Crossfire beats SLi in Doom3!!!

you think they could fit another GPU onto something this big already:

1117609653IPc8G4ZOfL_1_1_l.jpg
 
Sly said:
So ATI would have had a better shot if they had gone the Dual Core route instead. Why did they decide to do dual slot? :confused: So far we've only seen pics of 6800 dualcore cards (Not sure if they're released yet), but ATI would have had been able to do this from the beginning. But instead, simply rigged together a dualslot solution. Why would they do that? :confused:

Honestly that's what I was waiting for them to do. However, from a marketting standpoint I can kind of see why they did it this way.

In order to make a dual GPU card work as effectively as SLi, ATi would have had to add another 256MB frame buffer to the card for the second GPU. Thus making the card a power hungry monster, that generates alot of heat and costs a fortune. Just like the SLi on one card video cards we've been seeing with 6600GT's and 6800Ultras. It would have had to be done the same way.

However, there is more money to be made and somewhat more flexibility in a dual card design.
 
Sir-Fragalot said:
Honestly that's what I was waiting for them to do. However, from a marketting standpoint I can kind of see why they did it this way.

In order to make a dual GPU card work as effectively as SLi, ATi would have had to add another 256MB frame buffer to the card for the second GPU. Thus making the card a power hungry monster, that generates alot of heat and costs a fortune. Just like the SLi on one card video cards we've been seeing with 6600GT's and 6800Ultras. It would have had to be done the same way.

However, there is more money to be made and somewhat more flexibility in a dual card design.


They didnt have the R&D ready for that. The next chip R520 is a monster with support of up to 32pipes and the most transistors seen on any graphics core ever. Quite simply it is not something they would want to run MAXX form. Dual core like the X2 is something else and would require again, more work then CrossFire to make a card that wasnt embarassing such as what manufacturers are doing with dual core Ultras and GTs.

Maybe in a year or two, but right now the cores are pushing too much heat and size to make a decent dual core card.
 
Why is everyone saying it is really early for ATI..nobody knows how long they have been working on it
 
FanATIc said:
Dual core like the X2 is something else and would require again, more work then CrossFire to make a card that wasnt embarassing such as what manufacturers are doing with dual core Ultras and GTs.

Maybe in a year or two, but right now the cores are pushing too much heat and size to make a decent dual core card.

I don't know. Given the cash, I'd run dual dual core 6800GTs or Ultras. The performance would be impressive, I'm sure.
 
Graphics cards are already "multi core". Each of those pipelines acts like a different core. This was touched on with another recent thread....
 
FingerSlut said:
anyone notice the dual core nvidia cards still have the sli bridge connecters?

That's because that's how the GPU's communicate still, through SLI.....
 
Skrying said:
That's because that's how the GPU's communicate still, through SLI.....

That makes no sense. Why would you include the connector (and necessary traces) on the PCB if both chips are on the same board?
 
trinibwoy said:
That makes no sense. Why would you include the connector (and necessary traces) on the PCB if both chips are on the same board?

Because the cards dont know how to communicat any other way except through SLI. SLI is an instruction set more than anything else. There are instructions on the GPU's on how to commnicate with each other, the brdige is included on the card because it allows them to commnicate.
 
trinibwoy said:
That makes no sense. Why would you include the connector (and necessary traces) on the PCB if both chips are on the same board?

Because, once sufficient PCI-e lanes exist, get ready for Quad SLI :D You'll have two GPUs on each card SLI'd together for 4 GPU's to render your graphics. Just think of the performance of AFR with four chips rendering each subsequent frame (GPU 1 - frame 1, GPU 2 - frame 2, GPU 3 - frame 3, GPU 4 - frame 4 and so on). This will yield greater scalability (AFR that is). Not sure how SFR or SuperTiling would scale to 4 GPU's though.

AND "your" existing dual GPU card will already support it. Just like all the PCI-e 6800GT and 6600GT products where SLI capable before the SLI motherboards were even available.

I think this is the secret SLI version 2 that the Inquirer made allusions to. All that remains is providing to x16 PCI-e Graphics slots in SLI mode instead of x8.
 
HighTest said:
Because, once sufficient PCI-e lanes exist, get ready for Quad SLI :D You'll have two GPUs on each card SLI'd together for 4 GPU's to render your graphics. Just think of the performance of AFR with four chips rendering each subsequent frame (GPU 1 - frame 1, GPU 2 - frame 2, GPU 3 - frame 3, GPU 4 - frame 4 and so on).

This will yield greater scalability (AFR that is). Not sure how SFR or SuperTiling would scale to 4 GPU's though.

That would also demand more driver work and therefore CPU work. The overhead would increase more, so with each GPU you'd get less of a benefit. But of course there would be a benefit.
 
Skrying said:
That would also demand more driver work and therefore CPU work. The overhead would increase more, so with each GPU you'd get less of a benefit. But of course there would be a benefit.

Well, to solve that riddle, someone needs to bench the CPU load of the existing SLI solution in comparison to non SLI. It's currently possible to bench, then just add to the score of CPU use increase to speculate on the impact of 4 GPU's. As I understand it, the overhead of AFR (preferable to SFR) is quite low, so I don't expect it to be as high as you think it might.

In any event, it's a minor point, as if you are going to go SLI 2, you'll most likely be dual-core X2. While games may not be multi-threaded, XP can be for the driver support and the other core will easily handle any overhead increases here.
 
Skrying said:
Because the cards dont know how to communicat any other way except through SLI. SLI is an instruction set more than anything else. There are instructions on the GPU's on how to commnicate with each other, the brdige is included on the card because it allows them to commnicate.

Are we talking about the same thing? Are you talking about this -

 
HighTest said:
Because, once sufficient PCI-e lanes exist, get ready for Quad SLI :D

No I meant what Skyring said makes no sense. He claimed that the connector is used by the two chips on board to communicate.
 
Skrying said:
Because the cards dont know how to communicat any other way except through SLI. SLI is an instruction set more than anything else. There are instructions on the GPU's on how to commnicate with each other, the brdige is included on the card because it allows them to commnicate.

hahahahahahahaha

Ok smart guy, then how do the dual chip cards without an SLi connector work? (In case you haven't noticed, most of them don't have the connector and work just fine, so now that they are appearing WITH the connector, it's a pretty safe assumption that quad SLi is on the way)...

:eek: :rolleyes:
 
Skrying said:
That would also demand more driver work and therefore CPU work. The overhead would increase more, so with each GPU you'd get less of a benefit. But of course there would be a benefit.

The CPU hit will be greater, but probably not so much that it became limiting...thus far it only appears to be about a %5 CPU hit maximum...and I'm guessing that double the cards won't quite be double the performance hit...

I don't think that many current games would really benefit from such a setup, but a game like Unreal3, which is multi-threaded and highly graphic intensive, might benefit from it dramatically, and the performance hit wouldn't be as noticeable since dual core processors could better handle the load...
 
^eMpTy^ said:
Ok smart guy, then how do the dual chip cards without an SLi connector work? (In case you haven't noticed, most of them don't have the connector and work just fine,

It's not even necessary to point that out (although pics are plentiful on the web for verification). The bridge connector is just a bunch of traces for signals to travel. All logic is embedded in the GPU. So of course if the chips are on the same PCB the traces will just run directly between each other and there would be no need for a connector for a simple dual-GPU card. 4x SLI is a whole other story.
 
The SLI bridge is already built onto the board. The connector on top could used for quad SLI. I dont know how you took my remarks, but you seemed not to understand them the way I meant them. The connectors on the dual cards can be used for connecting to another set most likely, or maybe they do need that to actually follow the traces, who knows. I'm not sure how we got to this point, but I'm fairly confused because I believe we were both taking each others words wrongly.

Anyway, Empty, GROW THE FUCK UP. Even if that gets me banned, you seriously need to get a life. Your whole fucking life seems to revolve making Nvidia look good, get over yourself child, please.

I think I'll just keep myself to reading for awhile. Good job Empty, you have ruined someones experience at a great site.
 
trinibwoy said:
It's not even necessary to point that out (although pics are plentiful on the web for verification). The bridge connector is just a bunch of traces for signals to travel. All logic is embedded in the GPU. So of course if the chips are on the same PCB the traces will just run directly between each other and there would be no need for a connector for a simple dual-GPU card. 4x SLI is a whole other story.

Yes, exactly. The cards that have dual GPU's (not dual core, people need to quit calling them dual core, they are no such thing. Pipeline mechanics not withstanding.) are showing up with the SLi connector for the purpose of quad SLi. This technology should be scalable and therefore depending on how the motherboard manufacturers and video card manufacturers impliment the technology, we very well could see quad SLi systems. Two cards with two GPU's each.

Be expecting to spend alot of money.

FanATIc said:
They didnt have the R&D ready for that. The next chip R520 is a monster with support of up to 32pipes and the most transistors seen on any graphics core ever. Quite simply it is not something they would want to run MAXX form. Dual core like the X2 is something else and would require again, more work then CrossFire to make a card that wasnt embarassing such as what manufacturers are doing with dual core Ultras and GTs.

Maybe in a year or two, but right now the cores are pushing too much heat and size to make a decent dual core card.

According to ATi themselves AFR has been built into every GPU they've made since back in the MAXX days. It doesn't take that much R&D to put a dual GPU solution on a single card.

There are no dual core Ultra's or GT's. I don't get what your saying by that last statement.

"Embarassing such as what manufacturers are doing with dual core Ultras and GT;s?"

Theres nothing embarrassing about it. Cards like that are built for the enthusiest. Sometimes they are built for marketting a certain type of product image. Sometimes simply being able to produce a mammoth video card that would cost too much is worth it for alot of reasons. That doesn't mean they'd actually sell the damn thing. By building in that feature, they have chip scalability. Which is probably the sole reason they can do Crossfire at all. The 6800's were designed to be scalable.

When ATi says that their GPU's are scalable in the same fashion they aren't kidding. I believe them. However that doesn't mean they had plans to use it. From a hardware perspective, the Crossfire Edition master card is a half ass solution. Its not nearly as elegant as the SLi system. Where you get the board, and two identical video cards and theres no dongle outside the machine and the thing works. This master edition card idea just about proves they weren't ready to take on SLi.

The MAXX style card was a cheap way to boost performance. ATi knew it then and they know it now. I think the reason they don't do it is primarily due to price point, and the physical size of the card would be monsterous. Not something they could sell to the masses. A dual video card solution has numerous advantages from a product stand point. Also, due to their video card popularity, this will really allow ATi to break into the motherboard market like Nvidia has.

They've had mobo chipsets for awhile but have never managed to really capture the enthusiest crowd with them. I still don't think that they were ready for Nvidia's SLi when it was announced. The time its taking to release an answer to that product shows this also.

I have to hand it to ATi on the driver side of things if it works like they say it will. If I had both products to chose from, and Crossfire were as efficient as SLi, I would probably have bought that instead of SLi as its less of a headache to deal with the ATi drivers the way they've implimented them for Crossfire than Nvidia's driver profile mess.

The heat is definatley on for Nvidia to do something. I would expect really kickass things in their drivers in the next month or two to try and level the playing feild with ATi. If nothing else, ATi's driver implimentation has raised the bar for Nvidia and possibly even from a hardware perspective. They'll have to make damn sure that G70 is at least as good as ATi's R520 and at least as good driver wise.

This kind of competition is good. ATi embarrassed Nvidia's FX series with their 9000 series. Nvidia forced ATi to innovate harder when the 6800's came out. Now ATi's going to make Nvidia do an even better job with G70.
 
Skrying said:
Graphics cards are already "multi core". Each of those pipelines acts like a different core. This was touched on with another recent thread....


CPUs have quite alot of pipelines as well, start calling them all multi core and you'll confuse a lot of people.
 
Yeah, don't mix the terms up. Dual core refers to the whole coherent computation unit of a processing unit, also called the "core", and not a single pipeline or other small subset of computation units. Only when you have 2 of these interconnected in close proximity atomic-wise are they technically allowed to be called dual core processing units. :)
 
robberbaron said:
Crossfire! Don't get caught in the...CROSSFIRE!! CROSSFIRE! CROSSFIRE! CROSSFIYAAAHHH!!!!

edit: I found a game that Crossfire will beat SLI at every single time.
http://crossfire.real-time.com/


http://crossfire.real-time.com/screenshots/gallery1/ss01_JPG.html

Imagine this... SUPER ANTIALIASED!

Here's a demo of the new advanced shadow and lighting engine on X850 crossfire edition cards:
http://crossfire.real-time.com/screenshots/gallery1/ss07_JPG.html


ROFL

oh man good times, good times
 
FanATIc said:
CPUs have quite alot of pipelines as well, start calling them all multi core and you'll confuse a lot of people.

Oh man, that is so wrong it's not funny. CPU's have a single pipeline with multiple stages, just like each pipeline on a modern GPU. Closest thing to multiple pipelines on a CPU is Intel's HT.
 
Skrying said:
The SLI bridge is already built onto the board. The connector on top could used for quad SLI. I dont know how you took my remarks, but you seemed not to understand them the way I meant them. The connectors on the dual cards can be used for connecting to another set most likely, or maybe they do need that to actually follow the traces, who knows. I'm not sure how we got to this point, but I'm fairly confused because I believe we were both taking each others words wrongly.

Anyway, Empty, GROW THE FUCK UP. Even if that gets me banned, you seriously need to get a life. Your whole fucking life seems to revolve making Nvidia look good, get over yourself child, please.

I think I'll just keep myself to reading for awhile. Good job Empty, you have ruined someones experience at a great site.

My job here is done...

(was it something I said?)

And the reason for the misunderstanding was quite simple. Everyone else was referring to the SLi connector at the top of the board, god knows what you were talking about. But the whole point of the discussion was that that dual chip board has an SLi connector on it...which suggests, yet again, that quad SLi is on the way.
 
FanATIc said:
CPUs have quite alot of pipelines as well, start calling them all multi core and you'll confuse a lot of people.

I'm going to resist my urge to be wilding condescending (since you were telling me only days ago how stupid I am) and I'll just let you know that this is absolutely and completely wrong.

Every x86 processor on the face of the planet has one pipeline with the exception of Hyper-threaded pentium 4s and the new dual core processors...which have two. (Hyper threading is more like 1.5)
 
trinibwoy said:
Oh man, that is so wrong it's not funny. CPU's have a single pipeline with multiple stages, just like each pipeline on a modern GPU. Closest thing to multiple pipelines on a CPU is Intel's HT.

And true or untrue that each of those 20 stages (P4 used for example) is not referred to aa pipeline when talking about it in comparison to something else?



I canstantly see people refer to AMD Vs Intel as and i quote "less pipelines but longer". Used all the time. Makes it easier for idiots or "learning impaired" to understand. We can start a word pick game if you like, i just dont want to see someone who thinks they're "cool" by saying that in every thread. It will confuse a vast majority who read.
 
FanATIc said:
And true or untrue that each of those 20 stages (P4 used for example) is not referred to aa pipeline when talking about it in comparison to something else?

Completely untrue.

You're confusing the usage of the word pipeline. CPUs are said to be deeply "pipelined", meaning there are more stages in the pipeline, but at no time do they ever become more than one pipeline (except HT and Dual Core)...
 
^eMpTy^ said:
well atleast you know what 2T stands for...maybe you should stick to overclocking
People who run 2T don't belong on the [H] forums. ;)

On a side note, and on topic, it is nice to see NV go down on their beloved Doom3.
 
FanATIc said:
I canstantly see people refer to AMD Vs Intel as and i quote "less pipelines but longer". Used all the time. Makes it easier for idiots or "learning impaired" to understand.

In this case, apparently not easy enough.
 
R1ckCa1n said:
People who run 2T don't belong on the [H] forums. ;)

On a side note, and on topic, it is nice to see NV go down on their beloved Doom3.

hahahaha...apparently you didn't look very closely at those benchmarks...
 
Skrying said:
The SLI bridge is already built onto the board. The connector on top could used for quad SLI. I dont know how you took my remarks, but you seemed not to understand them the way I meant them. The connectors on the dual cards can be used for connecting to another set most likely, or maybe they do need that to actually follow the traces, who knows. I'm not sure how we got to this point, but I'm fairly confused because I believe we were both taking each others words wrongly.

Anyway, Empty, GROW THE FUCK UP. Even if that gets me banned, you seriously need to get a life. Your whole fucking life seems to revolve making Nvidia look good, get over yourself child, please.

I think I'll just keep myself to reading for awhile. Good job Empty, you have ruined someones experience at a great site.


And you ruined another thread, honestly it's always the same people who crap here. You're all hypocrits.

Start fresh.
 
FanATIc said:
And true or untrue that each of those 20 stages (P4 used for example) is not referred to aa pipeline when talking about it in comparison to something else?



I canstantly see people refer to AMD Vs Intel as and i quote "less pipelines but longer". Used all the time. Makes it easier for idiots or "learning impaired" to understand. We can start a word pick game if you like, i just dont want to see someone who thinks they're "cool" by saying that in every thread. It will confuse a vast majority who read.

Completely untrue. I have never seen anyone refer to a pipeline stage as a "pipeline". That would be even more confusing!!! The comparison has always been, as ^empty^ said, between a deeper/longer pipeline with more stages and lower IPC to a shorter pipeline with higher IPC in terms of P4 vs Athlon(64).

There are multiple pipelines on a GPU, each with multiple stages - http://www.beyond3d.com/articles/vs/index.php?p=12
 
trinibwoy said:
Completely untrue. I have never seen anyone refer to a pipeline stage as a "pipeline". That would be even more confusing!!! The comparison has always been, as ^empty^ said, between a deeper/longer pipeline with more stages and lower IPC to a shorter pipeline with higher IPC in terms of P4 vs Athlon(64).

There are multiple pipelines on a GPU, each with multiple stages - http://www.beyond3d.com/articles/vs/index.php?p=12

mmhmm

Is this starting to make sense FanATIc? Do you see why when people ask about "dual-core GPUs" it doesn't really make sense because they effectively already are dual core? This is also a reason people are finding ways to do other kinds of computation on GPU hardware, because of this massively parallel nature.
 
Back
Top