Sign in with
Sign up | Sign in
Your question

R700 speculations

Last response: in Graphics & Displays
Share
a b U Graphics card
June 20, 2008 10:20:28 PM

This thread is for speculation on the R700, the dual or x2 card coming from ATI. This card holds much mystery, as it has a great potential to change the graphics market as we know it. This all may be rumor, but theres many people interested in this possibility of change. What we may have is my poor explanation of it but here I go. The R700 will have a PCI-e 2.0 bridge but also possibly an innercomunication bridge for the 2 cores. Whats this mean? If the 2 cores communicate together, the speeds, or diminishing returns we used to see in typical x2 or sli/cf would be dramatically reduced. How this communication is doen is the mystery. Also, we all know that it will bring with it the first use of GDDR5. This memory may have the ability to aid this cross communication between cores, as well as hugely widen thruput/bandwidth. This is a poor interpretation of what may be, but its what I understand as the potential of this card, tho with many holes and of course, no direct knowledge. Heres a link of the core http://gathering.tweakers.net/forum/view_message/302819... If anyone has more to add, or any enlightenment, or corrections, please add them as needed

More about : r700 speculations

Anonymous
a b U Graphics card
June 20, 2008 10:26:39 PM

thats what I understand...

and I think the 4870 X2 may actually have 100% scaling... idk why but I do and that would mean that a 4870 X2 would be 1.4 - 1.5 times a 4850 crossfire... which would put its 1920 by 1200 crysis performance at very high at just over 30...

thats awesome... still no AA... but w/e

I'm still thinking these cards aren't needed... maybe in a year or two... but what game is there that a current card can't play? I mean besides crysis
June 20, 2008 10:32:50 PM

Yer i hope the R700 core will change the market, the results as ive seen from the 4850 series seem promising and i hope the core will compare close to the GTX 280 or better.

ATi needs a come back, and what better than the R700? (Accept when they release and X2 version)

The GDDR5 should atleast give the core tht little extra "umph"

Ive got my eye set on one (or 2) of these cards, and i just hope ATi succeed! The pontential is there as Jaydeejohn said, its just grabbing the bull by the horns and delivering.

kkthanksbye
Related resources
a b U Graphics card
June 20, 2008 10:43:13 PM

Currently, we see things like non compatible sli/cf games, where 2 cards simply wont work. This may change that. Another problem with multi card setups is micro stuttering, which causes delays between the cpu in a rhythm feeding to gpu in rhythm, which causes delays in the afr layout. One actually waiting for the next. If the OS sees this setup as 1 card, then that may be the solution to micro stuttering. This is important, as I said, because if this is successful, then multi core will be opened for the future, and the single core will quickly go the way of the dinosaur
June 20, 2008 10:50:02 PM

I can see what your saying, has this been incorperated into the design of the cards? If it has then hands down to ATI, instead of just a small percentage increase of performance, they could be looking around 75-100% which would be a new era, soon we will be looking at synchronous quad graphics with 4k wattage PSU's. This would be an amazing jump in the graphic development!!!

I hope they pull that off!!!

kkthanksbye
June 20, 2008 11:01:47 PM

gah! i forgot the link where i read this from...but i'll try to find it and post it up

but from what i've read, the 4870x2 aka R700 will NOT use a pci.e 2.0 bridge, it will communicate the two GPU chips through memory...

vaguely as that term is on the reading i read, to my knowledge, i dunno how this works nor do i think any of us do at this point

but i guess it suggest there is no pci.e 2.0 bridge like the pci.e 1.1 bridge of the 3870x2

and from what i remember, the article said there is no micro stuttering on the 4870x2 ... now to look for that article to post it :) 
June 20, 2008 11:06:24 PM

http://www.nordichardware.com/news,7845.html

that was the link i read...but elsewhere on the internet, you can goggle 4870x2 bridge or memory bridge.....and a lot of sites will pop up suggesting the same thing, that the R700 will communicate via memory thus eliminating micro stuttering :D  yay

count me in for one... -feeds piggy bank-
a b U Graphics card
June 20, 2008 11:29:20 PM

Nice. If they can really pull this off like we're speculating I think it will really validate their current strategy of making small, cool chips that are slower than NVidia but x2 are much faster. I be it saves them a bundle on design and testing.
June 20, 2008 11:49:58 PM

aznguy0028 said:
http://www.nordichardware.com/news,7845.html

that was the link i read...but elsewhere on the internet, you can goggle 4870x2 bridge or memory bridge.....and a lot of sites will pop up suggesting the same thing, that the R700 will communicate via memory thus eliminating micro stuttering :D  yay

count me in for one... -feeds piggy bank-


Same page, different article. The same rumor with a little more detail. Still a rumor though.

Link
June 20, 2008 11:53:51 PM

EXT64 said:
Nice. If they can really pull this off like we're speculating I think it will really validate their current strategy of making small, cool chips that are slower than NVidia but x2 are much faster. I be it saves them a bundle on design and testing.

The biggest saving will be manufacturing. Just look at intels mcm compared to amds monolithic design. They can throw away the defective chips, can pair up the good ones and can use a higher percentage of a wafer. The key is the scaling and a solid interconnect is the most important part for that. Given AMD is involved they should have solid experience at hand on that matter. Sadly, they will botch up the execution. Again.
a b U Graphics card
June 21, 2008 3:39:38 AM

Just to help try and end the confusion, it'll likely use the following PXL bridge chip;
http://www.plxtech.com/pdf/product_briefs/PEX8648_Produ...

Which has the nice benefit of the 'Dual Cast'; and along with GDDR5 clamshell mode means that at both ends (bridge and memory) they have the ability for I/O sharing.
a b U Graphics card
June 21, 2008 4:01:18 AM

Im hoping its both ends...on to reading
a b U Graphics card
June 21, 2008 4:08:47 AM

Glancing at it dual cast looks nice , and not sure if this is right, but read pacing could help in elimating micro stutter?
a b U Graphics card
June 21, 2008 4:40:05 AM

Read pacing would only help bring down the fast frames not bring up those that make micro-stuttering noticeable. Really the way to avoid microstuttering is to improve performance and play at setting that don't dramatically drop a single frame.

It would act like v-sync in a way, but that may or may not smooth out framerate overall if it gives the next frame time to render. But when pushing the boundaries when Micro-stuttering would be most noticeable I don't think it would help on that frame that take noticeably longer (like 3-10x as long) to render than the previous frame.
June 21, 2008 7:34:09 AM

Do we know if the GPU is going to 2 chips or a dual core?
June 21, 2008 7:57:27 AM

i have seen an gpuz of r700
but it reveal that the card uses gddr3
though only with gddr3 r700 kill gtx 280 perfect
nvidia may be in lots of troubles hehe
Anonymous
a b U Graphics card
June 21, 2008 2:22:10 PM

^ what...

r700 has gddr5

June 21, 2008 10:24:21 PM

I thought i'd quote myself from another thread and drop it here just because its pertinent to the topic:

What card beats what card at the high performance level might depend on whether or not Nvidia can get their SLI driver on the same level as ATIs new Crossfire that is speculated to be released with the 4870x2

If the 4850 is any indicator, it shows that the gpus scale well in Crossfire mode

Actually, AMD themselves have stated that their aim was to create scalable gpus for Crossfire.

What I think is going to happen in reality is this, and i'm going out on a limb to say this but its just sort of a gut feeling i'm getting.

4870 is not going to be a huge jump up single card vs single card over the 4850. There will be differences, but it won't be astronomical like people are expecting.

However, what the 4870 WILL be able to do if the 4850 is any indicator, is get /significantly/ better scaling in Crossfire than Nvidia can get in SLI. This may be where ATI will really rock Nvidia is in the multi-gpu scaling department. The 4850s scale incredibly well from what tests i've seen.

Meaning as thus, 2 4870s might not beat gtx280 SLI, but for the same price as gtx280 SLI - you could get 4 x4870 Quad crossfire on an AMD790FX platform - or wait for the 4870x2 for Quad crossfire/Tri crossfire on the Intel platform and achieve better performance per dollar at varying points of reference.

The 4870x2 is supposed to be superior to a Crossfire bridged gpu setup - its supposed to have a better controller, I don't totally understand this business yet cause there isn't enough information - but it is "supposed" to have a much better Xfire technology than the 3870x2 - This should put the 4870x2 Quad crossfire above Nvidia's 3way SLI gtx280 if it scales the way that they have implied it will.

Just pure speculation, but going by AMD's marketting this is what i've been gleaning.
June 25, 2008 10:14:06 PM

The 100% scaling idea is interesting to think about. In a single generation update, I don't think ATI can get 100% scaling across the board (I would love to be proven wrong with the mysterious 4870x2)...it is just too big of a leap. It should still be part of the big picture or the master plan for the dual GPU solutions, but in the mean time, they need to take the baby steps towards 100% consistent scaling across the board rather than 100% scaling.

As a consumer, a dual GPU card that can scale lets say at a solid 85% in EVERY game is very appealing. Personally, I don't want a card that gives me 98% scaling in a game that I don't play and 70% in the game that I do. However, if there was a guaranteed 85% scalability....that would change the market and be just what ATI/AMD needs to step out on top.
June 25, 2008 10:23:29 PM

Somehow I can just picture the 4780x2 releasing and trouncing benchmark after benchmark against the GTX 280 (and maybe matching SLI'd 280s), and then, simultaneously, hardware enthusiasts worldwide jump from their chairs, throw their arms up in the air and just shout 'YEAH!!!!'

I admit, my expectations are a tad high. It'd be way awesome if the 4870x2 is the 9700 PRO of this generation, but then again this is a fairly new idea and ATI did botch a few things with their last two generations, so it's hard to say what'll happen.

Go ATI!
June 26, 2008 12:12:11 AM

Here is a snippet from Anandtech.com.

Quote:
The CrossFire Sideport

Although AMD isn't talking about it now, the CrossFire Sideport is a new feature of the RV770 architecture that isn't in use on the RV770 at all. In future, single-card, multi-GPU solutions (*cough* R700) this interface will be used to communicate between adjacent GPUs - in theory allowing for better scaling with CrossFire. We'll be able to test this shortly as AMD is quickly readying its dual-GPU RV770 card under the R700 codename.


a b U Graphics card
June 26, 2008 12:16:14 AM

Yeah when I read that I was thinking more along the line of dual die single package future, and allowing them to test such a concept early.

We'll see, it's pretty easy to take these tidbits and turn them into a giant soufflé of hopes & dreams only to have it crash down with the reality of what they actually bring to market.
June 26, 2008 1:07:07 AM

TheGreatGrapeApe said:
Yeah when I read that I was thinking more along the line of dual die single package future, and allowing them to test such a concept early.


That would be amazing. Given the heat produced by each chip as shown in various articles, i'm not sure how they would cool that package. Probably underclocking combined with a dual-slot cooler.
a b U Graphics card
June 26, 2008 1:35:58 AM

How much, if any driver optimisations would help with temps at load?
June 26, 2008 1:41:57 AM

JAYDEEJOHN said:
How much, if any driver optimisations would help with temps at load?

You can optimize drivers, sure, but once they do, they have to choose either performance or heat. Sometimes they will be able to get better performance and less heat, but those circumstances are rare. Most likely it will only get hotter with better drivers since the card will be "more" fully used.
a b U Graphics card
June 28, 2008 1:48:37 PM

I remember there being said by AMD somewhere, cant find that link, that this pci bridge was the solution, something they hadnt pursued in depth in the past, well it now seems to have surfacedagain "The previous ATI dual-chip solution was very different, Peddie said. "The HD 3870 X2 was not a proprietary bus but a CrossFire connection. The CrossFire connection and the SLI connection are at the very, very end of the pipeline. Not the most efficient place to do an inter-processor communication. That's one of the reasons ATI has abandoned it."

Sounds like they moved up their inter-processor communication, and upped the speed also. Heres some more "For ATI, the execution of this chip-ganging strategy is the key. And this is where ATI appears to have been successful. "The inter-processor communications. Getting that to work has been the trick. This is what ATI has done. They've come up with this stellar way of doing inter-processor communications so they can in fact get the scaling," according to Peddie.

Note, this is John Peddie, yes THAT John Peddie. It appears ATI may have a jump on nVidia, as graphics demands has outstripped design shrinks, thus making a larger die and evetual dead end. Maybe this will be the beginning of a new graphics era http://news.cnet.com/8301-13924_3-9978746-64.html Also, pics of the X2 board can be found here http://news.cnet.com/8301-13924_3-9978746-64.html
a b U Graphics card
June 28, 2008 2:22:13 PM

The number theyre trying to get is 1.8 times performance. What this means is, not or unlike CF, this would be a permanent scaling. So, we wouldnt see the ups and downs that sli/cf brings. It would be consistent. If this proves out, 1 X2 4870 would be better than any G280 sli setup, and a CF x2 would simply blow anything weve seen away. Maybe Crysis will be conquered after all. At 2.5 times a 4870 in CFX4x, I think wetre there
Anonymous
a b U Graphics card
June 28, 2008 2:28:48 PM

I'm hopeful =D
a b U Graphics card
June 28, 2008 2:33:48 PM

To see 80 fps at 16x10 for Crysis, maybe we will have room for AA? heheh
a b U Graphics card
June 28, 2008 3:26:56 PM

Something to be considered. The PLX chip sounds like its at the center of all this capability (no pun intended) , but whats interesting is, I keep hearing in other places, that this bridge chip doesnt have enough thruput to do any real decent changes. Ive heard of their 12 port design, but what Im wondering is, I believe ATI has their own design coming from PLX as mentioned here. "Still PLX and probadly they have to use the 12 port device, since PLX does not offer something other."

"Depends on the partner. "
This I believe is very important in understanding the abilities of this new chip
a b U Graphics card
June 28, 2008 4:07:38 PM

JAYDEEJOHN said:
The number theyre trying to get is 1.8 times performance. What this means is, not or unlike CF, this would be a permanent scaling. So, we wouldnt see the ups and downs that sli/cf brings.


I think that's still very optimistic on their part, because while it may mean a 'near true 2X of shaders or TMUS/ROPS' as seen by the app, even doubling shader or doubling other components doesn't guarantee a X.x times performance boost across the board, just look at the 2.5X boos in most components of the HD4K over HD3K.

I understand what they're saying about it being less Xfire/Software dependant, but I think it's an over-optimistic statement to say it will be a global 1.8X improvement without and ups/downs.
a b U Graphics card
June 28, 2008 4:21:16 PM

Thats true. OK, so what if CF actually outperforms this new bridge connect? Would there be a old CF option?ala sideport?
a b U Graphics card
June 28, 2008 4:30:08 PM

JAYDEEJOHN said:
Something to be considered. The PLX chip sounds like its at the center of all this capability (no pun intended) , but whats interesting is, I keep hearing in other places, that this bridge chip doesnt have enough thruput to do any real decent changes. Ive heard of their 12 port design, but what Im wondering is, I believe ATI has their own design coming from PLX as mentioned here. "Still PLX and probadly they have to use the 12 port device, since PLX does not offer something other."

"Depends on the partner. "
This I believe is very important in understanding the abilities of this new chip


Well it's obvious that it's the PLX chip now that clearer pictures are out. However I agree, we don't know if it is the bridge we've all been looking at or an improvement on it. Notice they whited out the part #/descriptors in the Chiphell pics.

However having both the CrossfireX sideport and unfettered bridge bandwidth at your disposal is alot.
The main advantage of moving to a dual package solution in the future is going to be communication at chip speeds and much much lower latency.
a b U Graphics card
June 28, 2008 4:33:42 PM

JAYDEEJOHN said:
Thats true. OK, so what if CF actually outperforms this new bridge connect? Would there be a old CF option?ala sideport?


Oh yeah the bridge and CF are not mutually exclusive, I just don't know if they bothered to use the traces in the board. Since the sideport is built into the VPU it'll always be an option for both.

Remember it's likely to be a tradeoff between the two, but I'm doubtful there would be many situation where plain CF would beat a similar speced card, however the main benefit of the bridge and R700 solution of course is the memory properties, and until we find out their ability ot make the drivers work I don't know.
a b U Graphics card
June 28, 2008 4:38:32 PM

And reducing the latency would eventually eliminate micro stutter. Im thinking this PLX chip is wider/faster than what we now know, allowing for reductuions in latency as well as providing more interconnectability between the 2 chips. Just a guess, but why would DB comment on this, if it didnt mean just that?
a b U Graphics card
June 28, 2008 5:17:43 PM

I don't know about that.

See the thing about micro-stutter, I still don't think people realize sometimes a single scene/frame will be slightly hard to render than the one before or ater it, so while it might reduce it, it might never eliminate it.

However people wouldn't call it micro-stutter anymore, just frame drops (like frame tearing) where it was just an occasional thing. I think it'll alway happen to some extent when you push the very envelope of what one chip can render even with great chip level communication. It would be like exhausting an old pipeline, you didn't call that micro-stutter, you just called it a dropped frame.

I don't know Wavey can be pretty criptic at times, especially now being so deeply embeded he has to chose his words carfully enough, but it's definitely a least a 2 fold increase over the previous bridge with PCie 2.0 throughout.

If they've improved it beyond the 8648 that would be interesting, but likely a very hot chip.
a b U Graphics card
June 28, 2008 5:27:38 PM

One can hope!!
June 28, 2008 5:38:59 PM

TheGreatGrapeApe, you're in Calgary eh? I just moved from there to Germany last year :) 
a b U Graphics card
June 29, 2008 8:00:41 AM

Enjoy, I used to live in Boblingen just outside Stutgart when I was a kid, been back a few times since then.
June 29, 2008 10:01:52 AM

form the latest news we konw r700 will beat 280 quite easily
though nv has "the way"
June 29, 2008 11:33:31 AM

Has there been any updated word or rumors on the release date?
June 30, 2008 12:27:28 PM

No, but plenty of people will tell you otherwise.

I believe late Q3 was the last schedule offered up by AMD.
a b U Graphics card
July 2, 2008 1:34:50 AM

Last week of July. And latest rumor has it running faster than 4870 Xfired in Crysis
July 2, 2008 1:59:38 AM

Paris Lee said:
form the latest news we konw r700 will beat 280 quite easily
though nv has "the way"



"The way" or not, its already been shown that dual gpus beat out the gtx280 in average fps long before the 4800 series were released

But the problem is min framerate, not avg frames

I think the 4870x2 will be just as consistant as the gtx280 and possibly have a higher max framerate to boot; the fact is though when you're talking about this degree of power, you have to be thinking more about quality modes than just "fps in call of duty"

Nvidia has transparency antialiasing and their custom filter AA's /quality AA
ATI has a superior custom AA as far as I know, but no card was able to use it viably before the 4800 series - I don't know if the 4800s have something similar to transparency AA or not, or if its maybe built into one of their custom AA modes; info on this would be good.
July 2, 2008 10:00:22 AM

What prevents ATI from using the same method that Intel used in their early multicores. "Gluing" those cores together? Is it because the power comsumption? Or does this CrossFire Sideport support the same idea?


a b U Graphics card
July 7, 2008 4:32:58 PM

The latest rumor is we will see previews July 14th and release 2-3 weeks later. Cant wait
!