ATI 4xxx Series Thread

[Jax Omen]

Originally posted by: chizow
If it doesn't double 3870s performance than it'll be doomed from the outset, as GT200 does promise a true doubling of G92's main specs and overall performance.

Now you're just pulling stuff out of your ass. Provide one link to support this. Oh wait, you can't, because plausible GT200 specs haven't been leaked yet. It's all conjecture right now.

What we have is plausible 4870 specs. They look to significantly improve on the 3870's performance. Will it be enough to compete with GT200? Only Nvidia knows that right now.

 

[SlowSpyder]

Originally posted by: Jax Omen

Originally posted by: chizow
If it doesn't double 3870s performance than it'll be doomed from the outset, as GT200 does promise a true doubling of G92's main specs and overall performance.

Now you're just pulling stuff out of your ass. Provide one link to support this. Oh wait, you can't, because plausible GT200 specs haven't been leaked yet. It's all conjecture right now.

What we have is plausible 4870 specs. They look to significantly improve on the 3870's performance. Will it be enough to compete with GT200? Only Nvidia knows that right now.

Slightly off topic, but along the lines of what was said above, do we know if the GT200 is supposed to be a new architecture, or another sping of the G80? I didn't see anything concrete on that one way or another, I'm just curious if anyone had any idea. Thanks.

 

[Munky]

Originally posted by: SlowSpyder

Originally posted by: Jax Omen

Originally posted by: chizow
If it doesn't double 3870s performance than it'll be doomed from the outset, as GT200 does promise a true doubling of G92's main specs and overall performance.

Now you're just pulling stuff out of your ass. Provide one link to support this. Oh wait, you can't, because plausible GT200 specs haven't been leaked yet. It's all conjecture right now.

What we have is plausible 4870 specs. They look to significantly improve on the 3870's performance. Will it be enough to compete with GT200? Only Nvidia knows that right now.

Slightly off topic, but along the lines of what was said above, do we know if the GT200 is supposed to be a new architecture, or another sping of the G80? I didn't see anything concrete on that one way or another, I'm just curious if anyone had any idea. Thanks.

We don't know, but seeing how it's way past the point when we should have gotten a new high end gpu from Nvidia, it's highly likely that gt200 is what we've been waiting for. Although, I remember people speculating that the g92 was the next big gpu, and we all know how that turned out.

 

[chizow]

Originally posted by: Jax Omen

Originally posted by: chizow
If it doesn't double 3870s performance than it'll be doomed from the outset, as GT200 does promise a true doubling of G92's main specs and overall performance.

Now you're just pulling stuff out of your ass. Provide one link to support this. Oh wait, you can't, because plausible GT200 specs haven't been leaked yet. It's all conjecture right now.

What we have is plausible 4870 specs. They look to significantly improve on the 3870's performance. Will it be enough to compete with GT200? Only Nvidia knows that right now.

VR-Zone's latest Rumors.....

VR-Zone has a slightly different story to tell, though. According to the Singapore-based website, the GT200 will show up in graphics cards branded GeForce 9900 GTS and GeForce 9900 GTX. The GT200 GPU will reportedly have a billion transistors, and it will be hooked up to GDDR3 memory via a 448-bit memory interface on the 9900 GTS and a 512-bit interface on the 9900 GTX. Unlike Expreview's news post from last month, this new report hints that the GT200 won't only launch in top-of-the-line cards, since Nvidia offerings with the "GTS" suffix typically debut in the $300-400 price range.

So, what does AMD have in store to counter Nvidia's offensive? According to a rumor from a couple of days ago, AMD's next-gen RV770 graphics processor will hit Radeon HD 4850 and 4870 graphics cards in May. Both offerings will have 480 stream processors, and the latter will be based on GDDR5 memory.

The bolded portion is the best hint that GT200 is going to have it where it counts. We know with G80 and G92 that ROPs and memory controllers are tied together with 4 ROPs tied to every 64-bit controller. If GT200 holds true to this design, you're looking at 32 and 28 ROPs for the 9900GTX and 9900GTS respectively. Other reports indicate approximate doubling of everything else, like 1GB RAM, ~200 shaders, 100 TMUs holding to the same ratios as seen in G92.

While specs are certainly not as firm as RV770, we've heard its going to be launched around July and will outperform a 9800GX2. Given what we know, I think its a pretty safe bet GT200 will perform closer to double a G92 than RV770 will compared to RV670.

 

[Jax Omen]

That sounds obscenely expensive. How are they going to compete with AMD on the value/dollar market like that?



Don't get me wrong, I'd love to have something with double the performance of my card. I just... don't see why Nvidia would release such an expensive-to-make part like that again. Wasn't the entire point of G92 to reduce costs because G80 was too expensive to produce, with too low of yields?

 

[v8envy]

Originally posted by: Jax Omen
That sounds obscenely expensive. How are they going to compete with AMD on the value/dollar market like that?

They already are, with the 9600GT. With the GT200 cards performing at > 2x the level of a 9600GT we're back to a high end card weighing in at $600, with midrange $150 cards offering about half the performance. As it was, as it should be.

9900 GTS should still pack an impressive punch if it's targeted at the $300-400 price point.

Also, the GT100 is meant to be the value part. All signs point to that being another respin of the G80.

 

[chizow]

Originally posted by: Jax Omen
That sounds obscenely expensive. How are they going to compete with AMD on the value/dollar market like that?



Don't get me wrong, I'd love to have something with double the performance of my card. I just... don't see why Nvidia would release such an expensive-to-make part like that again. Wasn't the entire point of G92 to reduce costs because G80 was too expensive to produce, with too low of yields?

It certainly will be, but NV has shown success at moving big expensive chips in the past with G80. As long as the product performs and sells it'll make producing them worthwhile. I expect it to sell for no less than $600 similar to the G80 GTX or G92 GX2 which it replaces at the high-end.

I think the point of G92 was to produce a cheaper chip with similar performance of G80, which it clearly did. If you look at how much the GX2 and 9800GTX retailed for at launch (may carry similar relative costs to GT200) you can see that G92 did drop pricing considerably at the high-end. I don't recall reports of G80 being unprofitable, in fact I'd say it had a lot to do with NV's record profits over the last 18 months. Don't forget that NV managed to sell self-neutered G80s at a $300 price point over a year ago with the 320MB GTS.

 

[Jax Omen]

$300? I paid $230 for mine a year ago



I'm not sure I like the $600 high-end trend continuing. It was nice having the high-end being affordable to everyone >_>

 

[srp49ers]

The main reason nvidia was able to get away with selling $600 cards last year was because ati had nothing close in performance. I think its safe to say that is not going to happen again.

 

[Killrose]

we should be seeing some leaked reveiws any day now I would think

 

[woolfe9999]

I don't think the scattered and fragmentary GT200 information is near as reliable as the specs we have right now for rv770, but who knows.

One thing I'll bet on though, if we start seeing reviews of rv770 in say 2 weeks, we'll also suddenly see a proliferation of very impressive "leaked" GT200 specs all over the web. You know, the kind that will make you not want to buy rv770 and wait for GT200 instead. :laugh:

- woolfe

 

[Munky]

Originally posted by: Jax Omen
That sounds obscenely expensive. How are they going to compete with AMD on the value/dollar market like that?



Don't get me wrong, I'd love to have something with double the performance of my card. I just... don't see why Nvidia would release such an expensive-to-make part like that again. Wasn't the entire point of G92 to reduce costs because G80 was too expensive to produce, with too low of yields?

There are rumors that we will get yet another g92, based on 55nm process, and that gpu will fill in for midrange cards in the foreseeable future while the gt200 will be reserved for high end cards.

 

[bryanW1995]

Originally posted by: Ketherx
Well, I've looked for a crossfire board for intel and it's about $100 more than one for AMD (that I've found) so it'd be cheaper for me to go with AMD. I've always liked AMD anyways so it's no problem for me.

ix38 quad gt at newegg is $137AR shipped. you found an amd crossfire mobo for $37???

 

[bryanW1995]

Originally posted by: Extelleron

Originally posted by: bryanW1995
@extelleron: more and more rumors point to 850 core and 1050 shader, so...

(480*2*.85) = 816

816/496 = 1.65X . still a good improvement, but not nearly as impressive.

If shader clock is 1050MHz, then the shaders will be clocked at 1050MHz.

Not sure what you are getting at. It's (480*2*1.05), which is exactly what I said.

Originally posted by: thilan29

Originally posted by: biostud
The RV770 is supposed to have +800M transistors the g92 has ~686M but the G92b should be larger than the RV770, why is that?

Is G92b on the 55nm process or the 65nm process? That would explain the difference. Or maybe they've included some added functionality?? (ie. like the NVIO chip that was separate on the G80 but part of the die I think on G92.)

G92b will be 55nm and have (as far as we know) the same 754M transistors that G92 currently has.

sorry, I was thinking that you were comparing core clocks because I'm blind.

 

[Martimus]

Originally posted by: bryanW1995

Originally posted by: Ketherx
Well, I've looked for a crossfire board for intel and it's about $100 more than one for AMD (that I've found) so it'd be cheaper for me to go with AMD. I've always liked AMD anyways so it's no problem for me.

ix38 quad gt at newegg is $137AR shipped. you found an amd crossfire mobo for $37???

That is with a $90MIR. A great deal to be sure, but that $90 won't get back to you for a while (it took 6 months for me to get my corsair $30 rebate).

 

[Extelleron]

Actually I did run the numbers and while I don't think 2x performance is realistic given the released specs of RV770, I do think a 50% increase over RV670 is achievable, which puts me at my 15-25% increase estimate over existing G80/G92 parts. Considering a 3870 in CF or in X2 often fails to beat the 8800/9800 GTX/Ultra in games that don't scale particularly well, I'm not sure why you're so confident RV770 will approach doubling RV670's performance. Personally I think the improvements aside from the TMU additions are unnecessary and that ignoring ROPs is a mistake.

I find it hard to see your logic in thinking the RV770 will be ~50% faster when nearly every part of the GPU has seen a 100% improvement over RV670.

ROP is not that important; this is 2008, games are shader and texture dependant and performance does not scale with pixel performance. Look at HD 3870 vs. 8800GT.... 3870 has ~30% more pixel performance but the 8800GT wins every time.

TMU is a huge bottleneck in the R600 design; people saw that back in May when HD 2900XT launched and it remains true today. ATI has clearly taken a more aggressive route when it cames to shader performance than nVidia, but they completely ignored texture performance and that hurt them a lot. ATI's top end card has had 16 TMU since 2004. Despite increases in core clock & increased performance-per-clock, that isn't nearly enough to keep up with the rapid advancement in shader performance or memory bandwidth.

The only situations where 3870 CF / 3870 X2 don't scale well are games where Crossfire is the problem; this has nothing to do with the additional resources having no effect. A single GPU with 640SP/32TMU/32ROP will outperform two RV670 chips put together by a mile.

 

[bryanW1995]

so what??? it's a FREAKIN' X38 MOBO FOR $137!!! if you can't spare $90 for 6 mos then you don't need to build a computer in the first place. also, you'll make up for it by getting a q6600 instead of an overpriced phenom, AND you'll save a couple bucks a month on power to boot.

 

[Martimus]

Originally posted by: bryanW1995
so what??? it's a FREAKIN' X38 MOBO FOR $137!!! if you can't spare $90 for 6 mos then you don't need to build a computer in the first place. also, you'll make up for it by getting a q6600 instead of an overpriced phenom, AND you'll save a couple bucks a month on power to boot.

I was just saying that most people don't count a MIR in the cost, because it is money that you need to spend now, and won't get back for quite a while. That MB doesn't cost you $137, it costs you $227 with a promise that you will get $90 back if you jump through some hoops and are willing to wait for it.

If I were buying computer instead of saving up for a new house I would jump on that ABIT MB, but I have more important things to spend my money on (or not spend in this case).

 

[chizow]

Originally posted by: Extelleron
I find it hard to see your logic in thinking the RV770 will be ~50% faster when nearly every part of the GPU has seen a 100% improvement over RV670.

ROP is not that important; this is 2008, games are shader and texture dependant and performance does not scale with pixel performance. Look at HD 3870 vs. 8800GT.... 3870 has ~30% more pixel performance but the 8800GT wins every time.

TMU is a huge bottleneck in the R600 design; people saw that back in May when HD 2900XT launched and it remains true today. ATI has clearly taken a more aggressive route when it cames to shader performance than nVidia, but they completely ignored texture performance and that hurt them a lot. ATI's top end card has had 16 TMU since 2004. Despite increases in core clock & increased performance-per-clock, that isn't nearly enough to keep up with the rapid advancement in shader performance or memory bandwidth.

The only situations where 3870 CF / 3870 X2 don't scale well are games where Crossfire is the problem; this has nothing to do with the additional resources having no effect. A single GPU with 640SP/32TMU/32ROP will outperform two RV670 chips put together by a mile.

Well, here's looking it at a different perspective then. If TMU were identified as a crippling bottleneck with RV670, simply doubling them when everything else is doubled does nothing for your overall performance; you'll still be limited by the lowest performing part, which again will be TMUs. But since you've also neglected ROPs, you'll have to take that into consideration as a possible bottleneck as well.

If AMD really wanted to double RV670's performance, they wouldn't have just band-aided TMUs by doubling them, they would've increased them by a factor of 3 or 4 to increase the ratio of TMUs to everything else. Same with ROPs, although I don't know how much impact they have on this generation of ATI parts. I think its clearly obvious they're still the most important factor when it comes to performance with NV parts.

I'm really not sure what you're trying to get at with the rest of your statements about shader performance and memory bandwidth being so monumentally important in current games when that simply doesn't play out in games, at least with NV parts or with 2900XT. Its clear that having too little of either can have a serious negative impact on performance, but also that there's simply no advantage gained from excess shader performance and bandwidth.

 

[bryanW1995]

Originally posted by: Martimus

Originally posted by: bryanW1995
so what??? it's a FREAKIN' X38 MOBO FOR $137!!! if you can't spare $90 for 6 mos then you don't need to build a computer in the first place. also, you'll make up for it by getting a q6600 instead of an overpriced phenom, AND you'll save a couple bucks a month on power to boot.

I was just saying that most people don't count a MIR in the cost, because it is money that you need to spend now, and won't get back for quite a while. That MB doesn't cost you $137, it costs you $227 with a promise that you will get $90 back if you jump through some hoops and are willing to wait for it.

If I were buying computer instead of saving up for a new house I would jump on that ABIT MB, but I have more important things to spend my money on (or not spend in this case).

yeah, I know that you're right. esp a $90 mir like that is kind of a pain...but that does make that mobo cheaper than even the k9a2 platinum, which seems to be the mobo of choice for 790fx buyers who don't have an unlimited budget.

 

[bryanW1995]

If TMU were identified as a crippling bottleneck with RV670, simply doubling them when everything else is doubled does nothing for your overall performance

huh?? if tmu are your crippling bottleneck, and you double them along with everything else, wouldn't you double your overall performance?

 

[taltamir]

Originally posted by: v8envy

Originally posted by: Jax Omen
That sounds obscenely expensive. How are they going to compete with AMD on the value/dollar market like that?

They already are, with the 9600GT. With the GT200 cards performing at > 2x the level of a 9600GT we're back to a high end card weighing in at $600, with midrange $150 cards offering about half the performance. As it was, as it should be.

9900 GTS should still pack an impressive punch if it's targeted at the $300-400 price point.

Also, the GT100 is meant to be the value part. All signs point to that being another respin of the G80.

maybe a 55nm G80 derivative? well G92 but same difference.

 

[ArchAngel777]

Originally posted by: bryanW1995

If TMU were identified as a crippling bottleneck with RV670, simply doubling them when everything else is doubled does nothing for your overall performance

huh?? if tmu are your crippling bottleneck, and you double them along with everything else, wouldn't you double your overall performance?

I don't think that is what he meant. I think what he meant was that if the TMU was the problem with R600, then doubling everything would still leave it TMU bound, regardless of the improved performance and would therefore leave doubts that the 'only' problem with the R6XX series was a lack of TMU performance.

 

[bryanW1995]

no, he said that doubling EVERYTHING does nothing for your overall performance. He didn't say that it did nothing to help you find out if the tmu's were the bottleneck.

 

[ArchAngel777]

Originally posted by: bryanW1995
no, he said that doubling EVERYTHING does nothing for your overall performance. He didn't say that it did nothing to help you find out if the tmu's were the bottleneck.

That is what he said, but that isn't what he meant. This much is obvious. Were human, we make mistakes and do not always type out properly what we mean to convey.