out loud about how you could have traded a bit of coolness, for a bit more functionality and money's worth.

i know your happy with what you did, so that's all that matters.

i had to LoL ... forget "happy" ... i am ecstatic ... the damn thing works and it surpasses my expectations.


You cannot separate "apoppin" from "price" ... it is my nature. ... letsee ... what my performance "should have been" is *impossible* because of my choice of MB.

But id you are asking what performance hit i am taking *overall* - being limited by my 2900p in a 4x slot - compared to another 2900xt in a 2nd 16x PCIe slot - about 15% over all is my guess. But then i would be buying a more expensive and much more power-hungry card.

As to selling my 2900xt ... $200 ... and spending $350 more for an "ultra" or $250 more for a highly OC'd GTX ... you're right, i'd get the SAME - or slightly less overall performance than i get right now - for only $150 with my pro. Of course my electricity bill is up.

Wait ... how is that again? $200+150 = $350 for ???


=============

excellent work Mark. Excellent thread!

thanks ... but as you know, it was that 'fluke' of grounding the pins that made CCC function as it should .... and it functions very well with mismatched cards ,,, my own "step up" program.

 

[Keysplayr]

Originally posted by: apoppin



Wait ... how is that again? $200+150 = $350 for ???

A top of the line o/c'd GTS G92, with cash to spare.

 

[apoppin]

Originally posted by: keysplayr2003

Originally posted by: apoppin



Wait ... how is that again? $200+150 = $350 for ???

A top of the line o/c'd GTS G92, with cash to spare.

you really think GTS92 would be faster?
i doubt it ... it isnt faster than even the old ultra

and i seriously doubt i could still get $200 for my 2900xt

 

[Keysplayr]

Originally posted by: apoppin

Originally posted by: keysplayr2003

Originally posted by: apoppin



Wait ... how is that again? $200+150 = $350 for ???

A top of the line o/c'd GTS G92, with cash to spare.

you really think GTS92 would be faster?
i doubt it ... it isnt faster than even the old ultra

and i seriously doubt i could still get $200 for my 2900xt

It's difficult to tell. I mean, a new GTS trades blows with a GTX at stock speeds for both. Crossfired 2900XT's in 2 8X PCI-e slots would most likely beat out a GTX/new GTS. But the way you have your setup, a 2900xt CF with a 2900pro in a 4x slot? I really don't know if it would beat a GTX/new GTS.

And yes, I think you can easily get 200 dollars for your XT. It's still a great card. I could get about 225 to 250 for my GTS 640.

I'm talking of course of ebay. I keep an eye out for a 2nd GTS 640 for myself, and they are still a bit too pricey for me to grab a second one.

 

[apoppin]

unless i am mistaken, the 2900xt trades blows with the [old] GTS-640 still ... on the rarest of occasions in specific games, it even matched the Ultra.
-Generally, your card and mine fall about 20-30% behind an ultra
--G92-GTS is slightly faster then the GTX yet still behind the ultra

IF i got +30% performance increase overall by adding the 2nd GPU, i would say i AM into "ultra territory" - right now - never mind the tweaks i will make later

yes ... no? ... good deal for $150 more dollars?

 

[Keysplayr]

Originally posted by: apoppin
unless i am mistaken, the 2900xt trades blows with the [old] GTS-640 still ... on the rarest of occasions in specific games, it even matched the Ultra.
-Generally, your card and mine fall about 20-30% behind an ultra
--G92-GTS is slightly faster then the GTX yet still behind the ultra

IF i got +30% performance increase overall by adding the 2nd GPU, i would say i AM into "ultra territory" - right now - never mind the tweaks i will make later

yes ... no? ... good deal for $150 more dollars?

That depends. Is your Crossfire setup using AFR?

 

[BFG10K]

IF i got +30% performance increase overall by adding the 2nd GPU, i would say i AM into "ultra territory" - right now - never mind the tweaks i will make later

The 8800 Ultra is often more than 50% faster than a 2900.

 

[apoppin]

Originally posted by: BFG10K

IF i got +30% performance increase overall by adding the 2nd GPU, i would say i AM into "ultra territory" - right now - never mind the tweaks i will make later

The 8800 Ultra is often more than 50% faster than a 2900.

true ... but many times a single 2900xt is a lot closer then that and - rarely - it even beats it ... i did qualify it by saying "in the territory" ... and of course, there are times when Xfire will not scale at all
-it is still good bang for buck - even for a GTX' performance, - or i wouldn't do it at all; remember it was a "crapshoot" with many here posting and predicting it would be an utter failure due to the Pro being slow, 256-bit and in a 4xPCIe slot ... again - in my opinion - the experiment was successful with my 'el cheapo' bastardized mismatched Xfire exceeding my minimum expectations.

i mean ... for my 2900xt and Orange Box, i spent $330 back in May and just did a AMD Step Up for $150 more to get a significant improvement in all the games i have tested so far. DX10 games that were formerly unplayable at my resolution are now open to me.

--and i need to play some more games *tonight* ... to test it further, of course.

That depends. Is your Crossfire setup using AFR?

AFaiK - everything is enabled in CCC ... Why?

 

[superbooga]

Apoppin, you're overstating the performance of the 2900XT. The Ultra is at times 50% or more faster than the G80 GTS and 2900XT.

I'm not saying what you did was dumb ... it was a good cheap upgrade. But you could have sold your 2900XT and bought a G92 GTS for the same cost, and get at least the same, if not better, gaming experience. If you don't think so, I'll send you my G92 GTS and you can use it for a week.

 

[apoppin]

Originally posted by: superbooga
Apoppin, you're overstating the performance of the 2900XT. The Ultra is at times 50% or more faster than the G80 GTS and 2900XT.

I'm not saying what you did was dumb ... it was a good cheap upgrade. But you could have sold your 2900XT and bought a G92 GTS for the same cost, and get at least the same, if not better, gaming experience. If you don't think so, I'll send you my G92 GTS and you can use it for a week.

no i don't think so ... really

and ...

sure ... i'd *love* that
--but i bet you don't really mean it
:brokenheart:

Could you please - cheaper/easier/instead - link to some recent benches that compare 2900xt with ultra ... i remember the 2900xt was also recently compared to the 3750 and GTS and maybe ultra or GTX
?
where


oh, i am back on Vista 32 ... CCC decided that i would like 757/850 for my XT and 700/850 for my Pro

i am curious about FEAR ,,, i never experienced it ... but i did finish the game.

back in a few

 

[zeroburrito]

2900xt is just as fast as the 3870. even with his odd setup it should still be faster than a g92 gts. the g92 gts is about 10% faster than the 8800gt. 2900xt crossfire with anything would be faster than that. you're overstating the performance of the gts.

 

[Keysplayr]

Originally posted by: superbooga
Apoppin, you're overstating the performance of the 2900XT. The Ultra is at times 50% or more faster than the G80 GTS and 2900XT.

I'm not saying what you did was dumb ... it was a good cheap upgrade. But you could have sold your 2900XT and bought a G92 GTS for the same cost, and get at least the same, if not better, gaming experience. If you don't think so, I'll send you my G92 GTS and you can use it for a week.

:shocked:

 

[MegaWorks]

Here's a review.

It's an 3870 X2 review but it has two 3870 in action, I would say it's about the same as your setup.

 

[Sylvanas]

The 2900XT is about the same if not a little better than the 3870 (in UE3 based games) so if the G92GTS is about 10% overall faster than the 3870, then its also 10 quicker than the 2900XT, TWO 2900XT's in Crossfire is alot better than what you'd get from any ultra, GTS or GTX.

 

[apoppin]

Originally posted by: MegaWorks
Here's a review.

It's an 3870 X2 review but it has two 3870 in action, I would say it's about the same as your setup.

that is NOT a useful review as 3870 is not the same as 2900xt ,,, there was a recent one with a 2900xt

But let me explain further ,,, and Vista 32 continues to be faster than 64-bit on my rig with Crossfire

Lost Planet: Extreme Conditions - full retail game built-in demo. DX10/everything fully maxed in-game/1680x1050/4xAA-16xAF
Xfire -------- Snow - 30.8 / Cave 30.2 [Vista 32]
Xfire ---- Snow - 30.0 / Cave 29.0 [Vista 64]
1-2900xt -Snow - 19.6 / Cave 28.1 [Vista64]

with a single GPU the frames would bottom at 10 or 11 ... now it is 15-17 for much shorter dips ... i'd say i have much better than +30% in this case ... an average FPS of 19.6 > 30.0


==============

F.E.A.R. built-in Demo

16x10 everything maxed 0xAA/16xAF - SS on

Pro/Xt --- Vista 32 -- 35 Min/76 Avg/120 Max
Pro/xt - Vista 64 - 32 Min/70 Avg/113 Max
1-2900xt [64bit] - 30 Min/59 Avg/112 Max

a solid increase on the average, and a bit on the bottom and top ... but not so impressive as LP

=============


Vista32 HL2/Lost Coast ... everything Maxed at 16x10 - 4xMSAA/16XAF ... 3 Runs with Fraps
Min, Max, Avg
2900xt - 62, 226, 106.322
Xfire --- 68, 283, 141.051

pretty impressive scaling, wouldn't you say?

i think i'f give an ultra a run on my rig .... but how can we know for sure?
--send me your GTS

 

[superbooga]

There are certain games that the 2900xt and HD 38xx excel at, but the 8800 cards still give you the most consistent performance across all games. Is it NV's developer relations, good architecture, or both? Who knows. But the games that are the most taxing are the ones the 8800 does the best at. For older games I expect a dual-GPU from ATI to beat a single high end Nvidia GPU, but does 90 vs 110 fps really matter that much?

As yes, I am willing to send my GTS.