ATI Radeon HD 4000 Series discussion

[Helmore]

Well I don't want to be rude, but you can't really deny that you in the end turned out to be wrong. There are no serious issues.

Ok, so let's now all forget this little dispute has ever happened and talk about what R700 might bring.....

[hellcamino]

Seriously...all the doom and gloom I'm glad I bought an nvidia cards stuff is getting really old . For those that have " bad " cards post what you did to cause / rectify the problem or if it was simply bad from the manufacturer . For those of you with GTX280's...you should be doing the same thing as opposed to smack talking another product . I myself hold loyalty to NO brand , my only loyalty is to performance . The 4870X2 is shaping up to be one hell of a good card and at this time I am planning on buying a pair as well as card cover blocks for them .

[eleeter]

Would be nice to see this happen. XS has degenerated into name calling and personal insults for anyone who dares to question the latest fad.

You're not fooling anyone. You love to troll and knock AMD at every turn. You don't care about the facts, you just jump on the bandwagon and bad mouth as much as you can, all the while claiming you have "proof"

Guess what, you were WRONG, again. So stop trolling and being an asswipe.


http://www.xtremesystems.org/forums/...postcount=3982
http://www.xtremesystems.org/forums/...postcount=3994
http://www.xtremesystems.org/forums/...65#post3099665
http://www.xtremesystems.org/forums/...postcount=4037
http://www.xtremesystems.org/forums/...0&postcount=31

[disruptfam]

sorry if this have been asked before.But is a e8400 @ 4ghz a good match for a 4870x2? bottleneck wise?

[zerazax]

sorry if this have been asked before.But is a e8400 @ 4ghz a good match for a 4870x2? bottleneck wise?

Unless you game below 1680 x 1050, I doubt a 4870x2 will be bottlenecked by the CPU at all.

[annihilat0r]

Oops ceevee, looks like this wasn't a "major major problem"! People who thought their cards had died now got their cards back thanks to W1zzard's fix!

People getting their dead cards back on track is a happy thing. Except for you. You must be pretty sad the guys got their cards back. I don't think anyone could doubt that.

Happy gaming with your GTX 280 which gives something like 10 FPS more than 4870, if not beaten by it, for an additional fee of only $350! The way it's meant to be played, right?

[twwen2]

Oops ceevee, looks like this wasn't a "major major problem"! People who thought their cards had died now got their cards back thanks to W1zzard's fix!

People getting their dead cards back on track is a happy thing. Except for you. You must be pretty sad the guys got their cards back. I don't think anyone could doubt that.

Happy gaming with your GTX 280 which gives something like 10 FPS more than 4870, if not beaten by it, for an additional fee of only $350! The way it's meant to be played, right?

Ouch! That's gotta hurt

[perkam]

I'm gonna ask everyone to cease and desist with the ceevee-bashing.

His comments can be NV-biased sometimes, but they are far from being worth derailing the whole thread. This is the precise reason that I had asked for this thread to be removed. The longer it goes, the less of a purpose it will have. Members wished for this thread to be kept on for R700 discussion and that is what this is for.

There is NO general ATi or Nvidia discussion areas for exactly this reason. Please discuss the HD 4870 issues in the Video Card section, HD 4870 performance in the HD 4870 Review thread and likewise for the 4850.

Perkam

[Direwolf]

With 3rd party cooling the 4850/4870 cards will be reaching amazing clocks very soon. 950mhz core is pretty insane...

in short, I cant wait to see whats in store for us within the next two months

[annihilat0r]

Okay, I discuss R700:

There are rumors it'll be %15 faster than a 4870 CF setup and I say that should be impossible with the downclock it will have.

If the rumors are true, though, and the card sells for $499, and has no microstutter related issues (I can live with some games not benefiting at all, there are only a little such games anyway; but microstuttering destroys the gains and is there for most of the games) I will only then buy it.

Microstutter can be overcome only by delaying the GPU for a small amount of time so that both GPUs don't render the frames in the same time. And this will drag FPS down somewhat. So with the reported %15 increase, it's impossible to achieve.

[Caveman787]

I don't think it'll be downclocked...remember the leaked x12000 vantage score.

That was with 2 and the screenshot showed 750 mhz core....so It would seem that it won't be downclocked really.

[annihilat0r]

Then, a %5 increase in clocks resulting in a %15 increase in performance is definitely possible thanks to the GPUs being on a single card, but I still think solving the microstutter problem would mean a sizeable FPS loss.

And microstutter = no buy for me, after what I'm seeing in my 8800GT SLI rig.

[GoldenTiger]

With 3rd party cooling the 4850/4870 cards will be reaching amazing clocks very soon. 950mhz core is pretty insane...

in short, I cant wait to see whats in store for us within the next two months

When I was fiddling with the BIOS overclocking of my 4870, I had it at 870/4500 on stock cooling, 32ish % fan... it was getting very hot but running stable, I went to reflash to lower clocks to just game with but had a crash from heat after bootup at lower clocks, shut down and of course the rest is history... truncated BIOS = no bootup, until W1zzard came to the rescue . I'm not going to try those clocks again until a software OC utility comes, but once it does I have a feeling 45-50% fan would keep it stable . I had run it for a good 15 minutes without game issues until it was getting so hot I got worried.

[Calmatory]

Anyone have any estimates about the absolute max RV770XT chip can go? e.g. any possibility to reach 1GHz with DICE/LN2 + insane volts? What about air?

[cadaveca]

1ghz should be doable on water, provided cards can provide enough power. I tihnk the limit will not be the chips, but how much power they can have delivered.


Makes me wonder about boards with 4 slots PCI-E 2.0 8x, and 4x 4870's VS 2x "R700".

[twwen2]

Anyone have any estimates about the absolute max RV770XT chip can go? e.g. any possibility to reach 1GHz with DICE/LN2 + insane volts? What about air?

Yeah i'd be interested to know how it responds to voltage increases.

[Calmatory]

Then, a %5 increase in clocks resulting in a %15 increase in performance is definitely possible thanks to the GPUs being on a single card, but I still think solving the microstutter problem would mean a sizeable FPS loss.

And microstutter = no buy for me, after what I'm seeing in my 8800GT SLI rig.

If the clocks are increased by 5%, there can't be more than 5% gain. No way of doing that. 5% is the theoretical maximum, and it can never be exceeded. It doesn't matter if there are one or two cores.

...unless CF somehow starts to work more efficiently at different clockspeeds, which is more than awkward. And 5% core, 10% shader and 15% mem increase can't increase the performance more than 15% at max, no 30% performance increases. :P

[cadaveca]

Depends on how the card is truly limted, actually. You assume that clockspeed is the inefficiency, not something else...

[zerazax]

There's a lot more to how a card performs than just clock speed... once you factor in caches, pipeline efficiency, dispatch efficiency, etc. core clocks are just one part of the equation.

[grimREEFER]

im still microstutterring-agnostic.
so one german site writes a page theory about a phenomenon with crossfire 3870's that cant be measured by any means and all of a sudden its a truth?
am i the only skeptic here? there are many explanations for stuttering that haven't been ruled out. (lack of bandwith on the pci express slots they used? lack of memory caused by using a 32bit os and adding another 512mb of video memory?)
the site says its caused by an inherent flaw in afr rendering, so the 4870x2 should have microstuttering unless they come up with a new system, which would be a little chaotic. then they would have to work on making games compatible for crossfire and simultaneously make those games compatible for the 4870x2 and other newer x2 cards.

[Calmatory]

There's a lot more to how a card performs than just clock speed... once you factor in caches, pipeline efficiency, dispatch efficiency, etc. core clocks are just one part of the equation.

Still it can't be possible to gain over 100% efficiency, unless the clockspeed somehow makes the chip more efficient as in cache/memory management etc. Thus it is impossible to get more than 100% increase in performance from 100% overclock etc.

Depends on how the card is truly limted, actually. You assume that clockspeed is the inefficiency, not something else...

Still over 100% efficiency in gain is impossible.

Also another "funny" thing is that people now talk about GDDR5 overclocks. "If you OC from 900 to 1000, you don't overclock 100 MHz, but 400 MHz Instead! From 3600 MHz to 4000 MHz!", which is obviously false. Doesn't matter if you OC 4 MHz or 4 GHz, if the increase is only 4 %, besides the chips still run at 900/1000MHz, not 3600/4000.

[grimREEFER]

If the clocks are increased by 5%, there can't be more than 5% gain. No way of doing that. 5% is the theoretical maximum, and it can never be exceeded. It doesn't matter if there are one or two cores.

...unless CF somehow starts to work more efficiently at different clockspeeds, which is more than awkward. And 5% core, 10% shader and 15% mem increase can't increase the performance more than 15% at max, no 30% performance increases. :P

well, firstly, you should make some observations between clock speed and fps at different frequencies to see if you can assume percent change in clock speed equals percent change in performance.
my guess is that it doesn't, i think it's a far more mysterious relationship.

[zerazax]

Still it can't be possible to gain over 100% efficiency, unless the clockspeed somehow makes the chip more efficient as in cache/memory management etc. Thus it is impossible to get more than 100% increase in performance from 100% overclock etc.



Still over 100% efficiency in gain is impossible.

Technically it is. Eric Demers posted this

As for super-linear performance, it's possible -- The actual amount of cache is increased as well as the number of functional units. That means that it's very possible for data to stay in cache longer in xfire, than in regular rendering on a single card. That will lead to performance higher than 2x.

On average, we are seeing scaling in the 1.75 ~ 1.85 range for xfire. Of course, this is at higher resolutions and/or higher settings.

So it sounds like cache can cause performance to scale greater than linearly

[G0ldBr1ck]

R700 having a much better interconnect between GPU's than with 2 separate cards allready gives the R700 a performance advantage over Crossfired 4870's with matching clocks.....therfore annihilat0r is correct that with a 5% clock increase on R700 could very well have a 15% performance gain over 2 4870's in crossfire.

in theary thats: Sideport comunication=10% gain over xfire
5% higher clockspeed=5% gain
=15%

[disruptfam]

Unless you game below 1680 x 1050, I doubt a 4870x2 will be bottlenecked by the CPU at all.

thanks