AMD: R7xx Speculation

[Arty]

Going back to the leaked slides from Toms, it does mention 'Tera-scale' graphics engine. That would mean both the 4850 & 4870 should hit the 1TFlop mark ..

 

[Wirmish]

Photoshopping update...

R580 = 111 pixels (mm² = ???)
· Tech Report: R580 = 314.5mm²
· PCPerspective: R580 = 315mm²
· Hexus: R580 = 342mm²
· Beyond3D: R580 = 352mm²
· XBit Labs: R580 = 350mm²
· Nvidia Slides: R580 = 352mm²

RV670 (13.86mm x 13.86mm = 192mm²) = 80 pixels
RV770 = 102 pixels vs 80 pixels = +27,5% x 192mm² = 244.8mm² (15.65mm x 15.65mm)



EDIT: largon (XS) -> RV670 vs RV770 in-scale:

 

[ShaidarHaran]

Going back to the leaked slides from Toms, it does mention 'Tera-scale' graphics engine. That would mean both the 4850 & 4870 should hit the 1TFlop mark ..

Hmm, not necessarily... I mean, it is just marketing

 

[Arty]

I'm just throwing it out there, thinking out loud.

Btw, update on 9800GT. Posting it here since its 4850's direct competitor:

Still offering DirectX 10.0 and PCI-Express 2.0 support, the upcoming card features 112 Stream Processors, just like the 8800 GT, and a 256-bit memory interface.

memory clocked at 1800 MHz and a core set to 630 MHz, very close to what a reference 8800 GT would 'brag' with

Courtesy: TechConnect

 

[ZerazaX]

Well the 4850 could be intentionally made slower by lower bandwidth and limiting factors because ATi seems intent on having it actually slower than the 4870 this time around, and not as close as last time

 

[tacopaco]

I would have thought that since the 4870 can best a 9800 gx2 wouldn't the 4850 be more of a competition for the 9800 gtx? 9800 gt isn't really much of a challenge......
Also they released some low end 9 series mobile chips don't know if thats old news but i looked online, how is a 65nm m8800gtx different from its 9 series counterpart? both would be 512mb with a 256 bit bus and they wouldn't have the clock speed differences their big brothers have because the m8800 is already 65nm....

 

[trinibwoy]

I would have thought that since the 4870 can best a 9800 gx2 wouldn't the 4850 be more of a competition for the 9800 gtx? 9800 gt isn't really much of a challenge......

Well the answer to that question is based on the validity of the first part of your sentence. Where did you get that the 4870 is faster than a 9800GX2? There have been no rumours indicating that even based on Vantage (hope you're not going off those slides some site made up a few months ago).

 

[ShaidarHaran]

I would have thought that since the 4870 can best a 9800 gx2

I believe you are misinformed. No reliable source has ever claimed this. The GT200 will be faster than the 9800 GX2 in most (if not all) circumstances, and the 4780 *X2* should be of similar speed to the GT200.

wouldn't the 4850 be more of a competition for the 9800 gtx? 9800 gt isn't really much of a challenge......

The 4850 is said to be approximately equal to the 9800 GTX in performance.

 

[tacopaco]

I believe you are misinformed. No reliable source has ever claimed this. The GT200 will be faster than the 9800 GX2 in most (if not all) circumstances, and the 4780 *X2* should be of similar speed to the GT200.

i most likely am misinformed it sometimes gets hard to dig through the fakes to get some real info

The 4850 is said to be approximately equal to the 9800 GTX in performance.

that what i mean how would the 9800gt be a comparison?

 

[trinibwoy]

German site pulled together a few scores from various rumours:

In 3DMark06 reached the HD 4850, according to "Fudzilla" a score of 11,760, while the Nvidia card [8800 GT] compared to 10,800 3DMarks came - two values that only by a factor of 1088 separately. At this point not tested 9800 GTX would probably come around 12,500 points.
Similarly, results in the colleagues of "ITOCP" emerged that the HD 4850 P5847 points in 3DMark Vantage has achieved. At the same system, a 9800 GTX P5816 points obtained. So far so good: It is interesting, however, that the X-mode, in 1920x1200 instead of 1280x1024 and there is also the default 4-times anti-aliasing and 16 times anisotropische filtering active, the score against a 9800 GTX with points instead X2609 X2104 points, 24 percent higher.

So it seems the 4850 could either take less of a performance hit with AA enabled or simply handle the higher shader settings better than the 9800GTX.

http://www.hardware-infos.com/news.php?news=2124

 

[tacopaco]

the score against a 9800 GTX with points instead X2609 X2104 points, 24 percent higher.

so wait the 4850 has better anti-aliasing interesting....
could that mean the 4870x2 would have an edge on the gt200 in that department?

if the 9800gx2 and the gt200 are neck and neck whats keeping a much cheaper 4870 crossfire rig from completely destroying them?

 

[ShaidarHaran]

German site pulled together a few scores from various rumours:



So it seems the 4850 could either take less of a performance hit with AA enabled or simply handle the higher shader settings better than the 9800GTX.

http://www.hardware-infos.com/news.php?news=2124

FUD.

I have an 8800 GT in my E8400-powered system and even with everything stock it scores just under 12k so I instantly distrust their numbers when they claim it only pulls 10.8k.

 

[ShaidarHaran]

so wait the 4850 has better anti-aliasing interesting....
could that mean the 4870x2 would have an edge on the gt200 in that department?

if the 9800gx2 and the gt200 are neck and neck whats keeping a much cheaper 4870 crossfire rig from completely destroying them?

Because performance does not scale linearly with additional GPUs. i.e. adding a 2nd card does not double performance. The average boost seems to be somewhere in the 40-50% neighborhood for most modern games that are AFR-friendly.

 

[DeF]

FUD.

I have an 8800 GT in my E8400-powered system and even with everything stock it scores just under 12k so I instantly distrust their numbers when they claim it only pulls 10.8k.

Are you sure? At xbitlabs 8800GT scores 10925 with X6800

 

[tacopaco]

Because performance does not scale linearly with additional GPUs. i.e. adding a 2nd card does not double performance. The average boost seems to be somewhere in the 40-50% neighborhood for most modern games that are AFR-friendly.

I knew that sli and crossfire are never going to be 100% efficient but the rig would provide some serious competition imo
thats a second card right i have heard of 3 way sli being even less efficient and the 3rd card makes little difference but that could be just horrible driver support for three way and that most games don't support it well...

 

[CarstenS]

Quite intriguing. Just yesterday i ran a few 3DMark06s with my trusty old HD2900 XT on Beta-Cat 8.6 (supposedly) an also got 11.7xx-ish points on my OC-E8500.

 

[tacopaco]

hmm 2900.... 11.7 i run a similar rig (well my brother does) and he runs it with a 6850 oced to 3.8 and he gets 11 ish and he oc's the 2900 to very unsafe clock speeds like 70ish idle 97 load
hes not the brightest in the world but hey its not my rig

 

[Shtal]

FUD.

I have an 8800 GT in my E8400-powered system and even with everything stock it scores just under 12k so I instantly distrust their numbers when they claim it only pulls 10.8k.

My Radeon 3870 clocked @ 825MHz core / 2400MHz GDDR4 - combine with intel Quad Q6600 @ 3GHz w/4GB DDR2 @800MHz 4-4-4-12 scores in 3DMARK06 12.4k sometimes 12.5k points.


EDIT: If I clocked my Radeon 3870 @ 860core/2600MHz GDDR4, it scores 12.9k and my Q6600 @ 3GHz.

 

[ZerazaX]

3dMark06 scores are way too heavily affected by CPU to really be able to take any of those #'s as literal values - you can only compare it to other cards tested on the same rig

The Vantage scores are a much better poitn to look at - if the 4850 really scores 24% better than 9800GTX at 1920 x 1200 with 4xAA and 16XAF, then it looks like AF is fixed and that it won't be as bandwidth limited at high resolutions as the 9800GTX certainly is with those settings

Either way though, a card at $179 with that level of performance would be downright amazing

 

[JasonCross]

3DMark06 these days scales quite a bit with processor speed at default settings. It's almost impossible to make a useful 3DM06 comparison without running the cards in the same system with the same settings, unless you start cranking up the resolution and piling on the AA/AF.

One of the things they did right with 3DM Vantage is change the contribution of CPU tests to the final score at different profiles, and the physics/AI tests are a much smaller percentage of the score at Xtreme than at Performance, for instance.