AMD: R7xx Speculation

[ZerazaX]

Says 480 shaders on the GPU-Z:

http://publish.it168.com/2008/0610/images/1039036.JPG

So the SP thing is wrong, GPU-Z thing is wrong, or they do have a shader clock?
Looks like fill rates are vastly improved though

 

[no-X]

AlphaWolf: Some of the GPU scores are slightly lower than I've seen (lower CPU clock), CPU scores are better (higher number of CPU cores), but these results are very likely real.

 

[ZerazaX]

Wait, is it just me or are those GPU-Z fill-rates freakishly large/wrong? I thought we were in the tens of GTexel/s and GPixel/s, not hundreds

 

[AnarchX]

Wait, is it just me or are those GPU-Z fill-rates freakishly large/wrong? I thought we were in the tens of GTexel/s and GPixel/s, not hundreds

TMUs emulated via shaders? More probadly RV770 detection is not final.


Seems HD4800 to offer good performance for price, while offering top-notch number-crunching power, which could give also G200 a hard time.

 

[Anarchist4000]

Says 480 shaders on the GPU-Z:

http://publish.it168.com/2008/0610/images/1039036.JPG

So the SP thing is wrong, GPU-Z thing is wrong, or they do have a shader clock?
Looks like fill rates are vastly improved though

33% less bandwidth, 3790% increase in pixel fill rate, 7680% increase in texture fill rate over RV670. Yeah I'm a little doubtful.

 

[ZerazaX]

maybe ATI does employ magicians after all

Also, they claim use of Cat 8.5 so if those 3dMark06 scores are legit (seem in line with whats been heard), then it should surely improve with next round of driver releases

Actually, I'm wondering, what parts of GPU-Z are detecting the actual hardware (clocks definitely are) and how much is just a database? At this point, much of those specs are speculation and unless the card itself has areas where it tells you the true value (how likely is it though that the BIOS on the card actually stores die size) vs. what's currently speculated and will be updated once the next round is released?

 

[AlphaWolf]

or perhaps GPU-z just isn't working correctly for this product

 

[ZerazaX]

Right, which is likely, but that's why I'm asking how much of GPU-Z is actual detection of card specs vs. speculated specs put in a database?

 

[Raqia]

The 480 SP rumors might be based on a GPU-Z detection glitch.

 

[AlphaWolf]

Right, which is likely, but that's why I'm asking how much of GPU-Z is actual detection of card specs vs. speculated specs put in a database?

W1zzard almost releases a new version with every card release.

http://forums.techpowerup.com/forumdisplay.php?f=56

You'll notice rv770 or 48xx doesn't seem to have a release.

 

[Pete]

Those 3DM06 scores look slightly below an 8800GTS.

19x12 1x1  (slower CPU, old drivers on G92GTS)
3DM06        GT1   GT2   HDR1  HDR2  CPU
"4850"       29.3  37.9  41.5  45.2  4100(!)
8800GTS-512  37.5  39.3  40.1  40.3  2300
diff.        1.3x  1.0x  1.0x  0.9x

The 19x12 8x16 3DM06 #s might be slightly more interesting, though I forget how each card does 8xAA and I only have a 9800GX2 to for comparison:

19x12 8x16 (slower CPU on GX2)
3DM06    GT1   GT2   HDR1   HDR2  CPU
"4850"   20.7  22.6  27.8   22.3  4100(!)
9800GX2  44.0  36.7  41.1   31.2  2700
diff.    2.1x  1.6x  1.5x   1.4x

The Vantage Extreme scores put it slightly above a 9800GTX/8800GTS.

Though a teratexel fillrate is more than I was expecting. Practically all those #s are wrong, even the bandwidth (which should be a simple calculation from mem speed and bus width). Either it's disinformation, or an older GPU-Z database (assuming W1zzard has the specs and has already added them to current versions--I thought we'd established that GPU-Z is basically all database). Ah, it268 agrees GPU-Z's not right: "We test used by the test drive version amd_vista_radeonhd_4850_8.50.3_final_certified, in the latest version of the GPU-Z in a comprehensive manner and failed to obtain relevant information, but we can understand that the core and memory frequencies were 625 / 1986MHz."

I don't think GPU-Z invalidates real benchmarks #s, and if there are 4850s floating around (see OBR) then eventually something's going to leak. The question is, are those 3DM #s believable? I think the Vantage #s agree with CJ's.

Edit: Right, I forgot the RAM was clocked down.

 

[ZerazaX]

Those Vantage #'s seem close to CJ's definitely

GPU-Z definitely looks like the info is wrong unless TeraScale means Teratexel fillrate

Hard to compare 3dM06 across systems though unfortunately especially on a quad 3.0 with 3GB of RAM, and we all know the CPU affects 3dMark06 a lot. However, they were using Cat 8.5 so its likely that the drivers aren't optimized anyways so there should certainly be improvement in the next Cat release, which is likely with the cards.

Looks like the GPU-Z data is wrong though since w1z updates the database w/ each card release

BTW the bandwidth looks correct since current clocks show memory downclocked to 750 MHz w/ PowerPlay it seems... and I don't really remember the equation for calculating bandwidth but I got 750 * 2 * 256 * 2 / 16 = 48 Gb/s but that's an easy thing to calculate for that program anyways

 

[Shtal]

http://publish.it168.com/2008/0610/20080610041602.shtml

4850 3dmark06 scores on a bunch of different settings. No idea if they are real.

My Intel_core2 quad Q6600 over-clocked @ 3GHz scores 4700 points in 3DMark06 for CPU score test.

Configuration.
1. Asus P35 chipset motherboard.
2. DDR2 800 4-4-4-12.

lol: My Radeon HD 3870 825core/2400GDDR4 - Default Test scores 12,497 points in 3DMark06

 

[Shtal]

1800MHz GDDR5 slow? latacy on GDDR5 vs. GDDR3 comparison?

GDDR5 memory on a 256-bit memory bus running at 1800MHz (effective 3600MHz DDR speed) would deliver 115.2 GB per second.

 

[AlphaWolf]

1800MHz GDDR5 slow?

It's as slow as GDDR5 gets.

 

[ZerazaX]

Probably lower binned for reference cards anyways so they can get the cheaper easier to make stuff out and in big enough quantities

My bet is AIB's will add the super high clocked GDDR5 in a few months anyways

 

[Thorburn]

The 480 SP rumors might be based on a GPU-Z detection glitch.

GPU-Z doesn't calculate these things as far as I know, 90% of the details are filled in from a database lookup. So for GT200 and RV770 it's going to be the authors best guess at rumors.

 

[itsmydamnation]

http://www.auspcmarket.com.au/show_...uct_code]=VI-RV770PRO&input[category_id]=1204

for all the people in OZ, not bad bang for buck at all...........

now all we need is some damn specs!!

 

[Tchock]

Moar perlin noise. And other associated treats.

Feature Test 6 for the uninformed.

Cloth's framerate is the big anomaly I guess. Other than that, feature tests are screaming. Throw in more bandwidth and there's something really working here (I guess, and hope )

 

[fellix]

Feature Test 6 in Vantage is namely the Perlin Noise redux from 3DMark'06, so 4850 matching here the 3870X2 is expected.