9800 GX2 Benchmark is out!

[bunnyfubbles]

Originally posted by: nRollo

Originally posted by: reviewhunter
<3DMark06>

8800 GTX: 7378
8800 Ultra: 8154
9800 GX2: 9984

http://lly316.blogspot.com/200...800-gx2-benchmark.html

3DMark only. Given that 3DMark06 scores at 12X10 have very little to do with actual gaming performance, I wish people would stop using it in reviews. (although this really isn't a review)

but that didn't stop you from creating a thread when nVidia topped the charts in it

 

[ttnuagadam]

this isnt the kind of performance jump i was hoping for. Nvidia should have a single chip solution thats faster than this.

 

[Keysplayr]

Originally posted by: apoppin

Originally posted by: reviewhunter

Originally posted by: apoppin

not necessarily .. your examples compare video cards of the SAME generation - which why i asked for a comparison to 9600GT ... i don't think it is valid comparing different generations of even nvidia GPUs with 3DMarkXX.

Same generation?
Then they should be comparing against 8800GTS 512, since they are both based on G92.
I don't see any reason bringing in the mid-range G94-based 9600GT in.

OK, i will agree with that .. my point remains - they should NOT be comparing with G80

Why not? G80 is still the nvidia card to beat. Regardless of it's generation, it still is the top card. "Ultra is anyway". GTS 512 does a fine job against the GTX for the most part, but it still falls a bit short.

 

[apoppin]

Originally posted by: keysplayr2003

Originally posted by: apoppin

Originally posted by: reviewhunter

Originally posted by: apoppin

not necessarily .. your examples compare video cards of the SAME generation - which why i asked for a comparison to 9600GT ... i don't think it is valid comparing different generations of even nvidia GPUs with 3DMarkXX.

Same generation?
Then they should be comparing against 8800GTS 512, since they are both based on G92.
I don't see any reason bringing in the mid-range G94-based 9600GT in.

OK, i will agree with that .. my point remains - they should NOT be comparing with G80

Why not? G80 is still the nvidia card to beat. Regardless of it's generation, it still is the top card. "Ultra is anyway". GTS 512 does a fine job against the GTX for the most part, but it still falls a bit short.

no ... that is absolutely NOT what i mean

it seems i am doing a lot of "explaining " recently ... i must not be communicating my thought very well

ANYWAY, what i meant was that you cannot really "compare" 3DMark06 Scores of a G80 and a G92 to get a relative "real world" performance difference. Now IF they had used G92 GTS-512, they would be comparing the *same* [or a very similar] core and we could tell a LOT more

it is almost the SAME as benching it against 3870x2 and only using 3DMark ... you will not get a realistic idea of how they perform in games.

 

[shabby]

Is the gx2 going to need an sli mobo by any chance?

 

[apoppin]

Originally posted by: shabby
Is the gx2 going to need an sli mobo by any chance?

nope ... and i want one
-it would be going in my Crossfire MB .. 1-PCIeX16 slot required ... the other 4xPCIe slot can be empty for awhile

 

[reviewhunter]

that's the good thing about dual-core cards with built in SLI-bridge.

 

[BFG10K]

but it does give you a very clear idea about how cards scale: when you compare a 8800 UlTRA against 8800GT, or 3870 VS 3850, the difference you see in 3dmark is the difference you get in actualy games.

I don't think this is necessarily true.

 

[apoppin]

Originally posted by: BFG10K

but it does give you a very clear idea about how cards scale: when you compare a 8800 UlTRA against 8800GT, or 3870 VS 3850, the difference you see in 3dmark is the difference you get in actualy games.

I don't think this is necessarily true.

as a blanket statement you are right. 3DMark is not representative of many game's Real world performance

However, in the same rig, comparing a g80 8800Ultra to a g80 8800GTX in 3DMark will give you an 'idea' of relative performance in many modern games ... sometimes not .. it is 'rough'

in other words, comparing same gen CPUs ... i.e. 3870 vs. 3850 ... using 3DMarkXX .. is a LOT more realistic than [say] comparing HD3870 to 8800GT [which gives us absolutely 'nothing'] even in the same rig.

 

[Griswold]

Originally posted by: jaredpace
Speichertakt - lol, memory in German.

Actually, thats german for memory frequency / memory speed.

 

[massappeal]

I just want to see this card up against the new hd3870 x2 with GDDR4 and see who wins in terms of performance.

 

[reviewhunter]

Originally posted by: Griswold

Originally posted by: jaredpace
Speichertakt - lol, memory in German.

Actually, thats german for memory frequency / memory speed.

google translator: memory clock
yahoo babel fish: storing act

 

[bryanW1995]

Originally posted by: massappeal
I just want to see this card up against the new hd3870 x2 with GDDR4 and see who wins in terms of performance.

I don't understand why so many people are gung-ho about the gddr4 hd3870. 25% more memory bandwidth isn't nearly as important as 25% improvement in core speed. I'll tell you what, I'm using a 3870. It's currently at 850 core/2412 memory according to gpu-z v 0.1.7. That's 9.4% core overclock. I'm using catalyst 8.3. I need to run the memory at 2463 to get a 9.4% memory oc. I'll run 3dmark 06 and crysis demo with 9.4% oc on the core and no oc on the memory, then switch to 9.4% oc on memory and no oc on the core. If the core oc shows a similar improvement to the memory oc then I'll start to get a little excited about the gddr4 3870x2. If not, then at least all of you out there slobbering about it will have a better idea of what it can really do. I'll start right now and post my results when it's done.

 

[nullpointerus]

Originally posted by: reviewhunter

Originally posted by: Griswold

Originally posted by: jaredpace
Speichertakt - lol, memory in German.

Actually, thats german for memory frequency / memory speed.

google translator: memory clock
yahoo babble fish: storing act

Fixed.

 

[massappeal]

Originally posted by: bryanW1995

Originally posted by: massappeal
I just want to see this card up against the new hd3870 x2 with GDDR4 and see who wins in terms of performance.

I don't understand why so many people are gung-ho about the gddr4 hd3870. 25% more memory bandwidth isn't nearly as important as 25% improvement in core speed. I'll tell you what, I'm using a 3870. It's currently at 850 core/2412 memory according to gpu-z v 0.1.7. That's 9.4% core overclock. I'm using catalyst 8.3. I need to run the memory at 2463 to get a 9.4% memory oc. I'll run 3dmark 06 and crysis demo with 9.4% oc on the core and no oc on the memory, then switch to 9.4% oc on memory and no oc on the core. If the core oc shows a similar improvement to the memory oc then I'll start to get a little excited about the gddr4 3870x2. If not, then at least all of you out there slobbering about it will have a better idea of what it can really do. I'll start right now and post my results when it's done.

I cant wait, this will clear up any ambiguous perceptions. I appreciate you doing that.

 

[bryanW1995]

thank god my "fileserver" is right next to main rig. I don't think I could sit around for an hour or so and watch all these demos run again..

btw, I had to go down to 8 % oc. 3dmark 06 wouldn't run at a 9.4% memory oc . I've noticed that the past few cat releases have lowered my oc...bastards...

anyway, I'm doing the test at 2430 memory (default is actually 2250 so 8% oc exactly). core is 837 now instead of 850 (vs 776 default) for a 7.86% oc. curently almost done with the memory oc testing...halfway there!

on crysis I took the middle number from the last 3 demo runs. On all 3 video card settings these last 3 demo runs were within 8/100 of a second avg fps.

ok, all done. here's the data:

rig in sig with e6750 @3.4 on xp home
stock 3870 @776/2250

3dmark06 11171
crysis demo (14x9,all high, noAA/AF) 31.45



3870 @776/2430 (exactly 8% memory overclock)

3dmark06 11397
crysis demo 31.92



3870 @837/2250

3dmark06 11611
crysis demo 33.33


ok, so it's obvious that the core OC of 7.84% had a much greater impact than the mem oc of 8%. There was a 5.98% improvement in crysis and a 3.94% improvement in 3dmark with the core OC of 7.84 %. There was a 2.02% improvement in 3dmark 06 with the 8% mem oc and a 1.49% improvement in the crysis demo. how is that for scaling purposes?

core oc in 3dmark: 3.94/7.84 = 50.3% scaling.
core oc in crysis: 5.98/7.84 = 76.28% scaling.

mem oc in 3dmark: 2.02/8.00 = 25.25 % scaling
mem oc in crysis: 1.49/8.00 = 18.63 % scaling

according to this, the core scales twice as well in 3d mark and over 4 TIMES as well in crysis. Also, even using the higher scaling % of 25.25 from 3dmark06, a 25% memory increase would only net a 6.31% increase in performance.

Maybe someone with a 3870x2 can try to verify these tests for us with an 8% memory oc on his card?

 

[Avalon]

Sounds about right Bryan. I remember doing a similar test on a Radeon 9500 way back in the day, and found that a purely memory OC was half as effective as a purely core OC of the same amount. I'd expect the card to get an 8-10% performance bump by using GDDR4 with 25% faster mem clocks.

 

[reviewhunter]

the extra performance with gddr4 might just be enough to stop the 98GX2 from doing its damage

 

[taltamir]

if you are buying a 500$ card... or TWO 500$ cards for a quad core setup... every 6% counts...

The budget 3870 or 3850 should be DDR3, but there is no reason to make the top end slower.

 

[reviewhunter]

this is confirmed,
the 9800 GX2 is retailing at slightly less than US$600

 

[apoppin]

Originally posted by: reviewhunter
this is confirmed,
the 9800 GX2 is retailing at slightly less than US$600

i guess it is OK for a "status" card .. another 7800GTX-512MB ... if you really want one, you should also probably buy it the instant it is released

i am glad that $679 price is "just Frys" .. although, they will probably be $1,000 on ebay the week after ..
. . .and $200 in 3 months
:roll:

good luck with that!

 

[imported_dinGLeBeRRy]

Originally posted by: apoppin

Originally posted by: reviewhunter
this is confirmed,
the 9800 GX2 is retailing at slightly less than US$600

i guess it is OK for a "status" card .. another 7800GTX-512MB ... if you really want one, you should also probably buy it the instant it is released

i am glad that $679 price is "just Frys" .. although, they will probably be $1,000 on ebay the week after ..
. . .and $200 in 3 months
:roll:

good luck with that!

typo?

cant find the justification to pick one of those up now over a newer gen card. :T

 

[reviewhunter]

Originally posted by: dinGLeBeRRy
typo?

cant find the justification to pick one of those up now over a newer gen card. :T

7800gtx-512mb refers to an overclocked version of the card based on the same core.
I think that's what he meant.

 

[betasub]

Originally posted by: reviewhunter

Originally posted by: dinGLeBeRRy
typo?

cant find the justification to pick one of those up now over a newer gen card. :T

7800gtx-512mb refers to an overclocked version of the card based on the same core.
I think that's what he meant.

apoppin's reference to the 7900GTX-512 appears to be about the "status" the card had at the time - as much for its scarcity/price as its performance.

 

[apoppin]

i remember a certain member recommending 7800GTX-512 as an "investment" ... IF i remember it turned out badly ... someone here bought two of them i think
We were *told* it would be a "hard launch" but they were never in supply. They launched back in November '05 ... they were on ebay the next week for over $1,000 used ... no one could get any

-The details escape me ... didn't another much cheaper and faster card come out a really short time later?
---please .. refresh my memory

i having an episode of deja vu