ATI Radeon HD 4000 Series discussion

[ORBR]

Does the GPU-Z reading of the Catalyst drivers line up with 8.5? I don't have an ATI card so I can't tell if those are the correct ID #'s for 8.5 / Vista

in the right Catalyst 8.6 [ Vista 64 ] from one of my PCs



looks like the catalyst used mismatch the version string

[Miksu]

GPU-Z uses a database to identify the cards. If the DB isn't up-to-date, it shows incorrect information (well, clock rates are probably read real-time). Even the original author of the review says that the GPU-Z values are wrong:

"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."

So 480vs800 SP is still undecided. I would lean on the 800's side though...

[oohms]

So 480vs800 SP is still undecided. I would lean on the 800's side though...

Purely based off transistor count, 480 seems a lot more likely

[shadow of chaos]

so we have weak cards?

[Miksu]

Purely based off transistor count, 480 seems a lot more likely

And on the other hand the Perlin noise -score says it's 800

[spoof]

GPU-Z uses a database to identify the cards. If the DB isn't up-to-date, it shows incorrect information (well, clock rates are probably read real-time). Even the original author of the review says that the GPU-Z values are wrong:

"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."

So 480vs800 SP is still undecided. I would lean on the 800's side though...

If GPU-Z info is legit, the core and mem are downclocked, so this isn't the actual score ot 3D Mark06. Furthermore, the shot from gpu-z given from ORBR have different default clock freq, so one more missmatch.

[cadaveca]

gpu-z seems fine for clocks, driver is correct for 4850's, mem bandwidth ok, fillrates are not, do not scale with clockspeed changes.

[Hornet331]

If those are legit it looks like they've majorly tightened up on the AF issue. With 8xAA/16xAF:

the scores are legit, but who know maybe it shoped.

[SoulsCollective]

the scores are legit, but who know maybe it shoped.

u cn tel bi d pixls?

[Hornet331]

u cn tel bi d pixls?

fear my mad calculator skillz!!111

[865pe]

Rv770 > 800 Shaders or 480?

[Macadamia]

And it's also got 480 shaders. Damn, the 4000 series is going to suck. What a let down. That will likely go for the 4870 as well. If the 4850 scores X2700 in vantage, then the 4870 will maybe score X3100, which may even be optimistic. If the GTX 280 is scoring X4800, the 4870 is going to get pounded in a huge way.

And I sincerely doubt the 4870X2 is going to score X5500. I mean thats not far off from double the performance, and multi GPU configurations do NOT scale that well.

Err...

2800*1.3 = 3640 (Current correct GTX 260 scores are about X3700-3800, 80% of the GTX 280 which scores ~X4800)

I expect 4870 to get about X3600. And it's DEFINITELY 800 (160), no talk about anything else

[alexio]

RV770 = 800SPs

Don't ask for the source

[Genius_pt]

Hmm, doesn't sound too good for me. With 800sp this cards should have scored better !

[perkam]

Err...

2800*1.3 = 3640 (Current correct GTX 260 scores are about X3700-3800, 80% of the GTX 280 which scores ~X4800)

I expect 4870 to get about X3600. And it's DEFINITELY 800 (160), no talk about anything else

A. The 4870 is priced a third lower than the GTX 260, so there is NO expectation for it to get the same mark.
B. Even if it had similar performance, Vantage has had an Nvidia bias since launch and though that was remedied with the ATI vantage hotfix, you can't rely on Vantage alone for performance benchmarking.
C. 480 or 800, it will be a killer card for the price

The scores arent TOO bad at all, considering what the 4850 can do with voltmods.

Perkam

[alexio]

Hmm, doesn't sound too good for me. With 800sp this cards should have scored better !

They are simpler units than those on the RV670 I presume, but I can not confirm what the exact change has been. It has 800SPs according to my source so I'm pretty sure about that.

[AliG]

how could they be simpler units? If they were simpler units that did less intructions then ATI would have needed more shaders to achieve that 800 number (as they multiply the number of shaders by the number of instructions it can do basically for the final number), definitely not a simpler shader

[alexio]

http://www.beyond3d.com/content/reviews/16/8
Yes, a shader can be simpler if it is dedicated to a smaller amount of instructions.

[AliG]

You aren't getting the point, ATI gets its shader count by multiplying the number of shaders by its possible instructions. So for the R600 you got 64*5 shading instructions = 320. Now if the rv770 has 800 shaders, you HAVE TO HAVE either 160 * 5 shading instructions or more shaders. No other possible way for it to happen, otherwise it wouldn't be "800" shaders.

[Macadamia]

A. The 4870 is priced a third lower than the GTX 260, so there is NO expectation for it to get the same mark.
B. Even if it had similar performance, Vantage has had an Nvidia bias since launch and though that was remedied with the ATI vantage hotfix, you can't rely on Vantage alone for performance benchmarking.
C. 480 or 800, it will be a killer card for the price

The scores arent TOO bad at all, considering what the 4850 can do with voltmods.

Perkam

Why not?

Vantage is still really shader limited (3850 > 9600GT) but unfortunately RV670 doesn't have enough TMUs to not get bottlenecked.

RV770's popular TMU number aka 32, should be enough to not be bottlenecked. G80 proved that.

And also, 4850 is heavily bottlenecked by bandwidth. Can't wait for a 2.2Ghz DDR3 version to come, I smell some 9 series slaughter.

Plus, XFastest.com has REAL screenshots of the 4850 in Vantage, see the feature test scores. They confirm 800SPs, and much better scores in the 6 feature tests (more important for me than the overall score)

[AliG]

link? I generally find it myself, but its kinda hard when you can't read the page lol

[alexio]

You aren't getting the point, ATI gets its shader count by multiplying the number of shaders by its possible instructions. So for the R600 you got 64*5 shading instructions = 320. Now if the rv770 has 800 shaders, you HAVE TO HAVE either 160 * 5 shading instructions or more shaders. No other possible way for it to happen, otherwise it wouldn't be "800" shaders.

Yeah I get what you mean. It seems I used the word 'shader' wrongly. I meant ALUs, simpler ALUs. ATI would count this as 100*8 for example.

[Hornet331]

Why not?

Vantage is still really shader limited (3850 > 9600GT) but unfortunately RV670 doesn't have enough TMUs to not get bottlenecked.

RV770's popular TMU number aka 32, should be enough to not be bottlenecked. G80 proved that.

And also, 4850 is heavily bottlenecked by bandwidth. Can't wait for a 2.2Ghz DDR3 version to come, I smell some 9 series slaughter.

ahm im sure you know the G80 had 64TMUs so closed the gap from 1/4 to 1/2

[AliG]

no, g80 had 32, g92 had 64

[Macadamia]

http://xfastest.com/viewthread.php?t...&extra=&page=1


And yet, it is the FASTEST single GPU in texture fillrate when compared to the Digit life scores.
(RV770 gets higher than the 64TMU G92s, touche)