ATi 4870 X2 (R700) thread

[allies]

Originally posted by: deepinya
so ATI > Nvidia.....finally?

Good for them

Well the R700 isn't due out until the end of July/early August. If nvidia can pull a rabbit out of their hat before now and then they'll retain the performance crown. However, price/performance definitely is owned by AMD, and if I buy a card, I'll be supporting them.

 

[chizow]

Originally posted by: Rebel44

Originally posted by: BFG10K

When I already have more then 60 FPS I dont care if I have 75 instaed of 79....

So are all cards are equal as long as they?re over 60 FPS?

That was my reply to chizow´s BS about CPU bottleneck. When my bottleneck is my Q6600 @3.4Ghz I am pretty sure that my FPS are more then 60 FPS - and I dont care if I am CPU bottlenecked in 3D Mark or similar stupid benchmark...

That just reinforces my point about CPU bottlenecking. Chances are if you have a GT200, RV770 or G80/G92/RV670 multi-GPU config you're already bottlenecked to the point upgrading may yield little/no benefit. There's certainly nothing wrong with basing your purchasing decision on a 60FPS threshold if you game on an LCD and use Vsync. I certainly look at 60FPS minimums as ideal also, but knowing how fast a card performs without Vsync helps me determine whether I can enable features like AA or not in both current and future games.

As for BS with bottlenecking, here's another review with 4870CF that shows very little benefit with a 3GHz Q6600:

Tweaktown 4870CF Review

The main problem that we really have is at 3GHz on a quad core we do feel somewhat CPU limited, with little gains seen with the implementation of CrossFire in the real world environment. Fortunately, we will be strapping the two cards into a 4GHz test bed soon enough, which should hopefully let these cards shine.

 

[Mech0z]

Can I connect my Auzentech X-FI Prelude to a 4850/4870 and run it to my av reciever through a single HDMI and get both video and sound through that cable or does it not work with none onboard soundcards?

 

[ShadowOfMyself]

Originally posted by: jaredpace
Pretty sure that...

2x GTX 280 SLI > R700 by 20% average
R700 > GTX 280 by 15% avg in X-fire scaling games
2x R700 > 2x GTX 280, 3x GTX 280, & 2x 9800GX2 by who knows how much after driver improvements in 4-gpu scaling...

Lol, you did some horrible math there, youre assuming the GTX 280 only scales 35% in SLI? Thats horrible! If R700 was only 15% faster than GTX 280 it would get OWNED by GTX SLI, it should be about 50% faster on avg I would say, which will still put 80% scaling GTX SLI at 30% faster, sounds fine to me (nevermind the ridiculous Nvidia prices, just talking about performance)

 

[Tempered81]

Originally posted by: ShadowOfMyself

Originally posted by: jaredpace
Pretty sure that...

2x GTX 280 SLI > R700 by 20% average
R700 > GTX 280 by 15% avg in X-fire scaling games
2x R700 > 2x GTX 280, 3x GTX 280, & 2x 9800GX2 by who knows how much after driver improvements in 4-gpu scaling...

Lol, you did some horrible math there, youre assuming the GTX 280 only scales 35% in SLI? Thats horrible! If R700 was only 15% faster than GTX 280 it would get OWNED by GTX SLI, it should be about 50% faster on avg I would say, which will still put 80% scaling GTX SLI at 30% faster, sounds fine to me (nevermind the ridiculous Nvidia prices, just talking about performance)

Okay okay, so my math might be off here. I am referring to a situation like this COD4 benchmark: http://www.anandtech.com/video/showdoc.aspx?i=3334&p=19.

At 2560x1600 w/ 4xAA the GTX 280 scores 67.4 FPS and in SLI scores 109.7 FPS. I expect a R700 to run about 85 FPS in that situation. Hopefully 2x R700 in CFX would run about 130 FPS. Notice the GX2 is scaling about 64% here as well. I may just be an idiot @ math. What does that mean my percentages should be?

280 SLI scaling ~62%
R700 CFX scaling ~52% [modest]

R700 > 280 by ~26%
280 SLI > R700 by ~30%
R700 CFX > 280 SLI by ~23%
R700 CFX > GX2 SLI by ~12.5% Correct?

In my original post i meant to say:

2x GTX 280 SLI > R700 by 30% average
R700 > GTX 280 by 26% avg in X-fire scaling games

Hard benches should be out sooner than later. I NEED DATA!!!

 

[ShadowOfMyself]

I thought after all this time multi gpu scaling was much better... But judging by those benchs its not, and its not cpu limitation either because the GX2 scores higher.. quite strange

 

[lopri]

Originally posted by: chizow
Again, it doesn't matter what kind of interconnect is between the GPUs, if I had to guess that's what the CF sideport is for. You still run into the problem of the physical PCIE 2.0 slot the card is placed into. AFAIK you cannot make this larger than x16 if the physical traces and the slot itself are limited to 16 lanes. Even if you were passing everything through a single GPU and not splitting the PCIE bus you'd have ~2x data (approx. 4870CF) passing over an x16 bus rather than 1x data over an x8 bus. Normally you would think this sufficient but based on the Tweaktown review of 4850CF on P45 and X48 that's not the case.

I think we're talking different things. For CF/SLI, the communication between GPUs occurs through PCIe slots (which go to NB) and the CF/SLI bridge, but that is not the case for 9800 GX2 or HD 4870 X2. While I am talking about the inter-GPUs, it seems like you're talking about the PCIe slots (on motherboard) bandwidth capability, which are very different things.

If we talk about the PCIe slots on motherboard, then the truth is that we don't know how much the current x16 lane (2.0) is a bottleneck, until the PCIe 3.0 arrives. Can you say for sure that GTX 280 isn't bottlenecked by x16 lanes? What if there were x32 lanes? Will any of the current top cards perform better?

Now, we know that x8 lane brings performance loss. Not as much as the TweakTown shows. It's usually around 10% (~15% in 3DMarks) in my testing. But a loss nonethless. So if you break a x16 lane into two x8 lanes then put each GPU in there, then surely there will be a performance loss. In this case, the communication happens between two x8 lanes and it also has to go through the north bridge. (But I'm guessing the SLL/CF bridge compensate somewhat)

Two-GPU cards like GX2 or X2 do not communicate on motherboard. The GPUs communication between the GPU occurs independently from motherboard's PCIe lanes. So there is no spliting of a x16 lane there.

However, if your concern is that two combined RV770's performance will be too much for current PCIe 2.0, then that logic can apply to anything that's faster than a single RV770. We could also say that "GTX 280 is bottlenecked by a x16 lane"

This might or might not be true, but it cannot be proved at this time, thus kind of pointless.

 

[Munky]

Originally posted by: Sylvanas
Lopri:

Gen 2 performancePAK?

Dual Cast?s
Read Pacing?
Dynamic Buffer Allocation


Hmm..

Good find! Could this mean a shared memory pool? It seems like we won't be stuck with the 512mb framebuffer per GPU limitation if the PLX chip 'dynamically' assigns memory per the GPU's needs, which means this card is going to be doing alot to improve the Crossfire implementation. Guess I'll be holding out for this.

I don't think the PLX chip has anything to do with the memory system of the GPU. If the r700 was to use a central memory controller on a separate chip for both gpu's, it would result in horrendous latency, decreased bandwidth, and would pretty much offset any performance benefit of having a shared memory pool, IMO. From my understanding, the PLX chip is only used do divide the PCI-e lanes from 1 slot between 2 gpu's. The gpu's still access the memory directly, and if there was shared memory on the r700, it would likely be implemented on the gpu itself.

 

[Cookie Monster]

Just what are the advantages of a shared memory pool? It sounds good, but how is this going to solve issues when dealing with AFR?

But what im more concerned is heat and power draw. RV770s aren't cool nor power friendly (in retrospect with other cards from past generation). They produce quite an alarmingly high amounts of heat. Adding two of these GPUs and using one dual slot cooler could probably limit this cards clock speeds due to the TDP limit and the HSFs limit.

 

[Jumpem]

So where does this stand against the 9800GX2 and GTX 280?

 

[taltamir]

Originally posted by: Sylvanas
Lopri:

Gen 2 performancePAK?

Dual Cast?s
Read Pacing?
Dynamic Buffer Allocation


Hmm..

Good find! Could this mean a shared memory pool? It seems like we won't be stuck with the 512mb framebuffer per GPU limitation if the PLX chip 'dynamically' assigns memory per the GPU's needs, which means this card is going to be doing alot to improve the Crossfire implementation. Guess I'll be holding out for this.

with there being a GPU-GPU interconnect that completely bypasses the PLX chip, AMD should have no problem implemeting memory sharing in hardware, and rumors say that they have.
So regardless of which PLX chip they use, it can be expected to share the ram.

Originally posted by: Jumpem
So where does this stand against the 9800GX2 and GTX 280?

Annihilates both
By at least 50%

 

[Tempered81]

Originally posted by: Jumpem
So where does this stand against the 9800GX2 and GTX 280?

on top of them

 

[Jumpem]

Originally posted by: taltamir
Annihilates both
By at least 50%

Even though it is supposed to be cheaper?

 

[Stoneburner]

Originally posted by: Jumpem

Originally posted by: taltamir
Annihilates both
By at least 50%

Even though it is supposed to be cheaper?

If something sounds too good to be true, then it probably involves cutthroat competition between two implacable foes.

 

[Lonyo]

Originally posted by: Jumpem

Originally posted by: taltamir
Annihilates both
By at least 50%

Even though it is supposed to be cheaper?

HD4870 vs GTX260. HD4850 vs 9800GTX (until NV announced price cuts because they had to).
I think most people would agree the GTX260 and GTX280 are overpriced.
ATI pricing their cards lower isn't because it's not as good performance wise, but because they can.

The die size of a single RV770 is around half that of a G200? or whatever chip, which means they can get 2 for the price NV pays for one.
If we also assume better yields for it (because you don't lose 1 huge chip, you only lose 1 small chip out of 2), then ATI can sell their chips cheaper.
PCB wise, NV have a 512-bit memory bus which means expensive PCB, and so do ATI (ATI need to power 2 GPU's and have communication between them etc)
RAM price, ATI are probably losing out (since they run GDDR5 which is more expensive than GDDR3 AFAIK).

But basically ATI can probably produce the cards cheaper than NV do because their design is different, and since they can undercut NV pricing, they are doing. They won't be losing money on these cards, but they can stop people buying NV cards.

 

[bryanW1995]

it should be quite a bit faster than both 9800gx2 and gtx 280, but we'll know for sure soon enough.

 

[bryanW1995]

Originally posted by: Lonyo

Originally posted by: Jumpem

Originally posted by: taltamir
Annihilates both
By at least 50%

Even though it is supposed to be cheaper?

HD4870 vs GTX260. HD4850 vs 9800GTX (until NV announced price cuts because they had to).
I think most people would agree the GTX260 and GTX280 are overpriced.
ATI pricing their cards lower isn't because it's not as good performance wise, but because they can.

The die size of a single RV770 is around half that of a G200? or whatever chip, which means they can get 2 for the price NV pays for one.
If we also assume better yields for it (because you don't lose 1 huge chip, you only lose 1 small chip out of 2), then ATI can sell their chips cheaper.
PCB wise, NV have a 512-bit memory bus which means expensive PCB, and so do ATI (ATI need to power 2 GPU's and have communication between them etc)
RAM price, ATI are probably losing out (since they run GDDR5 which is more expensive than GDDR3 AFAIK).

But basically ATI can probably produce the cards cheaper than NV do because their design is different, and since they can undercut NV pricing, they are doing. They won't be losing money on these cards, but they can stop people buying NV cards.

I think that amd will be able to price this card closer to $600 if current market conditions continue. if we see it 30%+ faster than gtx 280 and 9800gx2 then amd would be crazy not to price this card accordingly imho. of course, I don't own amd stock and there is at least a small probability that I'll buy this or another card in the next few months, so I HOPE that amd goes for market share and keeps the price at $499.

How many people are going to buy 1 gtx 260 for $399 or 1 gtx 280 for $649 if 4870x2 is really $499???

 

[Munky]

Originally posted by: Cookie Monster
Just what are the advantages of a shared memory pool? It sounds good, but how is this going to solve issues when dealing with AFR?

But what im more concerned is heat and power draw. RV770s aren't cool nor power friendly (in retrospect with other cards from past generation). They produce quite an alarmingly high amounts of heat. Adding two of these GPUs and using one dual slot cooler could probably limit this cards clock speeds due to the TDP limit and the HSFs limit.

At the very least it should solve the problems at very high rez where each gpu is starved for memory due to having duplicate data between the two of them. But ideally, having shared memory would be a step toward implementing a more elegant rendering method than AFR; I'm hoping some sort of tiling or SFR implementaton would be available on the r700.

 

[Stoneburner]

Munky, why is SLI just NOW a good idea? Has gtx280 resolved the common issues with SLI?

 

[TC91]

Originally posted by: Stoneburner
Munky, why is SLI just NOW a good idea? Has gtx280 resolved the common issues with SLI?

i think hes talking about SLI 8800gt's (or similar) being a better option than a single gtx 280 for the price and performance.

 

[Munky]

Originally posted by: TC91

Originally posted by: Stoneburner
Munky, why is SLI just NOW a good idea? Has gtx280 resolved the common issues with SLI?

i think hes talking about SLI 8800gt's (or similar) being a better option than a single gtx 280 for the price and performance.

Yep. Up until now a single new high end card offered about 2x the performance of two older generation cards, often with better image quality and features thrown in. So there was practically no reason for someone with the older card to get a second one as an upgrade to SLI. But now it actually makes sense to go SLI with cards like the 8800gt (and similar Ati cards in crossfire) for better performance and lower cost than a single gtx280, and you won't even be missing out on any new gaming features.

 

[Jessica69]

Originally posted by: munky


Yep. Up until now a single new high end card offered about 2x the performance of two older generation cards, often with better image quality and features thrown in. So there was practically no reason for someone with the older card to get a second one as an upgrade to SLI. But now it actually makes sense to go SLI with cards like the 8800gt (and similar Ati cards in crossfire) for better performance and lower cost than a single gtx280, and you won't even be missing out on any new gaming features.

So very true. You gotta feel bad for all the suckers, errrrrr, consumers that bought the GTX280 already......visions of $$$ being flushed down the commode come to mind.

The pain........the pain of it all.........

 

[Cookie Monster]

Well if i had the money and lots of it, id sure grab the GTX280 (along with a 30inch screen). Why wouldn't you not? If your budget minded, then its crazy. They would lose sleep over the amount of $$$ they just flushed down the toliet. Maybe even go insane.

 

[BTRY B 529th FA BN]

Sign of the times

 

[biostud]

Originally posted by: BTRY B 529th FA BN
Sign of the times

and nvidia think that will help selling more cards...........