some ATI 4800 details

We've been through all that, but if that was really the case, why up the count to 480 or 800 SPs, if all they need are more ROPs and TMUs ?
Point is, the 800 SPs number seems way too far fetched, even more so when the rumored performance numbers don't fit it. With 800 SPs and assuming that all R600/RV670 problems were taken care of (as rumors seem to indicate), a single HD 4870 should be taking the GTX 280's crown and leading by quite a bit.
Click to expand...
Stop trying to compare # of SPs across architectures. It's meaningless. And I don't kinda-sorta meaningless. I mean it really says absolutely nothing. Take that in for a minute.

Anyway, 800 SPs is CONFIRMED both by official slides and by running tests. Some guy over at B3D ran Perlin Noise tests which proved 800 SPs. Go check it out instead of spewing BS.
 
It's likely either 800SPs or a shader clock domain. The tests and achieving 1 TFLOP with the 4850 would support one of those options.
 
DeadSkull said:
Has the 40% asskicking gain over the gt been confirmed?
Click to expand...

How about you read the last few pages of this thread before posting? There's some 4850CF benchmarks and they look extremely promising.

I still don't understand how they could increase shaders by 150% and still fit in such a small die. Amazing :).
 
Kaldskryke said:
How about you read the last few pages of this thread before posting? There's some 4850CF benchmarks and they look extremely promising.

I still don't understand how they could increase shaders by 150% and still fit in such a small die. Amazing :).
Click to expand...

Dude, I was just making fun of the argument above my post by asking another stupid question people already know the answers too. Yes I have read most of the stuff from last few pages and I am aware that 4850cf beats the new 280. But so does a 8800gt sli rig at lower resolutions including 1680x1050.
 
A lot has been made of multi-GPU solutions and their issues and it’s been a while since I used a “true” SLI solution, meaning two separate physical cards, and at least with SLI, there are still issues with certain games. Sure you get faster average rates, but diminished game play. Performance charts came become useless. For instance, a lot of reviews will lead you to believe that Crysis runs better on a 9800 GX2 than a GTX 280. After a couple of hours of testing last night, that’s just not the case on my rig. In fact even SLI GTX 280 had problems.

I will be very interested in seeing if AMD has solved these multi-GPU issues with ALL games. AMD might very well be at the top of the performance charts come release time, with inferior game play. Just like the 9800 GX2 vs. the GTX 280.
 
It's pretty confirmed if they are saying 1 TFp on the 4850 and the Perlin noise as well

Theres always the chance of shader clocks but they havent been detected by ATI CCC or GPU-Z
 
Saw a French review that says 800SPs. AA hit seems to be roughly on par with the 9800GTX so it looks like the performance has been fixed there. Seems to vary a bit from game to game. Some tests position it between the 260 and 280 and others have it just below the 9800s.
 
Anarchist4000 said:
Saw a French review that says 800SPs. AA hit seems to be roughly on par with the 9800GTX so it looks like the performance has been fixed there. Seems to vary a bit from game to game. Some tests position it between the 260 and 280 and others have it just below the 9800s.
Click to expand...
Between the 260 and 280 performance levels would be nuts at that price point. I wonder what the 4870 can do...?
 
Anarchist4000 said:
Saw a French review that says 800SPs.
Click to expand...
Thanks for the link. The 4850 seems to beat the 9800GTX with AA in most tests. Without AA, the 9800GTX seems to pull ahead a bit in more tests.

The 4850 beats the GTX 260 without AA in: Rainbox Six: Vegas, GRID (also with AA) and Bioshock.

Very promising.

Test system was:

Intel Core 2 Extreme QX9770
Asus Striker II
4 Go DDR3 1066
Windows Vista SP1
Forceware 177.34
Catalyst 8.5
Catalyst beta for Radeon HD 4850
 
It looks like there's still a few games where ATI's AA just dies, like Stalker (page 4) and bioshock(page 6).
 
I think it works but just wasn't enabled. Something along the lines of DX10 having to change AA settings and not the control panel so it was disabled by default.
 
berzerker said:
Thanks for the link. The 4850 seems to beat the 9800GTX with AA in most tests. Without AA, the 9800GTX seems to pull ahead a bit in more tests.

The 4850 beats the GTX 260 without AA in: Rainbox Six: Vegas, GRID (also with AA) and Bioshock.

Very promising.

Test system was:

Intel Core 2 Extreme QX9770
Asus Striker II
4 Go DDR3 1066
Windows Vista SP1
Forceware 177.34
Catalyst 8.5
Catalyst beta for Radeon HD 4850
Click to expand...

Talk about eye opening! That French review has made me a believer! The 4850 routinely beats the 9800 GTX and even manages to match and in some cases beats the GTX 260... simply amazing! :eek:

Here are the benchmark results.

All tests 1920x1200 & 1920x1200 w/4xAA

Enemy Territory: Quake Wars
img0023558bk8.gif


Half Life 2: Episode 2

img0023559iw2.gif


Bioshock DX10
img0023563cd9.gif


Company of Heroes DX10
img0023564oq4.gif


World in Conflict DX10
img0023565vd1.gif


Crysis - Very High DX10
img0023566qu4.gif


Power Consumption
img0023572rv9.gif
 
looking at the french scores... its pretty clear the 4870 will beat the gtx260 for less money

and the 4870x2 will cream the gtx280

again for less money.... i hope this is spot on because this helps bring these crazy prices down
 
DeadSkull said:
...But so does a 8800gt sli rig at lower resolutions including 1680x1050.
Click to expand...

What evidence are you basing that on? I've read four GTX280 (/GTX260) reviews and none of them show that. If we say the delta between the GX2 and two GT's is 15% (which is on the conservative side, Its more like 20 or 30), your still looking at the GTX 280 outscoring two 8800GT's across the board.

But hey, its all speculation right? We'll see how the HD4850 performs, and Tweaktown seems reasonably impressed.
 
POPEGOLD said:
looking at the french scores... its pretty clear the 4870 will beat the gtx260 for less money

and the 4870x2 will cream the gtx280

again for less money.... i hope this is spot on because this helps bring these crazy prices down
Click to expand...

Yea, the 4870x2 with the AA fixed and shared memory looks like ATI could have a really great card to challenge the GTX280.
 
MrWizard6600 said:
What evidence are you basing that on? I've read four GTX280 (/GTX260) reviews and none of them show that. If we say the delta between the GX2 and two GT's is 15% (which is on the conservative side, Its more like 20 or 30), your still looking at the GTX 280 outscoring two 8800GT's across the board.

But hey, its all speculation right? We'll see how the HD4850 performs, and Tweaktown seems reasonably impressed.
Click to expand...

Look at anandtech review at 1680x1050 it gets beat by the 8800gt sli setup and even at 1920x1200 it wins some and loses others. Only at 2560x1600 does it really start to pull ahead.
 
Doodies said:
Look at anandtech review at 1680x1050 it gets beat by the 8800gt sli setup and even at 1920x1200 it wins some and loses others. Only at 2560x1600 does it really start to pull ahead.
Click to expand...

I stand corrected. But I'm shocked. The hard numbers tell a different story even with the 8800GT SLI hitting 200% over a single 8800GT. 240SPs as compared to 224, 120 TMUs as compared to 112, 140GB/s as compared to 110, Pixel fill rate should be identical. Whats going on here???

I think were going to see some serious extra performance from later drivers if the current ones are allowing two 8800GT's to catch up.
 
MrWizard6600 said:
I stand corrected. But I'm shocked. The hard numbers tell a different story even with the 8800GT SLI hitting 200% over a single 8800GT. 240SPs as compared to 224, 120 TMUs as compared to 112, 140GB/s as compared to 110, Pixel fill rate should be identical. Whats going on here???

I think were going to see some serious extra performance from later drivers if the current ones are allowing two 8800GT's to catch up.
Click to expand...
the GTX 280 has 80 TMUs not 120.
 
gs29 said:
Stop trying to compare # of SPs across architectures. It's meaningless. And I don't kinda-sorta meaningless. I mean it really says absolutely nothing. Take that in for a minute.

Anyway, 800 SPs is CONFIRMED both by official slides and by running tests. Some guy over at B3D ran Perlin Noise tests which proved 800 SPs. Go check it out instead of spewing BS.
Click to expand...

Across architectures ? It's the SAME architecture, which is why 800 SPs doesn't seem reasonable. There is no confirmation yet, but I would like to see those "official slides". You mean the suspicious looking slides, with extreme compression artifacts around the 800 number ? :rolleyes:
 
Silus said:
Across architectures ? It's the SAME architecture, which is why 800 SPs doesn't seem reasonable. There is no confirmation yet, but I would like to see those "official slides". You mean the suspicious looking slides, with extreme compression artifacts around the 800 number ? :rolleyes:
Click to expand...

Its 800 unless you want to explain to me how those 40 blocks on here correspond to 96 vec 5 shaders?
popImg.php


Its 40 quad ALU's so 40 x 4 = 160 vec 5 = 800 SP's

Heck, there might be 40 TMU's to boot depending on what those things at each end are
 
Across architectures ? It's the SAME architecture, which is why 800 SPs doesn't seem reasonable. There is no confirmation yet, but I would like to see those "official slides". You mean the suspicious looking slides, with extreme compression artifacts around the 800 number ?
Click to expand...

I was talking about RV770 vs GT200. You were whining about RV770 not beating GTX 280, remember?

I like how you have nothing to say about the perlin noise tests or Zerazax's excellent point on the die shot. You hold on so bitterly against the 800 SP number despite having obvious proof shoved in your face. What's your agenda? Nvidia PR & damage control?
 
gs29 said:
I was talking about RV770 vs GT200. You were whining about RV770 not beating GTX 280, remember?
Click to expand...

I was ? You must've said it for me then...:rolleyes:

I did say that with 800 SPs, even the HD 4850 would be doing much more than beating a 8800 GT 512...as in, taking on a GTX 280.

gs29 said:
I like how you have nothing to say about the perlin noise tests or Zerazax's excellent point on the die shot. You hold on so bitterly against the 800 SP number despite having obvious proof shoved in your face. What's your agenda? Nvidia PR & damage control?
Click to expand...

What obvious proof ? There's is NO obvious proof. We don't even know at what frequency the SPs are set. Multiple rumors indicate that RV770 decoupled the shader domain frequency from the clock frequency, which was not the case of R600 and RV670, but we do NOT know that shader frequency yet.

And I can say the same thing about you i.e. ATI PR guy or something of the sort, especially given your join date. You've done nothing more than drive this "800 SPs" rumor home, stating that there is "obvious proof", but there's no such thing, so who exactly has an agenda here ? :rolleyes:

The proof is in a thread next door, where someone that actually bought a HD 4850 and ran a few tests for us, can shed some light on the issue. I already popped the question there. Let's wait for the answer.
 
Silus said:
We've been through all that, but if that was really the case, why up the count to 480 or 800 SPs, if all they need are more ROPs and TMUs ?
Point is, the 800 SPs number seems way too far fetched, even more so when the rumored performance numbers don't fit it. With 800 SPs and assuming that all R600/RV670 problems were taken care of (as rumors seem to indicate), a single HD 4870 should be taking the GTX 280's crown and leading by quite a bit.
Click to expand...


ATI's has been increasing the number of SP units they have at the vrs adding more ROP/TMUs in the last few geneerations. Maybe they are trying to counter TESLA and some of what nV is doing with CUDA? And there is no way in HELL that a single 4870 would ever beat the GTX 280 in any game at the settings most H's play at (high res with AA/AF as that is all bandwidth/fillrate limited which the GTX 280 has more of in BOTH!). Come on Silus your smart enough to know that the # SP != absolute performance and thats all my orginal post was trying to point out....
 
Jbirney said:
ATI's has been increasing the number of SP units they have at the vrs adding more ROP/TMUs in the last few geneerations. Maybe they are trying to counter TESLA and some of what nV is doing with CUDA? And there is no way in HELL that a single 4870 would ever beat the GTX 280 in any game at the settings most H's play at (high res with AA/AF as that is all bandwidth/fillrate limited which the GTX 280 has more of in BOTH!). Come on Silus your smart enough to know that the # SP != absolute performance and thats all my orginal post was trying to point out....
Click to expand...

But that was precisely my point, when I answered your post. Why increase the number of SPs by 2.5 times as the ones in RV670, if what was crippling their performance was the low ratio of TMUs and ROPs ? By that line of reasoning, even 480 SPs shouldn't be needed, much less 800 SPs.

Obviously SP count is not what will determine the final performance, but it IS one of the key aspects to achieve it, in an unified architecture and increasing them by 2.5 times, should yield it much better results than just sitting in between a 8800 GT 512 and a 8800 GTS 512.

Either that, or there's some serious efficiency problems as was mentioned before.
 
What obvious proof ? There's is NO obvious proof. We don't even know at what frequency the SPs are set. Multiple rumors indicate that RV770 decoupled the shader domain frequency from the clock frequency, which was not the case of R600 and RV670, but we do NOT know that shader frequency yet.

And I can say the same thing about you i.e. ATI PR guy or something of the sort, especially given your join date. You've done nothing more than drive this "800 SPs" rumor home, stating that there is "obvious proof", but there's no such thing, so who exactly has an agenda here ?

The proof is in a thread next door, where someone that actually bought a HD 4850 and ran a few tests for us, can shed some light on the issue. I already popped the question there. Let's wait for the answer.
Click to expand...
Oh, I have no love for AMD or disdain for Nvidia. I was going to buy a 280 until it turned out it was horse shit disguised as a video card.

The proof is there: a picture of the GPU die and perlin noise tests that matched the performance of 800 SPs. So let's see: a picture and a test vs. old rumors. Brilliant. But here's something for you. Your beloved company itself mentions the 800 SPs:

http://forum.beyond3d.com/showpost.php?p=1178466&postcount=4014

AMD’s Radeon HD 4850 has 800 shader cores. The GeForce 9800 GTX+ has only 128 cores. Doesn’t this put you at a disadvantage?

Core counts are as about as useful as MHz ratings. In the end, it comes down to real world performance. Our cores are designed to be exceptionally efficient and operate at near full utilization.

If you were to normalize 3DMark Vantage performance, per core, it would give you a better sense of the relative performance of different cores.
Click to expand...
 
gs29 said:
Oh, I have no love for AMD or disdain for Nvidia. I was going to buy a 280 until it turned out it was horse shit disguised as a video card.

The proof is there: a picture of the GPU die and perlin noise tests that matched the performance of 800 SPs. So let's see: a picture and a test vs. old rumors. Brilliant. But here's something for you. Your beloved company itself mentions the 800 SPs:

http://forum.beyond3d.com/showpost.php?p=1178466&postcount=4014
Click to expand...

I really don't know if I should be laughing or be just in awe with your "proof"...

So "someone" asks some NVIDIA guy (I'm assuming he/she is a NVIDIA person, since your example is pretty vague..), that if the HD 4850 has 800 SPs they are in trouble and so that means NVIDIA said it ? :rolleyes:

Who is this "someone" making questions and where did he gather the 800 SPs number ? From the flacky slides with compression artifacts all over the 800 number or some place else ?
Where's the rest of this "interview" ?
Also, do you have the shader domain frequency for the RV770 Pro chips, so that I can do the math for you ?

The only person that can end this debate at this point, is the guy that posted his results for the HD 4850 in another thread. He's the one that can confirm or deny the 800 SPs in the HD 4850. Too bad he didn't answer my question yet. As for your "proof", it's laughable to say the least.
 
Interview? Christ... are you blind? It's some PR statements that Nvidia sent out to accompany the 9800GTX+ announcement; it's akin to an FAQ.
 
gs29 said:
Interview? Christ... are you blind? It's some PR statements that Nvidia sent out to accompany the 9800GTX+ announcement; it's akin to an FAQ.
Click to expand...

It wasn't the NVIDIA guy that said it...:rolleyes:.

Anyway, I finally saw a site, in one of those early previews, to show actual specifications:

http://www.guru3d.com/article/amd-ati-radeon-hd-4850-review-force-3d--powercolor/3

And they do mention 800 SPs, but the transistor count is way past the rumored 800+ million. It's actually almost 1 billion. Now that's something I consider proof.
 
What "the NVIDIA guy" are we talking about? It was a notification as I clearly stated; not an interview, no "guy" saying anything. Are you just desperate to save face? You bitterly held out against the 800 SP number long after everyone else accepted it. Get over it already.

christpunchersg: The slides Zerazax posted above are pretty much the sum of what we know about the R700 right now. Only the insiders know for sure.
 
You must log in or register to reply here.
Back
Top