Printable View
TMUs help AF. ;)Quote:
Originally Posted by Gir92
AA is influenced by fillrate, but in ATI's case it's mostly just sample rate.
Not that it matters much though- 8x Box AA or 2x Narrow/Wide Tent AA are extremely good, and Edge Detect 4x can match 16xQ or even more.
Currently you need a better raw speed for those modes, and the 4ks are gonna bring that for a start.
Not really.Quote:
Originally Posted by Macadamia
It's the TFUs.
Faster/more TMUs can accelerate it in some cases ( lets say that the cards TF was bottlenecked by the TMUs for example ).
AMD ATI Radeon HD 4850 pictured :
http://www.bit-tech.net/news/2008/05...850-pictured/1
:up:
Looks like that early photo of the PCB with the small silver square was not fake after all.Quote:
Originally Posted by whocaresbg
haha yes so ironic we had it monhts ago :ROTF:Quote:
Originally Posted by STaRGaZeR
Shaders did AA for thee R600 architecture - but supposedly they doulbled the clocks for AA operations per second anywaysQuote:
Originally Posted by Gir92
It certainly looks like the 4850s are ready to roll
Beautiful card imo. Hopefully these are gonna perform at least like 8800GTS 512MB or slightly better I'd hope.Quote:
Originally Posted by whocaresbg
Quote:
Originally Posted by whocaresbg
Nice work finding it! Looks like a nice card. Í´m getting one for sure.
Looks so nice, Cant wait for 4870 and 4870X2 pics and benches, I am needing one badly.Quote:
Originally Posted by whocaresbg
Hmm, nice card? It looks like HD3850 ...Quote:
Originally Posted by fng77
That's exactly the point, it doesn't consume much power at all considering how much power is stored in it, either way based on how much you support nvidia I don't think your opinion will hold much water to anyone planning on buying an ati cardQuote:
Originally Posted by OBR
every socket 775 chip looks the same.... doesnt mean they all perform the same.Quote:
Originally Posted by OBR
I want some Zerotherm models please!!!
I want them now!
Quote:
Originally Posted by OBR
Thats exacly why I think it looks nice :)
The 4870 is going to look something like this (old news. I know =)Quote:
Originally Posted by ANP !!!
They had to write "FAST!" on the back of the card, does this imply a low performing card/failure?Quote:
Originally Posted by whocaresbg
I want one of these. Right now. What is the official release date now? At gpureview.com it still says 2008-05-31:shakes:
Just found an update here:
http://www.tomshardware.com/news/amd...deon,5514.html
The shipping to retailers has started according to toms :shocked: :up:
"We were told that AMD/ATI will decide on final pricing during the Computex tradeshow (which is scheduled to open on Monday) in order to have all the puzzle pieces for a June 18 in place. It certainly does not look like a paper launch at this time and the June 18 is an as-of-now date, since the company could revise is launch schedule to avoid a launch right on the heels of Nvidia’s new high-end parts"
yep
Quote:
AMD Starts Shipping ATI Radeon HD 4850 Video Cards
The ATI Radeon HD 4850 cards are reportedly already shipping to OEM partners and retailers. According to TG Daily everything is going as planned and AMD/ATI is aiming for a sizable launch of the new product generation, with Radeon 4850 512MB boards leading the charge. The final prices for all Radeon HD 4 series cards will be officially declared during the Computex 2008 tradeshow which will open doors on Monday. Higher-end Radeon 4870 cards with 512MB of onboard GDDR5 memory are expected to ship in volume sometime this summer, with flagship Radeon HD 4870 X2 to follow soon after that.
http://www.techpowerup.com/61735/AMD...deo_Cards.html
ugh 512MB ddr5 on the 4870
thought we were gonna get 1GB cards? :(
Yes - they are shipping later, GDDR5 takes its sweet time compared to GDDR3 which is already in place.Quote:
Originally Posted by Miwo
ATI does AA in the SPs, which have been increased to 480 on the HD4000 series for the 4870 variants, and they have also unlocked the shader speed from the core, so it is more like nvidia cards in that respect now, and hopefully, there will be a way to overclock the shader speeds, which will come at 1.05GHz, next to the 850MHz core.Quote:
Originally Posted by Gir92
The shader domain is still a question right now, it would be strange for AMD to do that. Also CJ and Wombat have stated that the clocks are not correct either.Quote:
Originally Posted by 003
We will eventually. AMD/ATi reference is 512mb cards but AIBs will be doing 1Gb models once GDDR5 is a bit more plentiful, possibly one of the reasons that they pushed R700 back.Quote:
Originally Posted by Miwo
I'm confused. Are you saying that the 850MHz core speed and 1050MHz shader shader speed are incorrect? And are you also implying that the shader speed may still be the same as and tied to the core speed?Quote:
Originally Posted by LordEC911
That would be a real shame. :down: :shakes:
I think he means that we don't really have any idea if they are separated or not, and that ATI hasn't settled on final clocks.Quote:
Originally Posted by 003
However, i think with Computex starting essentially in a couple days, the clocks will be close to final if not already settled there
If they're already shipping to Retail/OEM then AMD know the final clocks & they have been set. At least for the 4850.Quote:
Originally Posted by zerazax
Yep 4850's should be already set, especially since the box pictures and all that are out there
if it wernt for fast cars/moterbikes women and new gpu's life would be soooo dull...
it'll feel like christmas when these cards arrive...
not long now :)
Original graph from JC:
http://resources.vr-zone.com//newspi...erformance.jpg
After some modifications...
http://i114.photobucket.com/albums/n...GTX200GPUs.png
Catalyst 8.4:
· A picture is worth a thousand words -> WiC avg FPS.
Catalyst 8.5 :
· Call of Juarez DX10: Performance increases up to 12% on systems containing an ATI Radeon HD 3000
· World in Conflict DX10: Performance increases up to 25% on systems containing an ATI Radeon HD 3800.
· Higher performance gains are noticed on systems contianing an ATI Radeon 3870 X2.
Remember the part where they had 800SPs?Quote:
Originally Posted by 003
That's not entirely impossible either.
Overall, the 4870 Pro is 5% faster than the 9800GTX, right? And the 4870 XT is 25% faster than 9800GTX?
If yes, then I'd say it's a home-run for ATI (as they claimed to fix the AA issue).
that graph saying that the new nvidias are fatser than the ati's, was produced by nvidia, of course it's going to say that nvidia is a better card!
You say that like it's questionable/debatable :rolleyes:Quote:
Originally Posted by philbrown23
it's a gt280 vs a 3870x2, of course the gt is portrayed as a better card.
Everyone would do so
There is no such thing like 4870 Pro and 4870 XT... You have:Quote:
Originally Posted by Bo_Fox
HD4850 (RV770pro)
HD4870 (RV770 XT)
HD4870X2 (R700, 2 RV770XT(?) same PCB)
madcho what a load of rubbish those graphs are! what exactly are those random numbers on the side? is it trying to say that the 3870x2 gets 1fps in 3dm vantage? also trying to draw a conclusion based on hypothetical scores?
go vr-zone
percent, 3870x2 is always 1 (100%).Quote:
Originally Posted by Origin_Unknown
ever seen scientific graphs? they quite often have such a scale as well. you just divide the measured value by the one you want to compare it to.Quote:
Originally Posted by Origin_Unknown
in these graphs it's the values of the HD-3870X2.
why do they do this? quite easily, you can see the differences very well and you have a value without unit. for instance in 3D06 you got points, in most other cases you got FPS. this way you can put several bars in one graph...
wrong was, that NVIDIA began with 0.8, but thats marketing!
Latest rumors, but i don't know where it comes from.Quote:
Originally Posted by Bo_Fox
HD 4850 => 10 Prozent schneller als 9800 GTX
HD 4870 => 40 Prozent schneller als 9800 GTX
HD 4870 X2 => 30 Prozent schneller als 9800 GX2
From here. ;)Quote:
Originally Posted by ownage
Gut, das ist sehr schnell!Quote:
Originally Posted by ownage
Quote:
Originally Posted by ownage
Quote:
# HD 4850 => 10 percent faster than 9800 GTX
# HD 4870 => 40 percent faster than 9800 GTX
# HD 4870 X2 => 30 percent faster than 9800 GX2
http://translate.google.pt/translate...pt-PT&ie=UTF-8
http://www.hardware-infos.com/news.php?news=2110
Quote:
RV770 PR SAMPLING PROGRAM
http://img135.imageshack.us/img135/15/49077760ig9.jpg
http://www.ati-forum.de/allgemein/ne...-presentation/
Quote:
How we reported some days ago, the new Catalyst 8.6 contains the support of the RV770 and more further models of the new 4000 series. Today we can show with a save source you further models, under all the IDs of the 4400 series:
http://img228.imageshack.us/img228/2...eriesidcs4.jpg
http://www.ati-forum.de/general/news...e-4000-series/
:up:
HD4870 40% faster than 9800 GTX, yet HD4870 X2 only 30% faster than 9800 GX2?!Quote:
Originally Posted by ownage
I find this strange. That would mean HD4870 X2 scales extremly poorly.
I call this rumor BS.
What is the current ETA on these cards please?
sub'ed
HD4850 middle June. HD4870 July I guess due to GDDR5 shortage.Quote:
Originally Posted by Telo
seconed. :yepp:Quote:
Originally Posted by SimBy
It is called diminishing returns from the second GPU, and can be seen in early GX2s as well when drivers had not fully matured.Quote:
Originally Posted by Hornet331
THe GX2 has been out for a while, hence the lower result due to tweaks that may still be necessary on the 4870X2 to optimize the performance from both cores.
I am NOT stating that the source is correct, just that the above argument cannot be used as a basis for discrediting it.
Perkam
Unless of course HD4870X2 clock is lower than HD4870. But that would be a step back from HD3870X2 which actually had slightly higher clock than HD3870.Quote:
Originally Posted by perkam
Anyway drivers could be the reason too.
On the other hand I find the performance of HD4xxx kinda high. Actually much higher than what I was expecting. Thats also a reason why I call it BS.
Thanks much for the quick reply. I'll sub this thread and keep an eye out for the 4870x2 to replace my HD2900XTs. Thanks again.Quote:
Originally Posted by SimBy
http://www.fudzilla.com/index.php?op...7625&Itemid=34
Quote:
Pre-Computex 08: They will do it the right way
We've learned that RV770XT and Pro won't do Full Scene Anti Aliasing via Shaders. Developers hated it but RV670's design flaw was responsible for it.
RV770XT and PRO will end up significantly faster, especially with FSAA compared to RV670 generation. The numbers should be higher than you usually see from generation to generation, but we don’t know more details about it at press time.
RV670 was improved and fixed R600, or should we say R600 done right, whereas RV770 is an improvement of the stuff that ATI has implemented.
So then you can't get 24x edge detect AA, that's a shame.
I think fudo is full of :banana::banana::banana::banana: again
http://www.fudzilla.com/index.php?op...=7626&Itemid=1
"We have just confirmed that ATI HD4800 series powered by RV770 chips will support 7.1 surround sound audio with AC3 and DTS and the new cards will be able to drive HD video and 8 channel audio via single HDMI cable.
The RV770 based cards will use DVI-I to HDMI adaptor and the reference design of RV770XT and PRO includes two DVI outs. At same time RV770 / HD 4800 generation won’t have any issues supporting DisplayPort but this is a matter of a partner choice as DisplayPort is still not that important or accepted.
The possibility is there and display manufactures do plan to embrace this new connector.
The new RV770 chip also supports xyYCC color:end quote
New Logos:
http://www.fudzilla.com/index.php?op...=7627&Itemid=1
The new logos are better than the old ones. I don't like having to use DVI adapters since I can't get 75 hz on my monitor if I do that...Quote:
Originally Posted by fng77
Haha. OK. AA is not done in the shaders. Then where? In the ROPs? In its 16 ROPs? :rolleyes:Quote:
Originally Posted by informal
Same here. would be better with a separate (real) HDMI port.Quote:
Originally Posted by Nuker_
What a twist!! :eek:Quote:
Originally Posted by STaRGaZeR
We shall soon see what the real deal is...
Um... yeah. Where else do you think it would be done?Quote:
Originally Posted by STaRGaZeR
ROPs are not a bottleneck for RV670, and 16 ROPS @ 850MHz should be enough from what I see. It would give RV770XT only 18% less pixel power than the GTX 260. Even though it has 28 ROPS they are only clocked at 575MHz.
If only I could step-up to HD 4870 with the eVGA Step-up program. :p: HD 4870 is sounding better and better everyday, and at $300 versus $450 for the GTX 260.
4870 if really 40% faster tahn 9800GTX will a beast for the price
go 4870 \o/
its my #1 choice for my new build atm, Unless the GTX 260 is priced very well, or a lot lot better
hey everyone, by what percent is a 9800gtx faster than a 3870? and can u use that number to predict benches?
well, no, u cant, cause there is no really fixed percent it is faster than it. so u guys are reading into the percent thing a little too much.
But the fact remains that HD4870 > 9800, RV770 > GT200 price-wise, and R700 > GT200 performance-wise, and probably price-wise aswell. :DQuote:
Originally Posted by grimREEFER
Is that a fact, all of a sudden?Quote:
Originally Posted by Calmatory
I mean, everyone is jumping the gun over a bunch of wildly differing powerpoint slides..
At least wait a couple of days for Computex, where we'll see both card series in action.. :)
So any word on the relsase date?
Give it two more weeks and a few days.
AA not in the shaders may sound good if there were any proper location to do it instead. I just can't see it. Wait and see...
I doubt they would do it unless it was that big of an issue - increased shader count + speed + supposed double clocks/sec of ROPS would be more than enough. Besides, are they really going to remove 24x aa and edge detect?
Naw, was just a spiky comment. But then again, it seems possible. At least I don't mind if X beats Y by +/- 5 % difference, some of it goes to the margin of error, alot of it goes to the drivers, and in some cases there are just too many factors which count to be able to give accurate readings of the performance. Well, I say that pricing(and PR!) is something which is going to declare the winner, IMO.Quote:
Originally Posted by bowman
Quote:
Originally Posted by zerazax
maybe they do the AA up to 8x in RBEs and the shaders allow to go up to 32x with somes filters.
& maybe it's always shaders and fudzilla lie :rolleyes:
Anyway Narrow-tent, Wide-tent and Edge-detect all produce blurry images for me. So nothing to worry about if they are disabled.
i hate this i hit toronto June 18th - 30th, and i doubt either side will have cards on shelf , really getting tired of this 7600Gt!
I'm thinking Fud and/or his source misunderstood something.
Why do I think they caught wind of the z being upped to 4x and they assume that AA is back on the ROPs.
Box MSAA on ROPs, tent/edge detect on shaders I guess.
But it doesn't matter I guess, considering both ways will be samely fast.
And it's getting very ridiculous that the specs haven't leaked out a bit. :down:
Well, 32 TMU's on a single core is the first time ATI has ever done that. Everything from ATI up till now were no more than 16 TMU's. Nvidia already upped it to 24 with their 7800 cards. X1800, X1900XTX, and even the HD 2900 and 3800 cards only had 16. Nvidia's 8800GTX upped it again to 32 (64 texture filtering processors).
I do not think the ROP would be too limiting, as evidenced by the 9800GTX cards. Thanks to excellent overclockability, those cards can equal the 8800GTX cards with 32 ROP's in nearly every game with FSAA except at 2560x1600 resolution which is hardly used by anyone. Remember, the 9800GTX's are also severely limited by 256-bit memory bandwidth (and also only 512MB buffer).
We know for sure that the RV770XT will have nearly twice the bandwidth with GDDR5 (and it looks like there will be a 1GB version widely available). I think ATI just wanted to keep the die size and power consumption as small as possible and the first thing to ignore was the ROP's. The X2 version will have a total of 32 ROP's anyways, since that is what ATI is really betting its horses on.
The question is: Will the R700 (4870 X2) actually beat the GTX 280 in overall games at 1920x1200 with 4xAA?
I've heard some say that it will no longer be Crossfire scaling but a "hardware infusion" of both chips so that it can act like a single core processor altogether. But now I see from the heatsink design that it will still be two separate chips a few inches apart from each other, so it is probably still going to be nothing more than what the Crossfire drivers can do.
I imagine that hardware infused might also mean that the cores can share all the RAM between each other, who knows? :)
Found this and it doesn't look like it was posted before.Quote:
Originally Posted by DeathReborn
+900mhz on the HD4870 is pretty much set now.Quote:
Originally Posted by Guru3d.com
That 1050-1075 core clock definitely seems to be doable...
More news, or rumors I might say. Not much new:
http://www.overclock3d.net/news.php?...mour_roundup/1
http://my.ocworkbench.com/bbs/showth...916#post431916
:rofl::rofl::ROTF::rofl::ROTF::rofl::rofl: :clap: :clap:
And by the way, I'm an ATI fanboy...:wth:
So? ATI HD4870 scored higher than a 8800Ultra and almost doubled HD3870 score, what's wrong about it?Quote:
Originally Posted by mcbalaban
He seems to be confused,performance is just fine.And also Vintage is useless BS marketing benchmark.We need performance figures in real world gaming.Quote:
Originally Posted by Luka_Aveiro
That is old, and proven fake.Quote:
Originally Posted by fng77
Vantage numbers for either cards are about as useful as a library in Mariah Carey's house.
We have no idea how to translate 3d07 (vantage) scores into 3d06 and 3d05 numbers because there is far too little consistency across the different cards. 3D 06 numbers will STILL be the ones that matter this time around, as it has gone through a few revs. FM is already talking about a new revision coming out so 3d 07 numbers will be useless till those come out.
Perkam
I am still wondering how much memory bandwidth will come into play. Whether or not the gains of the 4870 over the 4850 are a result of the DDR5, or simply just the higher Core/Shader clocks of the 4870, or a combination of both. If its anything like the 3850 vs 3870, most of the speed improvements come from the increase in core speeds. If the same pattern follows, it may be much more economical from a Cost/Perf perspective to buy a 4850 (hopefully AIBs will add 1GB DDR3 variants), and voltmod the snot out of it.
:lol2:Quote:
Originally Posted by perkam
Well... I think CJ has said that the ATI parts are made up, he spread the original GTX scores though.Quote:
Originally Posted by perkam
Flame me for trying, but I think one segment where the 4870 will be VERY competitive with the GTX 260 is 3DMark Vantage, while in other titles it's back down to price.
Vantage is very shader-heavy. It does strain on texture a little, but not to the extent that it does to the shaders. All in all, High/Xtreme is quite a good representation of what DX10 future titles will be :cool:
Nearly every single shader heavy game or synthetic benchmark still runs faster on G80 including Vantage.
Higher theoretical Flops + lower efficiency != Faster
Fillrate, Z and tex (it still strains the units, just not as much to give G94 a lead). I think RV770 addressed that.Quote:
Originally Posted by trinibwoy
There's no competition that G80+'s shaders are more efficient, but R600+'s limited components gave its shaders too much of a bad name.
Yea, sure guys... I wasn't lying when I said I'm an ATI fanboy, I just got so used to ATI being over-hyped that auto-sarcasm has gotten the best of me. :yawn:
In any case, I'll be much nicer being pleasantly surprised if the 4k series performs 20% better than that chart (or downright religiously ecstatic for any number higher than that), than it will be when it performs only 5% better than that.
Think about it! What if all ATI fans chaged their attitude to mine?...
Then we could have threads like this with folks saying stuff like "Man, I can't wait to buy the next ATI card... We'll get owned by nVidia so badly that the cards wil reach S3 prices within a week!"
And that, my friends, would leave the nVidia camp with nothing troll about!
Catch my drift? ;)
More news(from questionable source i must admit :) ):
http://xtreview.com/addcomment-id-53...rice-drop.html
Quote:
It was considered that the video card radeon HD 4870 and Radeon HD 4850 will be proposed for $349 and $249 moreover in the first case buyer will obtain video card with 1Gb GDDR5 memory . According to extract from official AMD document , dated 26 May, the company notified its partners about reduction in the recommended prices of video card HD 48XX .
In particular, Radeon HD 4870 with 512mb GDDR5 memory will be proposed for $329, and Radeon HD 4850 with 512mb GDDR3 memory will be proposed for $229. Thus, AMD reduced the initial prices by $20.
Associates also report that Radeon HD 4870 will be approximately up to 40% faster than GeForce 9800 GTX, and Radeon HD 4850 will prove to be 10% faster than the same Nvidia video card .
Yes RV770 will benefit if those areas are improved. However, my point was that even if you isolate the shaders R6xx's performance falls short of its potential. So RV770 will certainly benefit from shader improvements as well.Quote:
Originally Posted by Macadamia
Check the perlin noise test in Vantage - it's pretty much a pure math test and the architectures are in a dead heat.
http://www.digit-life.com/articles3/video/vantage2.html
Also, take a look at the geometry shader heavy particle simulation test. This is supposedly a strong suit of the R6xx architecture yet it falls way behind. Just pointing out that you shouldn't get too caught up in marketing or theoretical numbers.....
Having said that, I fully expect HD4870 to be a more compelling option that HD3870 was. The pricing alone belies that possibility.
That can never happen with the state amd is in right now, if the 4870x2 outperforms the gtx 280, we'd definitely know it with a $150+ boost in price. However if amd had plenty of money, then yes they may want to do what they did with th athlons againQuote:
Originally Posted by Calmatory
http://www.hexus.net/content/item.php?item=13600
WTF Radeon 3950 as well? Revised RV670 or cut-down RV770 cores?
Interesting to see the SP counts are hidden for an AIB partner while the Nvidia cards are already out and about - either ATI is really confused or unsure or they're keeping a tight lip on the SP counts and have something up their sleeves
probably that's the rv670 a12 revision
If I remember the RV670 loved vertex, was neutral on pixels and hated geometry (shaders).Quote:
Originally Posted by trinibwoy
More like tasks the compiler can split up properly :rofl: but still, GS isn't widely used in current DX10 titles. The cloth simulation test is an eyedropper though.
I wonder if the cache for synchronization in every (cluster of MPs?) ALU of the G80+ helped here, considering RV670 has to reach the registers instead of the sideway cache.
Yeah it has a considerable triangle setup advantage but since games are mostly polygon shy this doesn't help.Quote:
Originally Posted by Macadamia
The parallel data cache is used by CUDA and geometry shaders AFAIK. However, it is no faster than accessing the register file (this is the fastest memory on a chip and is closest to the ALUs).Quote:
I wonder if the cache for synchronization in every (cluster of MPs?) ALU of the G80+ helped here, considering RV670 has to reach the registers instead of the sideway cache.
CJ at B3D posted this about the R700 Vantage scores:
Wow, that would be way more than the GTX280 score leaked by him..Quote:
But for now I'll leave you with this... what does a HD3870X2 score in Vantage on the Extreme Preset? Some reviews I found on the net say a HD3870X2 scores about X2600 with Cat 8.3... Well according to a source of mine who I trust and who has proven to be correct most of the times before...
you can double that... and some more for the R700 Vantage X-score...
Yeah, wasn't ATI going to be releasing a revised HD3000 lineup along with the HD4000 series, that is supposed to be able to clock higher than the current ones?Quote:
Originally Posted by AliG
Yeah the rev A12 cards. Fud claimed it was canned, but then again Fud knows nothing about ATI hardware it seems.Quote:
Originally Posted by 003
If true though, that should give a good indicator of where the 4850 and 4870's slot. If a 39xx series is faster than the 38xx series, it should place it close to the 8800GT and so 4850 would be a bit faster, probably close to the 9800GTX as hinted by rumors.
Damn that would put it somewhere around the X5600 range!! :hitself:Quote:
Originally Posted by zerazax
If that is really true, I am going to be buying dual 4870X2s for quad CFX.