4870 in CF vs potential 4870x2

[satanpro]

I realize the 4870x2 has very little information out about it, but could someone who knows a lot about such things comment about the performance between two 4870 cards vs a 4870x2? I would appreciate information regarding the 3870 vs 3870x2 if that's relevant to the 4800 series.

I've read these kinds of questions asked in other forum topics, but never answered in a reasonably complete way.

Thanks!

 

[hannibal]

The 4870 eats a lot of power (read run hotter), so expect to see bigger difference in clockspeeds (compared to 3870 counterparts)
On the other hand it is rumored that 4870x2 has a new way of communicating between gpu's...
We all are waiting to see what happens.
All in all two separate 4870 should have more computing power / GPU, but 4870x2 may scale better, so it's guite difficult to predict this.

 

[Kaldor]

If ATI can get their Xfire scaling working right in more than just a few games, the 4870X2 stands to be the best single card solution out there. 2 4870X2s stands to be the gold standard.

But like Hannibal said, "it's guite difficult to predict this."

This is a wait and see. It all depends on ATIs driver department now.

 

[badgtx1969]

Rumor says 4870x2 will have a new and improved PCIe 2.0 bridge and better GPU interconnect. Not a lot is known as to what enhancements were built into RV770 for R700, since most of the architecture is a closely guarded secret. My guess is we'll have to wait until the release of the 4870x2, and I bet we'll see some cool new things including improved scalling over two 4870s.

http://www.nordichardware.com/news,7878.html

 

[satanpro]

So am I correct to assume comparisons between the 3870 and the 3870x2 are not valid here, seeing how new technology to optimize the two gpu card is rumored to be present on the 4870x2?

One thing I'd like to clarify: are the best case scenario predictions that the 4870x2 will out-perform two 4870 cards in crossfire? Or is it only a matter of how close to the performance of two 4870 cards in crossfire the x2 can get?

I think the answer to that question will dictate whether I'll wait for the x2 or not.

Thanks for the continued help.

 

[Guest]

^ boom that's the question for me... except for the price... a 4870 X2 is supposed to be a 100 - 150 $ cheaper... even if its not exactly equal to 2 4870's... i'd get the 4870 x2.... but I think the 4870 x2 will be better than 2 4870s

EDIT: Just remembered this... the system will see a 4870 X2 as a single card... all the communications between the chips will occur through the PCB... so this will mean that 2 cards can work as one in a non crossfire supported game... that way where a non crossfire game gets a little scaling.... the 4870 X2 would still get full scaling... so that's why I think the 4870 X2 is better even if its slightly worse

 

[jevon]

Yeah, CrossfireX scaling shouldn't be an issue with the 4870x2 since for intents and purposes, it's read as a single card. Where it lands on the performance and energy/heat scales remains to be seen, but unless you're using a 24"+ LCD I'm not sure if it's the best purchase for you.

 

[homerdog]

The 4870X2 should scale significantly better than a pair of 4870s. There's even talk of it being seen as one card by the OS and scaling across the board (unlikely IMO unless they ditch AFR...). Anywho here's what all the fuss is about:

See that little box in the bottom right labeled "CrossfireX Sideport." That is supposed to be the key in allowing the two RV770s on the 4870X2 to share the framebuffer; that's right, 1GB actually means 1GB with R700.

Also there has been speculation that R700 will not suffer from any micro-stuttering issues, meaning the FPS you measure will be the FPS you perceive.

Needless to say R700 has me very VERY intrigued

 

[Guest]

^ are you saying the Crossfire bridge is still there? As far as I know the cards communicate through the memory... but yes micro-stuttering shouldn't be an issue...

Jevon I do in fact have a 24" monitor and i'm still rockin' my 8800 gts 640...

PS. got it for 250 to... great deal XD

 

[TheGreatGrapeApe]

Aslo add to that a few other items discussed in JDJ's R700 thread, and this could be a whole new ballgame for CF scaling.

http://www.tomshardware.com/forum/251872-33-r700-speculations#t1830674

Take note of the new features for GDDR5 and the likely PCIe bridge candidate.

 

[TheGreatGrapeApe]

Like I said in that thread I still think there'll be a PCIe bridge, but the updated bridge is far more advanced.
(edit might as well just post the direct link to bridge specs/ features): http://www.plxtech.com/pdf/product_briefs/PEX8648_Product_Brief_v0%2095_16Jun08.pdf

BTW, micro-stutter isn't caused by or aleviated by memory alone, it's a question of getting the GPUs to synch correctly and render close to similar intervals. Shared memory would help improve performance which will help avoid micro-stuttering, but I don't think it will simply solve micro-stuttering, which is just like any framerate drop on a single card, only it's got another card to try and keep up with so it's created this new phenomenon.
The reason there's micro-stuttering onn AFR and not SFR, is that SFR ha to wait to display both cards, and when one is slow both are stalled, and then it's just a low framerate, not inter-frame low/hi framerates.

 

[satanpro]

Thanks for the information everyone. I appreciate the advice from you guys who are much more technically proficient than me!

 

[homerdog]

^ are you saying the Crossfire bridge is still there?

Yeah but as Ape said it's significantly more advanced.

BTW, micro-stutter isn't caused by or aleviated by memory alone, it's a question of getting the GPUs to synch correctly and render close to similar intervals. Shared memory would help improve performance which will help avoid micro-stuttering, but I don't think it will simply solve micro-stuttering, which is just like any framerate drop on a single card, only it's got another card to try and keep up with so it's created this new phenomenon.

I know it sounded like I was implying that shared memory was the reason for the elimination of microstutter, but that's not what I meant. I'll make that bit a new paragraph

For those interested, here are some good explanations of micro-stuttering:
http://forum.beyond3d.com/showpost.php?p=1171992&postcount=2669
http://forum.beyond3d.com/showpost.php?p=1172004&postcount=2674

 

[moomoopro]

good info thanks

 

[lameness]

If the os sees the 4870x2 as a single card then inst it conceivable that you could have 4 X 4870s?
lol that would handle crysis quite nicely thanx

One last question: Is there DEFINITELY going to be a 4870 x2? All there is so far is hearsay right?

I mean the 4870 power usage isnt too far behind the gtx280 and everyone was in stitches about the prospect of having a gtx280 x2 ...that woulda been daft they said....far too much power and heat chucked out by two cores too close...its impossible.

And yet here we are with the 4870, ac hip that despite being great, uses a lot of power and runs hot enough to boil a kettle. But still we are as confident as ever about the prospect and credentials of the 4870x2?
Does ATI have some trick up their sleeve, or are people being a little optimistic?

 

[satanpro]

I assume it's a sure thing that there will be an x2, it's been in all the AMD slides. If FUD and Inquirer mean anything, they've hinted that the X2 is coming sooner than August.

 

[homerdog]

If the os sees the 4870x2 as a single card then inst it conceivable that you could have 4 X 4870s?
lol that would handle crysis quite nicely thanx

One last question: Is there DEFINITELY going to be a 4870 x2? All there is so far is hearsay right?

I mean the 4870 power usage isnt too far behind the gtx280 and everyone was in stitches about the prospect of having a gtx280 x2 ...that woulda been daft they said....far too much power and heat chucked out by two cores too close...its impossible.

And yet here we are with the 4870, ac hip that despite being great, uses a lot of power and runs hot enough to boil a kettle. But still we are as confident as ever about the prospect and credentials of the 4870x2?
Does ATI have some trick up their sleeve, or are people being a little optimistic?

There will definitely be a 4870X2. It is called R700.

As for power concerns, most of what I've read indicates that the PowerPlay stuff isn't working yet in the current drivers, so idle consumption is higher than it should be. As for absolute TDP under load, they could always downclock R700 a little like NVIDIA did with the 9800GX2. The small reduction in clockspeed shouldn't make much difference if this card ends up as fast as we think it will.

And 4 of any card is a bad idea, that would give you a minimum input lag of 4 frames; maybe that's alright for benching, but a delay like that isn't acceptable in any game that I play.

 

[satanpro]

As for absolute TDP under load, they could always downclock R700 a little like NVIDIA did with the 9800GX2. The small reduction in clockspeed shouldn't make much difference if this card ends up as fast as we think it will.

This concern is the exact reason this thread was created. Will this downclock potentially reduce the performance of the X2 below that of two 4870 cards in crossfire?

 

[lameness]

This concern is the exact reason this thread was created. Will this downclock potentially reduce the performance of the X2 below that of two 4870 cards in crossfire?

Im pretty sure that nobody other than an ati technician can answer that question, and if they do, chances are they dont have much evidence to back it up.

I think they will have to reduce the clocks of 4870x2 by a fairly significant margin (50mhz or more). But then as you have heard scaling is rumoured to be better. Personally i suspect that there wont be much between them. If best performance is your only concern then 2 seperate oced (factory or otherwise) 4870's will be the fastest (i guess).

 

[satanpro]

If best performance is your only concern then 2 seperate oced (factory or otherwise) 4870's will be the fastest (i guess).

Thanks for the advice. Can some others confirm or deny this as being probable?

 

[homerdog]

Thanks for the advice. Can some others confirm or deny this as being probable?

R700 > 2 4870s. Refer to my previous post for reasons.

 

[duzcizgi]

As far as the rumor goes, R700 is a dual core system which works much like an SMP. They'll share the framebuffer memory and synchronize directly through the Crossfire bridge.
I guess one of the GPUs would be the "master" and the other slave. To the OS, only master would be visible. Slave being utilized only when necessary. For example during idle, slave GPU being shut down totally. But when it enters 3D acceleration mode, it kicks in.
Of course, this is where the rumor goes.

 

[homerdog]

Looks like R700 will use AFR so a lot of my early optimism has resided. I still think it will scale significantly better than a pair of 4870s, share the framebuffer, and be free from microstuttering but application dependency and input-output lag should remain

 

[satanpro]

What about the memory size?

Running at 1920x1200 and looking to use max settings possible, including texture packs for relevant games, would I benefit more from two 4870 1gb editions (providing they appear) over the 1gb shared 4870x2?

If the x2 can share memory and use the 1gb all to itself, unlike previous x2-style cards, let me know. I'm not clear on what 'sharing the memory' on the x2 actually means.

Thanks!

 

[lameness]

As far as i understand the rumour is that x2 basically works like a normal card only the core is split in two...ie two chips. The memory is thought not to split ie yes it will have 1gb of memory not 2x 512.

I will believe the rumours when the card is launched. As far as iam aware ATI have not confirmed anything (has this changed anyone?)..... only that it will have improved scaling. I really want this card to be as good as the rumours, but rumours are seldom true.

All the rumours said that the 4850/4870 would have 480 shaders (seemed reasonable to me)...thats rumours for you.

A good saying that i like to remember is "nobody knows anything" 9 times outa 10 rumours are wrong, and when they are right its mostly by chance rather than design.