AMD: R7xx Speculation

[trinibwoy]

SB I agree with most of that. They are all great on paper like so many other things ATI has done in the past. However, until reviewers start talking about it and more people actually start running obscure CF configurations I wouldn't consider the advantages of CFX to be doing AMD much good. They have a long way to go before CF has anywhere near the market penetration that SLI has at the moment. Where AMD can make a big splash is if they address the core failings of multi-GPU like redundant storage and poor scaling. It looks like they are moving in that direction so hopefully we see something from them soon.

I don't accept the equal scaling though. It looks like CF still falls flat in considerably more cases than SLI.

Morgoth, on the other hand you can also get 2x8800GTS for much less which will beat all of AMD's offerings in a lot of cases.

 

[aaronspink]

The big advantage IMO is with respect to multi-display support but again you're talking about a very small (yet arguably vocal) number of consumers who have both a multi-GPU and multi-display setup. Another issue that Nvidia should be very aware of is the increasing attractiveness of AMD and Intel based motherboards....that could be the most dangerous threat to their multi-GPU throne.

It always surprises me that people can function with a single display. Once you star using a multi-display system its incredibly hard to go back and certainly with LCDs multi-display is very nice. Seriously, why would you spend $1k+ on an SLI/CF setup and not $200 for a second LCD?

Certainly for me, the general lack of support for multiple monitors in SLI/CF is just ridiculous. Followed closely by the requirement for SLI of only nvidia motherboards. I want a stable motherboard from a vendor that has a history of things like NOT having data corruption and crappy drivers. I'm not sure the Nvidia strategy for SLI wrt chipsets has really helped them as much as they think. Certainly not financially since it has very much limited to a small niche subset of the market.

Aaron Spink
speaking for myself inc.

 

[Dave Baumann]

Certainly for me, the general lack of support for multiple monitors in SLI/CF is just ridiculous.

Extended desktop is a feature of CrossFire X. Current WHQL supports it with CrossFire enabled.

 

[Rys]

And it's the best thing about multi-GPU to happen since multi-GPU was introduced. Every day that ticks past without SLI supporting it is yet another reason (and you don't need many) to just ignore SLI completely.

 

[AlexV]

Morgoth, on the other hand you can also get 2x8800GTS for much less which will beat all of AMD's offerings in a lot of cases.

As with everything, it's a tradeoff. A comeback would be that 3X 3870s are dirt cheap currently and should diss it out/beat the GTSs in a number of scenarios. It's not very clearcut up to a certain point along the product lines, once you go over that it's all nV. And then one has to factor the need for an SLi supporting chipset (which for Intel users might be problematic). My humble opinion is that ATi are more competitive now then they've been in quite a long time, albeit still not equaling their time of glory. As for splash-making, there are some neat things still in the pipeline WRT CFX...as I'm sure there are some for SLi.

 

[nicolasb]

Actually they never made an ultra high end card with the R600- it targeted the GTS, not the GTX. The R5XX was their last attempt at an ultra high end card

No, R600 targeted the high end. It just missed its target. By a tragically huge distance.

Once ATI realised that the best R600 could do was be beaten by the 8800GTS most of the time, they lowered the R600 selling price to GTS levels. But that was certainly not the original intention - not with a chip that size and a 512-bit bus.

 

[Novum]

SLI and Crossfire is all about the Software, and NVIDIA is way ahead of ATI at this point. Also I can't see why Crossfire should be better on a hardware-level than SLI.

 

[LordEC911]

No, R600 targeted the high end. It just missed its target. By a tragically huge distance.

Once ATI realised that the best R600 could do was be beaten by the 8800GTS most of the time, they lowered the R600 selling price to GTS levels. But that was certainly not the original intention - not with a chip that size and a 512-bit bus.

They were also aiming at a higher clockspeed... which is why I think it "missed" so bad.

 

[compres]

CFX is IMO ahead of SLI right now: Better Intel chipsets, multi monitor, better scaling and flexibility. IMO only single chip nVidia is better than ATI now. I would not invest in a intel/nvidia system and most people I know agree with me on this one. So some posts here leave me scratching my head as to why nVidia is ahead on multi-gpu. Care to elaborate someone?

 

[INKster]

CFX is IMO ahead of SLI right now: Better Intel chipsets, multi monitor, better scaling and flexibility. IMO only single chip nVidia is better than ATI now. I would not invest in a intel/nvidia system and most people I know agree with me on this one. So some posts here leave me scratching my head as to why nVidia is ahead on multi-gpu. Care to elaborate someone?

Humm, because Nvidia sell's more SLI-ready GPU's than AMD does Crossfire-ready ones ?
Just a thought.

Having supported chipsets on the market is only part of the equation.

 

[compres]

Humm, because Nvidia sell's more SLI-ready GPU's than AMD does Crossfire-ready ones ?
Just a thought.

Having supported chipsets on the market is only part of the equation.

What good is SLI ready GPU with no SLI ready chipset? Even the 8600s are sli ready, wonder who slis them lol.

And I think they sell close to the same, nVidia makes better margins.

 

[INKster]

What good is SLI ready GPU with no SLI ready chipset? Even the 8600s are sli ready, wonder who slis them lol.

And I think they sell close to the same, nVidia makes better margins.

According to the last JPR report, Nvidia was close to selling twice the amount of GPU's that AMD did over the same quarter.
SLI and Crossfire are really just blips in the radar, when it comes down to it.

Why ?
Because it only makes sense to do it in the high-end, and high-end GPU users don't mind spending more on motherboards anyway.

No one really buys motherboards just to do SLI or Crossfire of 8600 GT's or HD3650's, it's ridiculous.
They can achieve the same level of performance (without the hassle of multi-GPU driver support) with a single 8800 GT or HD3870 for less than the combined price of two of the others, and for that, any mobo with a PCIe x16 slot will do.

If that, SLI and Crossfire serve only as halo-effect marketing ploys to lure low-end and mainstream buyers into their GPU families, but that's it.
The average joe doesn't even understand the concept of buying two graphics cards to increase performance in the first place.
The very aggressive market dynamics and tight product cycles "take care" of any multi-GPU product in a very short period.

 

[AlexV]

They were also aiming at a higher clockspeed... which is why I think it "missed" so bad.

It should've been significantly higher then it ended up being, given the architecture itself. How high do you actually think they wanted to reach?I think that Eric Demers gave a ballpark figure in one of his interviews, and it was nothing outlandish 800ish something for the 2900XT. Which still would've put it quite far from the GTX/Ultra parts, and not necessarily a GTS killer.

 

[ShaidarHaran]

It should've been significantly higher then it ended up being, given the architecture itself. How high do you actually think they wanted to reach?I think that Eric Demers gave a ballpark figure in one of his interviews, and it was nothing outlandish 800ish something for the 2900XT. Which still would've put it quite far from the GTX/Ultra parts, and not necessarily a GTS killer.

I for one don't believe Mr. Demers would admit to the monumental failure that R600 amounted to, given the true 900-1000MHz target clocks. "800ish" sounds like not such a far cry from the 750MHz R600 ended up at, and really just reeks of marketing spin. Afterall, if the target clocks were really only in the "800ish" range, then we're there already with RV670/R680 - and those parts are far behind the competition (strictly speaking from a performance perspective, generally with filters enabled). I just don't believe anything less than 900MHz was ever the goal for the high-end R6xx hardware.

 

[AlexV]

I for one don't believe Mr. Demers would admit to the monumental failure that R600 amounted to, given the true 900-1000MHz target clocks. "800ish" sounds like not such a far cry from the 750MHz R600 ended up at, and really just reeks of marketing spin. Afterall, if the target clocks were really only in the "800ish" range, then we're there already with RV670/R680 - and those parts are far behind the competition (strictly speaking from a performance perspective, generally with filters enabled). I just don't believe anything less than 900MHz was ever the goal for the high-end R6xx hardware.

Why?Because ATi must always have a superior part?Does being fair about what your design goals were=admitting a monumental failure?Why is it so beyond the realm of imagination that during the design phase, the R600 is how they envisioned a high-end part in the timeframe they were aiming for, should look?

The R600 was poor because the G80 was great...and ATi sure as hell had no involvement with the second part.

 

[Moloch]

Also they may have thought it was was gonna perform better than it did per clock not realizing their VLIW design would have been as hard to extract performance from?

 

[AlexV]

Also they may have thought it was was gonna perform better than it did per clock not realizing their VLIW design would have been as hard to extract performance from?

Possible, but not necessarily likely. See the delta between it and the R5xx parts before that, and then apply a similar increase to the G70...they might've simply assumed that's how things would pan out with nV, and were caught by surprise when that assumption proved quite wrong.

I don't think the R6xx are primarily limited by their shading-performance...they seem to be doing fine when entirely shader bound in synthetic tests. Too bad there aren't any games behaving like that.

 

[XMAN26]

CFX is IMO ahead of SLI right now: Better Intel chipsets, multi monitor, better scaling and flexibility. IMO only single chip nVidia is better than ATI now. I would not invest in a intel/nvidia system and most people I know agree with me on this one. So some posts here leave me scratching my head as to why nVidia is ahead on multi-gpu. Care to elaborate someone?

SLi works with more games. You can create SLI profiles. IMO, CFX lacks to SLI in scaling, how many games do you have to disable CFX so you dont have to take a performance hit compared to SLI. Now I'm not claiming to know for a fact, but I've yet to read a review where SLI gave a negitive performance gain months after release. CFX still has this issue with many games.

 

[ChrisRay]

I know its semi offtopic. But the SLI verses Crossfire debate here got me thinking. And I used several comments in this thread and elsewhere. Delivering some feedback to Nvidia regarding SLI in general.
As I may not agree with everything said here. I do think there is some useful feedback Nvidia does need to be hearing.

 

[AlphaWolf]

Why ?
Because it only makes sense to do it in the high-end, and high-end GPU users don't mind spending more on motherboards anyway.

No one really buys motherboards just to do SLI or Crossfire of 8600 GT's or HD3650's, it's ridiculous.
They can achieve the same level of performance (without the hassle of multi-GPU driver support) with a single 8800 GT or HD3870 for less than the combined price of two of the others, and for that, any mobo with a PCIe x16 slot will do.

I think the combination of hybrid crossfire and the 780G chipset may change that. It could become a very viable option for low end gamers to combine a ~$79 card with onboard video.