ATI 4xxx Series Thread

[JACKDRUID]

Originally posted by: Ocguy31

Originally posted by: Extelleron
The number of GPUs on the PCB of a graphics card does not in any way harm or negatively affect the performance and gaming experience that a card gives you.

Wow folks....before you read anything else that he writes, read this post. This is even a better quote than my sig now!!!!!

sigged!

 

[Extelleron]

Originally posted by: JACKDRUID

Originally posted by: Ocguy31

Originally posted by: Extelleron
The number of GPUs on the PCB of a graphics card does not in any way harm or negatively affect the performance and gaming experience that a card gives you.

Wow folks....before you read anything else that he writes, read this post. This is even a better quote than my sig now!!!!!

sigged!

Tell me how this is not true.

If I have Card A, that is made up of 1000 GPUs connected to eachother, and Card B, that has 1 GPU. And Card A gets 100 FPS AVG and 80 FPS min. Card B gets 100 FPS AVG and 80 FPS min. What is the difference to the end-user?

I'm failing to see your point here.

 

[Quiksilver]

Originally posted by: Extelleron

Originally posted by: JACKDRUID

Originally posted by: Ocguy31

Originally posted by: Extelleron
The number of GPUs on the PCB of a graphics card does not in any way harm or negatively affect the performance and gaming experience that a card gives you.

Wow folks....before you read anything else that he writes, read this post. This is even a better quote than my sig now!!!!!

sigged!

Tell me how this is not true.

If I have Card A, that is made up of 1000 GPUs connected to eachother, and Card B, that has 1 GPU. And Card A gets 100 FPS AVG and 80 FPS min. Card B gets 100 FPS AVG and 80 FPS min. What is the difference to the end-user?

I'm failing to see your point here.

That the 1000 GPU Card is seriously under performing. (I kid I kid)

Generally with multi-gpu cards you run into more driver problems and problems with older games compared to single gpu cards.

Also more heat.

 

[Extelleron]

Originally posted by: Quiksilver

Originally posted by: Extelleron

Originally posted by: JACKDRUID

Originally posted by: Ocguy31

Originally posted by: Extelleron
The number of GPUs on the PCB of a graphics card does not in any way harm or negatively affect the performance and gaming experience that a card gives you.

Wow folks....before you read anything else that he writes, read this post. This is even a better quote than my sig now!!!!!

sigged!

Tell me how this is not true.

If I have Card A, that is made up of 1000 GPUs connected to eachother, and Card B, that has 1 GPU. And Card A gets 100 FPS AVG and 80 FPS min. Card B gets 100 FPS AVG and 80 FPS min. What is the difference to the end-user?

I'm failing to see your point here.

That the 1000 GPU Card is seriously under performing. (I kid I kid)

Generally with multi-gpu cards you run into more driver problems and problems with older games compared to single gpu cards.

Also more heat.

That's definitely true, but that was not part of my statement. What I am saying is that, if there are no significant driver problems, and R700 beats GT200 in performance, you cannot discount it based on the fact that it is multi-GPU. Its multi GPU nature may make it less efficient (perhaps it will be only 50% faster than the HD 4870 in a game) but that does not harm the performance that is there. If R700 scales 0% but gets 80 FPS, while GT200 gets 70 FPS, then R700 wins, regardless of the fact that Crossfire did not work. If Crossfire does not work and performance is not there, and GT200 wins, then that is a game where R700 loses.

What I am saying is that people should look at the numbers when the review comes out and compare it directly to GT200 without stipulations. Multi-GPU Inefficiency does not affect the FPS that are already there, it just means the card does not reach its full potential based on specs.

 

[JACKDRUID]

Originally posted by: Extelleron

Tell me how this is not true.

If I have Card A, that is made up of 1000 GPUs connected to eachother, and Card B, that has 1 GPU. And Card A gets 100 FPS AVG and 80 FPS min. Card B gets 100 FPS AVG and 80 FPS min. What is the difference to the end-user?

I'm failing to see your point here.

alot of traveling.. repetitive processing if data isn't multithreaded..

if data is multithreaded, then there would be overhead to "put results back together"

 

[Calculator83]

Originally posted by: Extelleron
I'm not going to argue with you anymore Wreckage. It's clear you and I have different opinions on what AMD/nVidia are doing and nothing that either of us say is going to change that.

BTW Wreckage I just bought an eVGA GTX 280, because I am such a ATI fanboy.

Now I have to hope HD 4870 doesn't touch the GTX 280 at it's $300 price

Dude, Think about it this way,, Thats a bad bet you're making.

4850 = 200 ,, 4870 = 300

If we're assuming it is worth the 1.5x the money, Then we'd More or less see at least 25 to 30% gain in performance. WHICH means, as ATI has already quoted the 4850 as up to 75% of the 280 Performance. Than the 4870 MUST be at least 100% to 105%.

These are terrible Odds ur betting against. especially with 650 dollars.

I don't think ur this crzy.. U didn't buy one at all did you?,, U just said that to shut him up.. LIAR.

 

[Grinja]

Originally posted by: RussianSensation

Originally posted by: airhendrix13

From a Manufacturing standpoint, having a single GPU would be better for them. It runs cooler (usually), uses less power (again usually), and thus in the end would end up saving them money on cooling and such.

I think you are missing some key points:

1) A less complex gpu has higher yields which MORE than offset cooling costs. Compared to wasted wafer space and # of working gpus one can pump out of a 300mm wafer, cooling costs of a gpu are practically immaterial. Think about.

2) A less complex gpu requires lower R&D spend and possibly less respins to get the bugs out/appropriate frequency, etc. There is absolutely no question that it's a lot more risky from a manufacturing and business perspective to attempt to design a large monolithic core that will be outperform the competition.

- you have issues of downscaling those cores to smaller ones while when you have a mid-range gpu, you just package them together (software becomes the scaling problem)

- what happens if your gpu doesn't outperform the competition? We know how this worked out for HD 2900XT...

3) A dual-gpu card will not necessarily run any hotter or use any more power than a large die gpu like GTX 280.

In other news,

"The Radeon 4850 features a 625 MHz core clock and GDDR3 clock in excess of 2000MHz. Corporate documentation explains that the 480 stream processors on the RV770 processor offer considerable enhancements over the 320 stream processors found in the RV670 core, though AMD memos reveal little about how this is accomplished. "

So much for 800 shaders.........

Source

I have to agree with these points.
All very good reasons for the multi gpu approach ATI taken. Since R/D already goes into developing CF (tech or drivers), why not make use of this for multiGPU aswell ...

Multi GPU is here to stay be it on a single card or SLI/CF.

On a side not ... I read a rumour that AA was no longer being done in the shaders (which apparently was troublesome for the 2 and 3 series) , anybody know how this could impact performance?

 

[Kuzi]

Originally posted by: RussianSensation

Originally posted by: airhendrix13

From a Manufacturing standpoint, having a single GPU would be better for them. It runs cooler (usually), uses less power (again usually), and thus in the end would end up saving them money on cooling and such.

I think you are missing some key points:

1) A less complex gpu has higher yields which MORE than offset cooling costs. Compared to wasted wafer space and # of working gpus one can pump out of a 300mm wafer, cooling costs of a gpu are practically immaterial. Think about.

2) A less complex gpu requires lower R&D spend and possibly less respins to get the bugs out/appropriate frequency, etc. There is absolutely no question that it's a lot more risky from a manufacturing and business perspective to attempt to design a large monolithic core that will be outperform the competition.

- you have issues of downscaling those cores to smaller ones while when you have a mid-range gpu, you just package them together (software becomes the scaling problem)

- what happens if your gpu doesn't outperform the competition? We know how this worked out for HD 2900XT...

3) A dual-gpu card will not necessarily run any hotter or use any more power than a large die gpu like GTX 280.

In other news,

"The Radeon 4850 features a 625 MHz core clock and GDDR3 clock in excess of 2000MHz. Corporate documentation explains that the 480 stream processors on the RV770 processor offer considerable enhancements over the 320 stream processors found in the RV670 core, though AMD memos reveal little about how this is accomplished. "

So much for 800 shaders.........

Source

XS forums tested the 4850, GPU-z showed it having 800 SPs.

Scroll down to 4th post:

http://www.xtremesystems.org/f...howthread.php?t=191096

 

[Extelleron]

Originally posted by: Calculator83

Originally posted by: Extelleron
I'm not going to argue with you anymore Wreckage. It's clear you and I have different opinions on what AMD/nVidia are doing and nothing that either of us say is going to change that.

BTW Wreckage I just bought an eVGA GTX 280, because I am such a ATI fanboy.

Now I have to hope HD 4870 doesn't touch the GTX 280 at it's $300 price

Dude, Think about it this way,, Thats a bad bet you're making.

4850 = 200 ,, 4870 = 300

If we're assuming it is worth the 1.5x the money, Then we'd More or less see at least 25 to 30% gain in performance. WHICH means, as ATI has already quoted the 4850 as up to 75% of the 280 Performance. Than the 4870 MUST be at least 100% to 105%.

These are terrible Odds ur betting against. especially with 650 dollars.

I don't think ur this crzy.. U didn't buy one at all did you?,, U just said that to shut him up.. LIAR.

I bought one for $429 on eBay (check out Hot Deals. 35% off anything Buy It Now). No way would I have spent $650. For that price though, who could resist. If 4870 were on there for $300 - 35%, I would have bought that instead, but unfortunately not.

Originally posted by: Kuzi

Originally posted by: RussianSensation

Originally posted by: airhendrix13

From a Manufacturing standpoint, having a single GPU would be better for them. It runs cooler (usually), uses less power (again usually), and thus in the end would end up saving them money on cooling and such.

I think you are missing some key points:

1) A less complex gpu has higher yields which MORE than offset cooling costs. Compared to wasted wafer space and # of working gpus one can pump out of a 300mm wafer, cooling costs of a gpu are practically immaterial. Think about.

2) A less complex gpu requires lower R&D spend and possibly less respins to get the bugs out/appropriate frequency, etc. There is absolutely no question that it's a lot more risky from a manufacturing and business perspective to attempt to design a large monolithic core that will be outperform the competition.

- you have issues of downscaling those cores to smaller ones while when you have a mid-range gpu, you just package them together (software becomes the scaling problem)

- what happens if your gpu doesn't outperform the competition? We know how this worked out for HD 2900XT...

3) A dual-gpu card will not necessarily run any hotter or use any more power than a large die gpu like GTX 280.

In other news,

"The Radeon 4850 features a 625 MHz core clock and GDDR3 clock in excess of 2000MHz. Corporate documentation explains that the 480 stream processors on the RV770 processor offer considerable enhancements over the 320 stream processors found in the RV670 core, though AMD memos reveal little about how this is accomplished. "

So much for 800 shaders.........

Source

XS forums tested the 4850, GPU-z showed it having 800 SPs.

Scroll down to 4th post:

http://www.xtremesystems.org/f...howthread.php?t=191096

I still think RV770 has 800SP (how else would they have gotten 1 TFLOP? Either that or those "480" are completely different than those in R600) but remember GPU-Z is just a database. So if the guy who makes GPU-Z thinks it has 800SP, then it will show up that way in the program. It doesn't actually detect anything from the card except clocks.

 

[Grinja]

http://www.tweaktown.com/artic...the_gtx_280/index.html

Not sure if this has been posted already ...
The author seems to be really optomistic about these cards ... I hope we get some descent competition this time round.

 

[Kuzi]

@Extelleron

At the same link, check the 23rd post, shows an AMD slide. Does not seem fake to me, especially as you mentioned, how can they get to the 1 Teraflop number with only 480 SPs.

 

[Quiksilver]

Originally posted by: Kuzi
@Extelleron

At the same link, check the 23rd post, shows an AMD slide. Does not seem fake to me, especially as you mentioned, how can they get to the 1 Teraflop number with only 480 SPs.

480 Strong SP's? Instead of 160 Strong?

Also that slide in that post looks obviously fake or was saved using mspaint which would make no sense.

I'm willing to bet the 800SP mark applies to 4870X2 and not 4870.

 

[schneiderguy]

Originally posted by: Kuzi

Does not seem fake to me, especially as you mentioned, how can they get to the 1 Teraflop number with only 480 SPs.

Run the shaders at a higher clock than the rest of the card like nvidia does in G80 and its derivatives.

480 Shaders * ~1.1ghz shader domain clock = a little over 1tflop

Originally posted by: Quiksilver

480 Strong SP's? Instead of 160 Strong?

800sp = 160 vec5 sp
480sp = 96 vec5 sp

 

[Munky]

GPU-z just has a pre-baked database from which it reads info, it has no way of knowing how many shaders or texture units are actually present. So it basically means nothing...

But I'm still sticking to the 800sp theory. It's actually more plausible than the separate shader clocks theory.

Also, I bet AA is still done in the shaders. The problem with r6xx AA and AF performance is not in the shaders, but rather the ROP's and TU's. If Ati has addressed those issues (as the latest rumors seem to indicate...) then the AA/AF performance problem will be fixed. Shader-based AA is the future, there's no point going back to fixed-function AA. Sooner or later, Nvidia will also adopt this approach to AA.

 

[SlowSpyder]

Originally posted by: munky
GPU-z just has a pre-baked database from which it reads info, it has no way of knowing how many shaders or texture units are actually present. So it basically means nothing...

But I'm still sticking to the 800sp theory. It's actually more plausible than the separate shader clocks theory.

Also, I bet AA is still done in the shaders. The problem with r6xx AA and AF performance is not in the shaders, but rather the ROP's and TU's. If Ati has addressed those issues (as the latest rumors seem to indicate...) then the AA/AF performance problem will be fixed. Shader-based AA is the future, there's no point going back to fixed-function AA. Sooner or later, Nvidia will also adopt this approach to AA.

Along those lines, I'm pretty sure that I've read somewhere that DX10.1 requires AA to be done through the shaders. Anyone know if this is the case? I could be wrong, but I'm fairly sure I've heard/read that.

 

[semisonic9]

God, they really need to lock this thread and start again already.

~S

 

[ViRGE]

All of you need to take a chill pill. The next person I deem to be acting like a partisan hack is going on a vacation until after the RV770 launch.

If you make a claim, prove it. If you have a difference of opinion, agree to disagree. If you don't like what the other person has to say, don't just call them a fanboy.

 

[Stoneburner]

Originally posted by: Grinja
http://www.tweaktown.com/artic...the_gtx_280/index.html

Not sure if this has been posted already ...
The author seems to be really optomistic about these cards ... I hope we get some descent competition this time round.

I almost feel bad for the author. He seems ready to burst with information but is being held back. What has ati wrought with the 4850?

 

[SlowSpyder]

Originally posted by: Stoneburner

Originally posted by: Grinja
http://www.tweaktown.com/artic...the_gtx_280/index.html

Not sure if this has been posted already ...
The author seems to be really optomistic about these cards ... I hope we get some descent competition this time round.

I almost feel bad for the author. He seems ready to burst with information but is being held back. What has ati wrought with the 4850?

I don't know why AMD is holding back the info at this point, the new Nvidia stuff is out and for sale. Lift the NDA already.

 

[ronnn]

Found this
in this thread.

If a boring repeat, please delete.

 

[Hauk]

ATI 4850 CROSSFIRE QUIK REVIEW

IMPRESSIONS
This is my first experience with ATI. I've owned nV cards since the birth of Geforce. With that said, a quik review is in order..

These cards feature good build quality, with a single slot solid copper heatsink. In one pic below you'll notice exaust vents on the side as well as the back of the heatsink. Pics we've seen don't easily reveal these exaust slots. I'd say nearly half the exaust blows thru these directed towards side panel. You'll see I mounted a 120mm fan near the front of the card avoiding the exaust slots. The fans never spun up past default speed (whatever that is). There are no fan speed adjustments in CCC. Overclocking beyond 675 core caused 3DMark to lock up so I backed them to 660 core. There is no shader adjustment in the overdrive panel.

With good airflow they're idling around 63c, with load temps about 74c. Under load, the area of the heatsink you can touch will burn your digits after two seconds or so. They ran very stable however, with zero lockups 660 core.

The driver appears to throttle down the cards to 500 core 750 mem with no load. See CCC pic below under current clock settings. GPU-Z also reports the drop.

So far I'm impressed with the results. Perhaps final drivers will yield additional gains. I'm sure we'll see third party oc and fan adjustments. Perhaps shaders will be adjustable too.

I was ready to pull the trigger on a GTX 280 if it launched at $499. I also considered a GTX 260 but decided to see what kind of performance I could get for $400 bucks with ATI. I plan to sell these off when 4870 X2 arrives. I've thrown a few thousand dollars nV's way; but it looks like I may be seeing red for a while. ATI appears to be pulling it together this round and I've decided to support them at least for at least a generation.

Hope the info helps everyone. Here's to making an informed decision... :beer:


TEST SYSTEM

- Dell 24" Ultrasharp 2408WFP
- Antec P182
- C2D E8400 @ 3.80
- Asus Rampage Formula x48
- 8GB Muskin Redline DDR2 1000 @ 1015 5-5-5-12, 7 performance value (per AT Rampage board review)
- PowerColor 4850's in CF clocked to 660 core 1050 mem
- Thermaltake Toughpower 850W PSU
- Catalyst 8.6 Betas
- Vista 64 Premium SP1



PICS
4850
Twins
Exaust Vents
Installed in Rig
Installed in Rig
Overkill Airflow
CCC
CPU & GPU Screens



BENCHIES

3DMARK06:
18059 3DMarks
* This is with 8.6 beta, perhaps finals will improve score further

VANTAGE build 1.0.1:
P10431
GPU = 11666
CPU = 7916

X5314
GPU = 5226
CPU = 7831


MASS EFFECT:
Testing tonight but was able to run 1920 x 1200 highest settings, vsync disabled, averaged 62fps
Fraps = 62fps


CRYSIS:

1024 x 768, dx9, high, no AA
Min = 28.68
Max = 96.29
Avg = 69.44

1280 x 1024, dx9, high, no AA
Min = 32.59
Max = 86.31
Avg = 62.67

1680 x 1050, dx9, high, no AA
Min = 26.83
Max = 56.83
Avg = 43.84

1900 x 1200, dx9, high, no AA
Min = 24.93
Max = 56.83
Avg = 43.84

1024 x 768 dx9, high 4x AA
Min = 26.28
Max = 81.06
Avg = 59.21

1280 x 1024 dx9, high, 4x AA
Min = 25.47
Max = 65.51
Avg = 49.07

1680 x 1050 dx9, high, 4x AA
Min = 47.53
Max = 47.53
Avg = 36.39

1900 x 1200 dx9, high, 4x AA
Min = 16.16
Max = 37.83
Avg = 29.97

 

[Quiksilver]

Originally posted by: Stoneburner

Originally posted by: Grinja
http://www.tweaktown.com/artic...the_gtx_280/index.html

Not sure if this has been posted already ...
The author seems to be really optomistic about these cards ... I hope we get some descent competition this time round.

I almost feel bad for the author. He seems ready to burst with information but is being held back. What has ati wrought with the 4850?

hmmm...
The GTX 280 is slower than a pair of CrossFired 4850s for the most part

That sounds intriguing.

I take it crossfired 4870's are gonna wipe the floor, and the 4870X2 should be about 10-15% faster than the crossfired 4870's due to crossfire optimizations and no crossfire bridge and what not.

I say this si gonna get interesting.
I'm wondering if 4870X2 is aiming at demolishing the GTX280 or the SLI'd GTX280...

 

[Janooo]

Originally posted by: SteelSix
I'll post this in the upcoming 48xx Review & Benchmark Thread when created...

I'm just starting to benchmark 4850's in Crossfire and want to share early results. Full benchmarks including Crysis to follow...
...

I hope you are not planning to sleep tonight.
Keep it up trooper.

 

[taltamir]

Originally posted by: ronnn
Found this
in this thread.

If a boring repeat, please delete.

finally a demo showing an awesome looking character in AN ENVIRONMENT instead of being rendered in a void.

 

[Foxery]

Well, this should make for an interesting product launch - individuals testing cards before the news media is allowed to print. The few hints I've seen so far suggest that ATI will greatly embarass nVidia in the "bang for your buck" area, so they'd better have a good reason for the delay.

I wonder what the overall impact on their sales figures will be when there is early availability to those who know where to look, but also a fair share of consumers jumping on nVidia before all of the facts are in.