ATi 4xxx Series Review

[AzN]

Originally posted by: Extelleron

Originally posted by: Aberforth

Originally posted by: airhendrix13
PLACE YOUR BETS! I bet the 280GTX will be marked down $75 by the end of next week. It's $649.99 right now at the egg, I expect to see it it at $574.99 by Saturday 12pm.

Anyone else got a guess?

Does it matter?...just CF your 4850 for $400 and beat 280 by a fair margin.

Except in games that don't scale with Crossfire, like Bioshock, ET:QW, and The Witcher. Even when CF does scale.... the 4850's don't really beat the 280. GTX 280 & HD 4850 CF are pretty much tied most of the time, the advantage is of course the 4850's are cheaper.

I'm still going to take a GTX 280 over HD 4850 CF at the same price. What will be awfully temping at the high end is HD 4870 Crossfire. The HD 4870 will be powerful enough as a single-GPU that when CF doesn't scale it won't be such a big deal. And when it does scale, the results will be pretty amazing.

3870 was the same way until AMD fixed most of CF issues with RV670. Just give it some more time so they can iron out the issues.

 

[AzN]

Anyway I think ATI engineers really have outdone themselves. With current generation of games it performs as good as 9800gtx. Really good AA performance. Very efficient and lower power consumption than 9800gtx. With 1 teraflop of processing power it will outlast a 9800gtx for sure when more shader intensive graphic heavy games start to appear.

 

[dajini92]

Dang... just spent the past couple hours reading through this entire thread, and all the links posted. From what I've seen, the performance scale (typically) goes 9800gtx, HD 4850, 9800gtx+, in order from lowest to highest. Now there are some tests where the 9800gtx pulls ahead of ATI's 4850, but as a general rule the ATI performs better. Sheesh, and I had just decided on getting an 8800GTS... guess I'm gonna wait until mid-July and get a 9800gtx+ instead, or maybe an ATI 4870?... these are interesting times as far as video cards go, and as everyone else said, there's never been a better time to be a buyer. =D

 

[golem]

Originally posted by: Azn

Originally posted by: ShadowOfMyself
Wow, the Nvidia fanboyism is strong here... Wreckage, Rollo, and Chizow too! I find it hilarious that you dont recognize how much the 4850 spanks the 9800 GTX in AA (which is what matters) in 80% of the reviews out there, and you cherry pick the one that shows different results... So predictable

I dont know how someone can be so blind, and for a company that doesnt give a shit about consumers

They love sticking up for 9600gt when 3870 was in same shoes as 9800gtx vs 4850 although 3870 was much stronger raw performer than 9600gt. Only when the situation is reversed for Nvidia do they stick up 9800gtx.

Well, then you are the same except for ATI. When 9600GT was a better performer with AA and higher rez, you stuck up for 3870...

 

[bryanW1995]

Originally posted by: nRollo

Originally posted by: bryanW1995
atreus21 was commenting that he's impressed with 4850 performance. He's not the only one...

@op 4850 xfire should be = gtx 280 when you factor in the games that DON'T scale very well. in games that do scale, it should be quite a bit faster. check out your favorites and see how they do in xfire, if they all scale well then you'll be gtg. what mobo do you have btw?

I think the AT review pretty effectively demonstrates why CF costs less than a GTX280-
they got no/very poor scaling in 3 of 7 games tested.

When you throw microstutter, mouse lag, and other multi card issues into the mix, it becomes apparent that the 4850s are priced appropriately.

That said, I think the 4850 looks to be a very solid performer at it's price point as a single card, and in some games transcends it's price point in CF.

ok, somebody pull up all of rollo's posts from the 9800gx2 launch period. I'm too lazy to do it, but something tells me that multi-gpu scaling issues weren't brought up then.

on a happier note, rollo and keys please tell your friends at the green team that putting f@h on the g80 and newer gpus was a GREAT idea. This makes it that much harder to not root for them. Of course, we do have 1:35 am emails to anand talking about the 9800gtx+ that remind us this is still a business...

 

[bryanW1995]

Originally posted by: LightningRider
What are the best ATi brands anyways? It's been so long since I've thought of buying an ATi card.

visiontek is the only one that I know of with a lifetime warranty. sapphire is the most popular brand, and HIS is pretty good, especially with their iceQ versions.

 

[AzN]

Originally posted by: golem

Originally posted by: Azn

Originally posted by: ShadowOfMyself
Wow, the Nvidia fanboyism is strong here... Wreckage, Rollo, and Chizow too! I find it hilarious that you dont recognize how much the 4850 spanks the 9800 GTX in AA (which is what matters) in 80% of the reviews out there, and you cherry pick the one that shows different results... So predictable

I dont know how someone can be so blind, and for a company that doesnt give a shit about consumers

They love sticking up for 9600gt when 3870 was in same shoes as 9800gtx vs 4850 although 3870 was much stronger raw performer than 9600gt. Only when the situation is reversed for Nvidia do they stick up 9800gtx.

Well, then you are the same except for ATI. When 9600GT was a better performer with AA and higher rez, you stuck up for 3870...

3870 was clearly better raw performer than 9600gt about 10-20% depending on the game. I can't say the same thing about 9800gtx vs 4850. Not to mention 3870 and 4850 has something both 9600gt and 9800gtx doesn't. Much better shader performance for future shader intensive titles to come. I've stuck with 8800gt performance no doubt over 3870 or 9600gt. It's called thinking logically something these guys can't do since they are probably being paid to do so.

edit:

If you actually analyzed some of these reviews 4850 has faster FP16 blending fillrate than 9800gtx for future titles that use more of HDR based scenes. It actually surpasses 280gtx while 9800gtx has more of the 8bit integer texture fillrate for older titles. Now look at all the shader power of what 4850 can do. Easily matching or surpassing 260gtx. We aren't that shader limited to begin with but it's slowly creeping as each future title raises the bar. 4850 and 3870 will only get better. Something I can't say about its competition. It is the same reason why I chose a 8800gs over 9600gt. More texture fillrate and shader even though AA performance sucks. Easily the better raw performer when overclocked.

 

[bryanW1995]

Originally posted by: Aberforth

Originally posted by: airhendrix13
PLACE YOUR BETS! I bet the 280GTX will be marked down $75 by the end of next week. It's $649.99 right now at the egg, I expect to see it it at $574.99 by Saturday 12pm.

Anyone else got a guess?

Does it matter?...just CF your 4850 for $400 and beat 280 by a fair margin.

I think $550...oops, nevermind...

chizow, why are you in here thread crapping? you bought a gtx 280. you are the TARGET MARKET for the gtx 280: somebody upgrading from an 8800gtx. You made the right decision for you. Unless you're gunning for focus group membership I don't see why you continually trash ati cards. Don't get me wrong, daamit has given us plenty of reason in the past 18 months to warranty castigation, but today puts an end to that. Did they hit a home run? no, probably the best way to look at it is that they hit the ball really far, but nvidia's outfielder knocked it back into the park with their 9800gtx+ release and forced them into a triple. Either way, it's still a fantastic card and a fantastic launch. It will be better for all non-focus group members who actually buy their cards (ignore this keys b/c I firmly believe in your impartiality) to have strong competition between gpu makers. Quit pissing on cards that you have no interest in and go start running f@h on your new gtx 280! btw, we're team # 198.

 

[bryanW1995]

Originally posted by: dajini92
Dang... just spent the past couple hours reading through this entire thread, and all the links posted. From what I've seen, the performance scale (typically) goes 9800gtx, HD 4850, 9800gtx+, in order from lowest to highest. Now there are some tests where the 9800gtx pulls ahead of ATI's 4850, but as a general rule the ATI performs better. Sheesh, and I had just decided on getting an 8800GTS... guess I'm gonna wait until mid-July and get a 9800gtx+ instead, or maybe an ATI 4870?... these are interesting times as far as video cards go, and as everyone else said, there's never been a better time to be a buyer. =D

you, my friend, have just been "nvidia'd"

 

[tuteja1986]

Originally posted by: bryanW1995

Originally posted by: dajini92
Dang... just spent the past couple hours reading through this entire thread, and all the links posted. From what I've seen, the performance scale (typically) goes 9800gtx, HD 4850, 9800gtx+, in order from lowest to highest. Now there are some tests where the 9800gtx pulls ahead of ATI's 4850, but as a general rule the ATI performs better. Sheesh, and I had just decided on getting an 8800GTS... guess I'm gonna wait until mid-July and get a 9800gtx+ instead, or maybe an ATI 4870?... these are interesting times as far as video cards go, and as everyone else said, there's never been a better time to be a buyer. =D

you, my friend, have just been "nvidia'd"

What most review fail to do is, do detail benchmarking. IF they did they would find out the 4850 are monster when using. It only few FPS from going from 4xAA to 8xAA. Also this thread haywire when they combined other threads to this thread so as your reading it gets weird.

Some game it looses half its FPS from going to 4xAA to 8xAA and in other its unplayable.
http://www.computerbase.de/art...tt_unreal_tournament_3

Unreal Tournament 3 1600x1200 4xAA/16xAF:
GTX 280 : 145
HD4850 : 65
9800GTX : 81

Unreal Tournament 3 1600x1200 8xAA/16xAF:
GTX 280 : 72
HD4850 : 63
9800GTX : 34

---------------------------------------------------------------
http://www.computerbase.de/art..._clive_barkers_jericho

Clive Barker's Jericho 1600x12004xAA/16xAF:
GTX 280 : 53
HD4850 : 29
9800GTX : 25

Clive Barker's Jericho 1600x1200 8xAA/16xAF:
GTX 280 : 26
HD4850 : 28
9800GTX : 7


---------------------------------------------------------------
http://www.computerbase.de/art...v770/8/#abschnitt_fear

FEAR 1600x1200 4xAA/16xAF:
GTX 280 : 147
HD4850 : 92
9800GTX : 83

FEAR 1600x1200 8xAA/16xAF:
GTX 280 : 93
HD4850 : 60
9800GTX : 49


---------------------------------------------------------------
http://www.computerbase.de/art...schnitt_call_of_juarez

Call of Juarez 1600x1200 8xAA/16xAF
GTX 280 : 30
HD4850 : 21
9800GTX : 20

Call of Juarez 1600x1200 8xAA/16xAF
GTX 280 : 24
HD4850 : 17
9800GTX : 6


---------------------------------------------------------------
http://www.computerbase.de/art...nitt_company_of_heroes

Company of heroes 1600x1200 4xAA/16xAF
GTX 280 : 84
HD4850 : 46
9800GTX : 47

Company of heroes 1600x1200 8xAA/16xAF
GTX 280 : 66
HD4850 : 42
9800GTX : 35

---------------------------------------------------------------
http://www.computerbase.de/art...#abschnitt_lost_planet

Lost planet 1600x1200 4xAA/16xAF:
GTX 280 : 58
HD4850 : 31
9800GTX : 34

Lost planet 1600x1200 8xAA/16xAF:
GTX 280 : 45
HD4850 : 29
9800GTX : 24

 

[dadach]

Originally posted by: ShadowOfMyself
Wow, the Nvidia fanboyism is strong here... Wreckage, Rollo, and Chizow too! I find it hilarious that you dont recognize how much the 4850 spanks the 9800 GTX in AA (which is what matters) in 80% of the reviews out there, and you cherry pick the one that shows different results... So predictable

I dont know how someone can be so blind, and for a company that doesnt give a shit about consumers

too funny to watch them squirm...cant buy stuff like that with any money ))))))

 

[Cookie Monster]

I dont know. Some of it looks fishy. I mean 1~2 FPS drops? are they even applying 8xAA to some of those games? Look at FEAR. 32 FPS reduction when 8xAA was used. Then you have other titles which produces 1~2 FPS hit. Doesn't make much sense to me, seeing as the GTX280 and 9800GTX all share a similiar performance hit across all the titles. Unless someone can actually confirm its doing 8xAA, the numbers look way to optimistic.

 

[tuteja1986]

Originally posted by: Cookie Monster
I dont know. Some of it looks fishy. I mean 1~2 FPS drops? are they even applying 8xAA to some of those games? Look at FEAR. 32 FPS reduction when 8xAA was used. Then you have other titles which produces 1~2 FPS hit. Doesn't make much sense to me, seeing as the GTX280 and 9800GTX all share a similiar performance hit across all the titles. Unless someone can actually confirm its doing 8xAA, the numbers look way to optimistic.

Well the G80/G82 series have long know problem with 8xAA, been talked about in forum few times maybe you weren't taking notice of it. Nvidia also recommends reviewer not to do 8xAA.

 

[Cookie Monster]

Well im not talking about that (yeh its true 8xAA is pretty bad on G8x/G9x hardware but thank god they have other AA modes that can act as alternatives). Im talking about if RV770 is actually doing 8xAA. How can a card take 35% percent reduction in some titles when 8xAA enabled where as it takes less than 5% hit in other titles?

note - I wouldn't claim things like "nVIDIA recommends reviewers not to do 8xAA". Such a flame bait right there.

 

[Sind]

That german site is wonderful, too bad it only goes up to 1600ish, would love to see the AA comparisons past that. And definately I wonder why the drastic performance hit with AA in some games vs hardly any in others from those charts.

 

[tuteja1986]

Originally posted by: Sind
That german site is wonderful, too bad it only goes up to 1600ish, would love to see the AA comparisons past that. And definately I wonder why the drastic performance hit with AA in some games vs hardly any in others from those charts.

Well that was their preview , they didn't have enough time to finish the review as NDA was pushed up. Then fews hrs later some Nvidia sent a emergency 9800GTX+ on very expensive fast delivery mail. The review is missing alot of things that it would have if they had finished it.

highest res setting , high AA mode benchmarking , Crossfire benchmarking , Heat , Noise , Video decoding , IQ testing and more detail analysis of the card. Which will come probably in next few day but this site doesn't do half ass review.

This article is called "Quick test: ATi Radeon HD 4850 (RV770)"

But their quick test is more detail than review done by many website.

 

[dadach]

nvidia is too sad...basically what they are saying to their customers is: "we have been ripping you off...come again" ...what other reason would there be to have so much headroom to drop prices?

 

[ShadowOfMyself]

Originally posted by: dadach
nvidia is too sad...basically what they are saying to their customers is: "we have been ripping you off...come again" ...what other reason would there be to have so much headroom to drop prices?

Cant really blame them for wanting to make as much money as possible as a company... Its just that their methods are horrible... I wish a new gfx company came up out of nowhere and took their spot, then we could all enjoy a healthy video card war with fairplay, not the low blows and shady tactics Nvidia uses

I mean, 9800 GTX+, seriously? LOL! Look out for AMD's 4850++ in a few days! It has plenty of headroom thats for sure... Pf!

Reminds me of when Nvidia marketed a card as having DX9+ and then ATi came up with DX9++, its kinda ridiculous

 

[MarcVenice]

About crossfire scaling in anands article. It scales VERY well in 4 games, and doesn't scale at all in the other 2-3 games. No scaling at all, means it's broken, and it does scale on a HD3870 X2. Looks like it can be fixed if you ask me. It's not ideal for sure, but for $400 to outperform a $650 card, I reckon it's worth it. It's also JUST 7 games, someone should compile a list of dozens of games with and without crossfire.

As for powerconsumption, I wouldn't crossfire 2 HD4850's with something less then a corsair 650tx or something similar. 1000w is mega pure overkill. Perhaps 750w to play it safe. For sli-ed 280gtx's, you might actually need a 1000w PSU in order to boot. I suppose that's a bit of a negative for nvidia, 1000w psu's dont exactly come cheap.

 

[SniperDaws]

Originally posted by: Azn
Anyway I think ATI engineers really have outdone themselves. With current generation of games it performs as good as 9800gtx. Really good AA performance. Very efficient and lower power consumption than 9800gtx. With 1 teraflop of processing power it will outlast a 9800gtx for sure when more shader intensive graphic heavy games start to appear.

It pains me to say this but for once Azn i agree with you, this doesnt mean we can go for drinks though ok mate!

 

[SniperDaws]

Whats the Diffrence between 8xAA and 8xQAA ? as in performance and quality.

 

[Aberforth]

Originally posted by: SniperDaws
Whats the Diffrence between 8xAA and 8xQAA ? as in performance and quality.

Quincunx AA is a supersampling algorithm that begins at the center(5 pixel grid),so it will have shared points. This will improve the performance but usually blurs the edges and doesn't improve the quality. Most generic AA method involves 2x2 grid along X-Y axis which improves the quality but reduces the performance- this method is also VRAM hungry.

 

[BFG10K]

Quincunx AA is a supersampling algorithm

Quincunx is not super-sampling. It's a post-filter that uses pixel sampling from neighboring pixels like ATi?s CFAA does

 

[AzN]

Originally posted by: SniperDaws

Originally posted by: Azn
Anyway I think ATI engineers really have outdone themselves. With current generation of games it performs as good as 9800gtx. Really good AA performance. Very efficient and lower power consumption than 9800gtx. With 1 teraflop of processing power it will outlast a 9800gtx for sure when more shader intensive graphic heavy games start to appear.

It pains me to say this but for once Azn i agree with you, this doesnt mean we can go for drinks though ok mate!

Now why would I drink with someone ignorant as you?

 

[AzN]

Originally posted by: SniperDaws
Whats the Diffrence between 8xAA and 8xQAA ? as in performance and quality.

8xCSAA uses 4 stencil samples using 8 quality value.

As for Q it uses 8 stencil samples and quality values of 8 and 16 just like regular CSAA.


Regular 8xCSAA is maybe 10% drop in performance compared 4xAA. It does improve ever slightly compared to 4xMSAA. Q uses more samples so it takes whole lot more vram and bandwidth which is much superior to 8xCSAA.