damn.....i'm almost sold :D :p:
Printable View
damn.....i'm almost sold :D :p:
Latest news is that it has been confirmed, 4850 and 4870 have 800 SPs:
http://www.nordichardware.com/news,7841.html
Yep been posted
And it was in an article somewhere from some engineer on the R600 project but basically he stated that R600 ALU's are cheap in transistor count
And I certainly echo the possibility that TMU's reach a point of diminishing returns I feel like. While G92 was heavily bandwidth limited, its doubled TMU count over G80 didn't give it any considerable advantage, even with clock advantages. Heck, the G80 GTX and Ultra do just fine with the 32/64 ratio compared to the 64/64 on G92.
Nvidia's seems to certainly have bet years ago that texture units would be a big concern (seeing as how they increased them greatly from G80 to G92 to GT200) and maybe less shader intensive whereas ATI believed shaders would rule the day and textures wouldn't be as big a concern. We'll certainly see soon though
I am not believing all this just yet, but I would love to ATI pull an R300 out of the bag with the GTX going the way of the 5900 :D
I do doubt the 800 SP rumours because 800SPs increases the die size dramatically, and since the process is the same as the 3870, we know that doubling the TMUs and more than doubling the SPs would be next to impossible without having a gigantic diagonal 90nm R600 size die:
http://pcarena.pl/uploads/files/wale...wsy/R600_1.jpg
As we have already seen the Rv770 die size and we know that the die size is only 50% larger, the 50% more SP theory (480 SPs) sounds more practical.
Those are my 0.02.
Perkam
that looks more like 10 yen :D
What would the SP's be doing anyway?
Werent there rumurs that the extra 320SPs would only do Physx, not giving extra graphical power
Quote:
Originally Posted by perkam
:rofl:
You've not been posting much lately, so I understand.
but Perkam......
:shakes:
Here, you can have them back....:rolleyes:
SP's alone to not make a gpu. ROPs, TMUs, cache, memory interface, and any other logic also account for die space. Since we are talking ati cards here, lets add in UVD/HDMI functionality, etc...doubling SP's @ 50% die increase seems more than reasonable to me.
I mean really, how else is R770 different from R670?
As I've shown before, die size & processing resources do not scale linearly at all.Quote:
Originally Posted by perkam
Case in point, RV635 (55nm/118mm^2) -> RV670 (55nm/192mm^2)
ATI fits 2.66x SPs, 2x TMUs, 4x ROPs, increased cache, plus twice internal ring bus size (512-bit vs 256-bit) in a die only 63% larger.
So it is not unlikely they can fit 2.5x SPs/2x TMUs, with all else the same, in a die 34% larger.
Comparing RV620 & RV670 just exaggerates the point.
RV670 is approximately ~3x RV620 and it features 8x SPs / 4x TMUs / 4x ROPs / 4x ringbus (512-bit vs 128-bit) / vastly more complex cache structure.
http://img.hexus.net/v2/internationa...CF0051-big.jpg
does anybody know anything about 3950 ?
lol well if it isn't ol' Cadaveca :p:Quote:
Originally Posted by cadaveca
What happened to your avvy with your old case in it :D
Anyhoo, they have also added 32 TMUs, and though your point is valid, 800 SPs is generally believed to be overkill unless SP efficiency is REALLY bad and they need to do it just to increase performance by significant amount.
As for your point about "what difference is there between 3870 and 4870 if they do that", take a look at the prices, they are almost identical to 3870 and 3850 prices at launch. If they really wanted to launch a 800 SP monster, they would have put the price similar to the GTX 260 at $449 and not $329 and $229. We have heard news that the reason for the low 4850 clock was because ATI does not want the 4850 to beat the 3870 by a large amount otherwise no one would buy the 3870 and that makes sense.
I just don't think ATI can put 800SPs on the same 55nm architecture as the 3800 series.
Perkam
The Gigabyte placeholders are pretty much the only thing we have seen of the HD 3950.Quote:
Originally Posted by eric66
It's most likely a RV670-based part with higher clocks than the HD 3870. I suppose it could be a chopped-down RV770 part, but I doubt it would be considering it is a HD 3000-series part.
Life is easier being a nobody than a somebody. ;)Quote:
Originally Posted by perkam
You are missing the point. That die increase has to account for something. TMU's? :rofl: you mean more cache?:ROTF:Quote:
Originally Posted by perkam
Ok, so we have accounted for 10%-12% of the extra 50%, were's the remaining 38%-40%?
What else in R770 can account for that very large transitor count? 666m transistors increased to 1b transistors...why?:shrug:
:welcome:
Cost is not something to be considered. We are still living out ATI's legacy...AMD products are not here yet in ATI line-up. This is not ATI anymore, Perkam, so old price models definately do not apply here. I mean really...3870x2 was top-level card...old ATI management would have never made such a silly move...you must rethink strategy here.
Well I'm all for it Cadaveca, and I hope you're right...it's just not easy being optimistic about the best case scenario ;)
In other news, HD 4870 or this...hmmm...decisions decisions :p:
http://www.newegg.com/Product/Produc...82E16814102723
Perkam
lol. Exactly how it has always been...tried to buy an X850XTCrossfire, x1900XTX or x1950 Crossfire master card, or say 2900XT 1GB? New? You'll pay about the same price.
It's called retirement...
Maybe you're right, i always assumed 800SP was BS because of the $300 price point, if it isn't then i hope ATI owns Nvidia in the mainstream market :up: Everyone else can have their inefficient expensive GPUs.
Come on 800SPs!!!!!!
I cant afford a GTX 260! So I (and probably the majority of enthusiasts ) will be buying in the 200-250 euro range
I feel a bit stupid trying to tone down the importance of my own news report, but here goes:
Does it really matter if it has 480, 800 or six gazillion shader processors when the only thing that matters is the number of frames per second you will get?
You have Vantage scores (don't go by the ones I posted, I barely trust those myself) and they are certainly good for a mainstream card, and I can certainly tell you that the current drivers suck in Vantage.
//Andreas
You will get what you pay for ( 2 year old high end performance ), which will be much slower than the GTX260..Quote:
Originally Posted by conzymaher
Anyone on a decent card like a 8800gts 640mb, 8800gt etc, would be a bit crazy to get a card that will only give a few extra fps in Crysis..
hmm once you've played crysis you move on you know? people want whats best , and bang for buck , heat output/ energy usage and features , like dx10.1?Quote:
Originally Posted by y2kbos
Lol wait for actual numbers before you actually start comparing these cards to cards of yesteryear
And not everyone has an 8800 yet - people with 6800's, 7900's etc. will certainly receive a huge boost
I found an Nvidia fanboy! Do i get a prize?Quote:
Originally Posted by y2kbos
What about the people who currently dont have a Graphics card and are building a new rig? The 4870 is going to be considerably cheaper than the GTX 260 and it will perform a hell of a lot better than the G92 8800gts (the next best bang for buck option)
Best?Quote:
Originally Posted by Papu
If you have an 8800gt/ 8800gts 640mb or similar, what is the point in getting a HD4870 when its only going to be as fast as an 8800gtx...
Energy usage/ heat will be similar to previous cards mid/high range cards you are upgrading from, dx 10 isnt even mainstream yet, let alone dx 10.1 ( nvidia's high end doesnt have it )
Well, to be honest, it does matter when shaders not only do vertex or shader processing....geometry processing...Quote:
Originally Posted by NH|Delph1
Anyway, R670 has 64 pipelines. Each pipe has 4 simple processors, and one complex processor. 320 total shaders, but only 64 really signifigant ones.
R770. IS it 160 pipes? I think not. TMU and ROP count do not match. 480 shaders? Again, we have the same problem. BOttlenecks galore in any current rumours out there...
I think we may all be in for a surprise. I'm gonna sit back and enjoy the show this time tho. It has been a LONG week!
You have a card.Post was meant for you, not anyone else..Quote:
Originally Posted by conzymaher
Doubtful..Quote:
hell of a lot better than the G92 8800gts
GTX280 yeah, HD4870 is too if you think 25% is " hell of a lot better "
Erm I don't have a card? I currently have no gaming PC at all.....
and secondly, you are just pulling benchmark figures out of your ass :D
We will see how well it performs soon....
RV 770 Pro naked :)
http://img140.imageshack.us/img140/8...ohd4850cs9.gif
RV 770 Pro/XT specs.
http://img140.imageshack.us/img140/3...cumentcvm7.jpg
Lets see tests first before making conclusions... if it scores the same as 8800GTS 512 I'll be happy only if it retails at a lower price range. If it's in the same price range it better be quicker if AMD/ATI wants to sell it to the masses...Quote:
Originally Posted by conzymaher
Sad but true, but Nvidia still got the advantange among the unknown masses... like Intel has with AMD...
Well you said you had a 8800gts on previous posting, thought you still had it..Quote:
Originally Posted by conzymaher
Performance will be near or above 9800gtx, if it was much better Ati would be pricing it higher..
Follow the money :up:
Genius, I never said a word about the slide text being aligned properly as the reason.. if anything I merely explained away the alignment issue FOR AMD.Quote:
Originally Posted by zerazax
people who are building new rigs? as was said not everyone jumped on teh OMG 8800 train.Quote:
Originally Posted by y2kbos
just because nvidia has it doesnt mean it isn't worthy? . they also haven't moved to 55nm process
heh :)Quote:
Originally Posted by y2kbos
But Ati need to get more market share, and Nvidia have nothing to compete below the GTX260 except for the G92b whenever that comes along, so the 4870 will either be alot better than the 9800GTX and reasonably priced OR not alot better and very cheap!
Either way its good for us :up:
And when I said it will be "a hell of a lot better" Anything even 20% better than the 9800GTX (that can already kick the crap out of any game) for similar money is awesome sauce \o/
hmm clean layout. :up:Quote:
Originally Posted by intel101
That's part right part wrong.Quote:
Originally Posted by annihilat0r
3d graphics do what is called a present call to display the image that has been rendered in the back buffer. If you have multi-gpu, often they don't present at regular intervals because you're not talking about 1 pipeline, you're talking about 2 pipelines trying to feign 1. So if they go:
10ms
Present
3ms
Present
10ms
Present
3ms
Present
Instead of the single GPU approach:
10ms
Present
10ms
Present
10ms
Present
10ms
Present
You *DO* get double the frames rendered and you're legitimately getting that higher framerate, but it doesn't *feel* that way because the presents are imperceivably close together. So while you do get double the framerate, sometimes (more often when framerates are low) you don't get the actual perception benefit (smoothness) of that higher framerate... it still chews through the frames but doesn't present them nicely for your eyes, so the benefit of getting more FPS is pretty much lost.
Keep in mind most of the time the few millisecond variation is no big deal (un-noticeable) but at lower framerates it can be.
Shouldnt be a problem than. When playing at insane res and AA's and the fps seem to drop, just add a 3rd card to get more fps again:ROTF::p:
Of course ATI says they won't have micro stuttering when their high end depends on the uptake of on Crossfire.Quote:
Originally Posted by mascaras
The microstuttering is from the present intervals of the driver, not the hardware.
GPUs get frames in batches, so 2 frames are sent simultaneously to 2 gpus to process, then they finish at almost the same time and have to decide when to present, rather than just presenting as soon as they have the frame as a single gpu does.
Dedicating transistor space to something like that that doesn't have many uses right now (no cuda on ATI, no physx or physics even) would be a huge waste of cash. There is no benefit to permanently dedicating some of your SPs to some task.. only bad things can come of that (waste for one).Quote:
Originally Posted by y2kbos
That document is fake or downright outdated... clocks are wrongQuote:
Originally Posted by intel101
I dont think the GPU's 'have to decide', I think it's pretty much like a frame is a transparant overlay or whatever. But 'deciding'... No.Quote:
Originally Posted by Sr7
No I didn't mean that.. I'm saying the driver decides.Quote:
Originally Posted by Rammsteiner
The frames are sent to the multi-gpus at the same time, so they do have to some intelligent spacing of the present call. Otherwise you'd end up with frames that take 10ms being sent at the same time, finished at the same time, and presented at the same time, and you still feel like you have 1 gpu since you won't see both of the frames. For scaling, you'd need the two gpus to present every 5ms instead of at the same time. And if they're sent at the same time, that tells you that something intelligent (driver) must be doing the offset on the presents.
Thanks for the explanation i've been waiting for. . .Quote:
Originally Posted by Sr7
Is this not hardware related as well as software related then? Or could drivers potentially sync this properly. I was also wondering - this makes sense if 2 gpu's are rendering alternate full frames but isn't it different when you get them to render different halves of the screen? (can't remember what it's called but i think you can set it that way)
cheers.
1200 Tfloops from Mr Andreas G. nordichardware it's true or fake?
haven't figured that out... am i missing something here???
1.2TFlops seems to be correct.Quote:
Originally Posted by MrHydes
I also think that 800SP is the final number and we still to understand the pipeline(shader) configuration . 5 way or 10 way VLIW, 32 TMU or 40 TMU,16 RBEs or 32 and so on .....
4850
4870
gtx260
gtx280
4870x2
eenymeeny miny mo, something here for everyone. im going to enjoy taking my pick. in about 1-2 month's time, if it ever happens.
I'd go for the 4870 for the price/performance if all current rumors happen to be real. But I well, I got better things to spend my money on at the moment, like better speakers, headset and a decent sound card. I guess I'll just wait for the next round in about a year time, RV870 and whatever NVIDIA has at that time.Quote:
Originally Posted by adamsleath
It's funny to see how suddenly everybody thinks the 800SPs are real :D
:D Yeah, it is. :yepp:
Well everybody may be a bit of an overestimation I think. I prefer to just hope for the best and prepare for the worst. To remain skeptical is another way to put it.Quote:
Originally Posted by STaRGaZeR
I just thought of something. XP can only support dual GPUs, so no quad CFX/ tri-SLI. But, if the 4870X2 is really not "crossfire on a card", and it uses a shared memory pool between the cards, then that means the card would be doing all the splitting of the load between the cores, and not the drivers. Which would presumably mean it would be recognized as a single card, correct? So, if you took two of them for crossfire, would it work in XP?
If windows does recognize the 4870x2 as having a single r700 gpu rather than rv770 and a bridge chip, then yes that could be possible, but otherwise, the 3870x2 dictates xp won't support more than 2 gpus
Only 2 GPUs would work.Quote:
Originally Posted by 003
DX9's hard limit is 3 frames (3 GPUs in AFR) BUT ATI locked the system down to 2 X2s for Vista CFX I think. 4870 + 4870X2 should work, but I'm not entirely sure.
And most of the time, Vista will have better ATI drivers (it's the trend for about 10 monthly releases already)
Yeah but if the 4870X2 identifies itself to the PC as a single GPU, then it would work, no? If it uses a shared memory pool, that means it is not crossfire on a card. Whether or not it still identifies as dual GPUs to the system is yet to be seen. I really hope it shows up as one because I do NOT want Vista. :down:Quote:
Originally Posted by Macadamia
If it used a shared memory pool that doesn't mean it's not crossfire on a card.. it just means that they didn't waste memory and therefore cut costs by not having to duplicate all memory contents across two sets.Quote:
Originally Posted by 003
It still doesn't mean that it's 1 big GPU. The determining factor for many GPUs to behave exactly as 1 big GPU would be if each GPU receives work for the same frame at the same time and outputs to the same place without the notion of SFR or AFR at play.
^
Super old I think.
And @003
Even when it shares RAM, DX9 limits the frames you can render (ahead?) up to 3. Even if 2 GPUs pretended to be 1, they'd still be rendering 2 frames with the chips they had.
or 2 gpu's render the same frame in same shared memory in 1/2 the time,may look to xp as 1 gpu and 1 frame?
That's called SFR(minus the shared memory part) and on current hardware performs significantly worse than AFR. Maybe shared memory could make SFR a lot faster?Quote:
Originally Posted by bill_d
what about super tiling ?
Resting in peace, gone forever.Quote:
Originally Posted by madcho
:cussing:
Any technology (SFR, Tiling), that splits the frames into smaller pieces will instantly kill geometry performance scaling across multiple GPUs.
So AFR here is just the lesser evil here, but it requires a lot of bandwidth and rapid response latency, to keep all the workload in proper time lock. And it's getting worse with the advance of the deferred 3D engines, with multiple render targets.
AFR create one picture before the CPU calculate it,
OR it create one picture after the cpu calculated it.
But considering a RV770 is likely enough powerful to handle geometry all nicely, wouldn't SFR work nicely?Quote:
Originally Posted by fellix_bg
The biggest problem now is probably just shader power, and MRTs.
Thats how RV770 will most likely look like:
http://img291.imageshack.us/img291/9...70800spuv9.png
Quickly Batman :eek:, to the Spreadsheet-Mobile !! :eek:Quote:
Originally Posted by w0mbat
:p:
Perkam
i seen only 200 shaders !!!
40 TMU?Quote:
Originally Posted by w0mbat
Hasn't it been pretty much confirmed that it will have 32 TMU? I suppose it is possible that those rumors were incorrect as well.
40 TMU + 800SP would be really awesome.
Then look again, its 160 5D in this Pic.Quote:
Originally Posted by madcho
lol your blind?Quote:
Originally Posted by madcho
its 5x4 clusters each with 8 sp -> 160 sp -> 5D -> 800...
... oops ^^
Another possibility: http://forum.beyond3d.com/showpost.p...postcount=2837
Why one 16 RBEs? Will that have any kind of impact on performance @ high resolutions coupled with AA?
RV 670 -> 320:16:16 ... RV 770 -> 800:40:16?
Doesn't seem balanced.
16 ROPs are completely sufficient.
meh im not sure about that, just look at 8800ultra vs 9800gtx -> in high res the ultra "stomps" the 9800gtx.Quote:
Originally Posted by w0mbat
That's different, the the 8 series cards ROPs perform AA functions and RV770 will use the shaders for that.Quote:
Originally Posted by Hornet331
The RBEs still render the pixel data... (If I recall correctly)
-EDIT-
LMAO, I read high resolution and was thinking AA for some reason. :confused:
Yeah BW limitation was G92s biggest problem @ higher resolutions...
well, yes and no, g92 did fine with 16 ROPs, but g80 beat it once you cranked up the max details and resolution, that's why the 8800ultra beats even the 9800gx2 in some cases with extreme detail etc
Very different, too many variables to make an estimation.Quote:
Originally Posted by Hornet331
Bandwidth, frame buffer, TMU&TFU and ROPs.
And that all has to do with the extra memory bandwidth the 8800 Ultra has.
even without AA 8800ultra beats 9800gtx in everything higher then 1600x1200Quote:
Originally Posted by LordEC911
9800gtx has higher shader clock, higher core clock, more TMUs same TAUs but lessbandwidth...
well even compared to the 8800gtx it loses sometimes in 1920x1200 and they have nearly the same bandwidth.
Im still concerned that 16 rops are not enough for the RV770 to do well in high resolutions.
I'd go as as to say that more than 16 ROPs is overkill for RV770.Quote:
Originally Posted by Hornet331
Quote:
Originally Posted by Hornet331
It's enough if they improved the Z and color fillrate capabilities.
RV670 has proven to be very flxible in Crossfire, reaching 2560 where the Ultra can't. It's pretty much obvious that if you want 2560 speeds, you have to pay $440 or $500 (4850 CF or 4870X2)
I have the same opinion, since RV670 do almost "nothing" in the ROPs and RV770 is supposed to be a derivative of it.Quote:
Originally Posted by Helmore
Pfff… you really don’t have a single clue what are ROP’s (or in ATI’s semantics - render backend) used for, and generally about anything slightest technical about CPU or GPU architecture?!Quote:
Originally Posted by STaRGaZeR
If they're still using shaders for AA then increasing RBE's won't matter that much (besides, they supposedly doubled the Z-rate)
And if you look at RV670, it seems to scale well with resolution, unlike the G92s which tank under higher res and settings
yes and but since it failed with detail, its still unclear whether or not 32 TMUs is enough
32 is enough. 24 is enough, possibly, too, if AA worked right, and not in shaders. r670 only works @ high res with no AA...even low res w/AA sucks.
No its about the lack of texture power, doesn't have much to do with AA not being done in the shaders. With double the texture power and double the Z rate, hopefully it should be enough for modern games with max detail. But something tells me if the 800 number is true, we'll be seeing 40 TMUs rather than 32 (same 4:1 ratio)
I would not be surprised if a future NVIDIA architecture did AA in the shaders as well. I'm basically expecting it to happen, it's more a matter of 'when' with NVIDIA. There is nothing bad about AA in the shaders, the ATI chips just lacked in Z-fillrate and lacked in texture filtering and other AF related things.Quote:
Originally Posted by cadaveca
Actually AF gave the biggest hit when turned on on ATI's chips, AA was actually pretty good for the amount of Z-fillrate as it actually scales better than NVIDIA's AA. (for the amount of Z-fillrate)
LoL. You mis-understand, really. TMU's are TEXTURE MEMORY UNITS...cache...Quote:
Originally Posted by AliG
The real problem with R670 is AA implementation. If they fix that, they can get away with 24 TMUs, although this is truly dependant on how many shaders there are...the number of shaders will dictate the number of TMUs.
Helmore, while I partially agree with you, it does not see mto make much sense that AF was a big problem. 16xAF @ 2560x1600 is not a show-stopper. AA WILL make an app unplayable, so while texture filtering and z-ops were a bit anemic, they weren't the real problem.
AA in shaders is essentailly the future, yes. When we get 4xMSAA standard...but that's still a couple of gens off yet...more likely to appear the same time as raytracing...
So ATI had a foray into it, didn't work well for them, but i think they understand the problems they are facing. Doesn't mean they have to beat the dead horse...
I just read out of boringness some wiki stuff about Ati, and it says 4870 X2 should be able to do Octa CF?:shocked:
Is that true?
First a question for you:Quote:
Originally Posted by Rammsteiner
How reliable is Wikipedia?
As reliable as its sources.Quote:
Originally Posted by Helmore
People usually won't add that for the hell of it, usually they read it on some site, mostly they don't bother adding source cites so it could be from anywhere from anandtech to fudzilla.
If one wonders, might as well just google it to find the original news post.
http://en.wikipedia.org/wiki/Radeon_R700
Editors at least follow wiki guidelines and post sources, just check them at the bottom. Mostly Fudzilla and Inq.
With the severely diminishing returns of going from 3 gpu crossfire to 4 gpu, can you really see them doing 8 :rolleyes: I guess that dude who rigged up 4 98 GX2's could do it.
:rofl:Quote:
Originally Posted by Nedjo
Nice try. And yes to the question. I'll talk seriously when I have the card in my hands.
And I'm still waiting your 3DMark06 3870CF 2,4GHz vs 3,0GHz numbers ;)
The octa CF is possible.
the CF of the R6xx can work with 2^n graphicards. ( the Trio of asus is not faster than a X2 ).
8 works 16 works, ... and go on !!!
Whatever you want to call it, they need a higher texture fill rate, put it that wayQuote:
Originally Posted by cadaveca
For what? Halo? Doom2?
Seriously tho, under PS2.0, ok. nVidia excels in this situation most often, but thier shader domain is twice as fast as ATI's, and hence the difference. TMUs aren't the issue...clockspeed is.
By your means, they would have to double TMUs to 32 to supply thier 320 shaders with the needed fillrate, and if the talk of 800 shaders is correct, then we'd be seeing 80TMU's, not the 40 you suspect.
Texturing is only important as much as you think in older apps, not in SM3.0...and both nV and AMD cards favor math-intensive processing for texture heavy-processing...and when it comes to lighting, RV670 excels, especially with phong.
Apps with TWIMTBP are coded to take advantage of nvidia's texturing power...they are old apps, after all. There's nothing wrong with rv670 and texture fetches...they just need to be twice as fast, not necessarily twice as many!
No that's not completely true, if that were the case, then ATI cards wouldn't die under any detail, especially AF. They do fine in pure resolution and that's why we know that its not a lack of shader power (though the clockspeed is why they aren't as powerful as nvidia's offerings in pure resolution)
And also, in order to double the texture fill rate, we're talking either double the core clock, which won't and will not ever happen, or doubling the count, therefore doubling the count wins
Um, TMU's do the texture processing? NO!!! The shaders do! So yes, that IS the case! More specifically, the inconnection speed between the TMUs and the shader core is much faster in nV's solution, hence the difference.
But when you add in math-intensive texture operations, like PHONG lighting, ATI excels, due to the pure shader count, and the efficiency of the dispatch processor. THis is why the increase of TMUs does not make a ton of sense, as you surmise, as there's more at play here than just the number of units, but also thier speed, and how they interconnect with the gpu, as well as thier size and precision!
what do you guys think the r700 will do in Open GL?!
vs. g92 I mean... still don't have a clue :p: