No, he meant just the opposite.
He makes a good point though. We are at a great point in technology where mainstream video cards offer great performance at most common resolutions.
EDIT: Double post :p
Printable View
consider a year ago that midranged cards at the $200 release msrp were the 8600GTS, 2600XT, and X1950Pro, we've come a LONG way and our cards are going to last a lot longer
Longetivity... Good point! I had X1950Pro for about month (200euros). Then 320MB 8800GTS lasted about two months (319 euros). Then two 2900CF (400euros) have kept me happy for about 6 months.
I'm just saying, back then at that price point, those cards were terrible compared to the high end offerings whereas today, those price points are giving us cards very close to the elite high end
Do you people think the 1GB version will beat the GTX 280?
Will the 512MB version handle 2560x1600 OK with 8x AF? What about 16x AF, or 8x AF/4x AA? Finally, what about all of that at 1920x1200 instead? I am thinking of getting a new monitor but I don't want to bother if it would mean my 4870 512MB GDDR5 wouldn't perform well on it... I ordered one not wanting to risk price gouging and a long wait on a 1GB version.
So I thought Nvidia said they were going to open up a can of whoop ass?
What happened there?
:rofl::ROTF:
So it won't make a HUGE OMGWTFBBQ difference you don't think? I'm thinking of grabbing a 30" or 24" LCD to upgrade my current S-IPS 20" and have a 4870 512MB GDDR5 on order. So, I'm concerned if I buy a big LCD right now, which I can afford, I might get stuck with a bad setup as I wait for the 1GB model.
been debating weather to put up with the long wait for the 1gb version or even an x2....
i think the performance from two 4870's in crossfire @1680x1050 will play all the latest games with all eye candy max'd....
thats going to suit me fine, anyone want to buy two 3870's.... lol
i've decided i'm going to drive over to ccl later and pick a couple up....
Yeah, they should be fine at 1680x1050 purecain :).
I'm just wondering about 1920x1200/2560x1600 myself, though I think they'll be OK there too *shrug*.
holy shmokes! the card flyes! it seems to be handling 8x Anti-Aliasing VERY well!!!! :up:
Thank you for this review cookerjc. I especially appreciate including Quake Wars benches.
For one thing quake wars is currently my favorite game. But also it is one of the most stressful OpenGL games out. I seem to remember that in the past ATI has had relatively weak opengl performance even when they have a relatively strong GPU. But now the 4870 comes within 1 FPS of the GTX280 at my native resolution (1920x1200) @ 8xAA/16xAF. I wonder if it is because of driver improvements or simply the raw shading power available + AA improvements in the R770? Or perhaps driver issues with the GTX280 in QW?
Either way I have almost definitely found my next card. I don't think I have run an ATI card regularly since the first generation AIW. I sure hope their drivers are better then they were back then...
Who said i didnt read the review???
Yeah 92C... how much can the card take? like 140C?
If the card works... and you are someone who doensnt care about cooling.
What is the mather? Because when the temperature rises more the cooler kicks up some more. Just like Nvidia... only Ati set the treshhold higher.
I agree 92C is quite hot, but than again... its the stock cooler :shrug:
If you want silence AND performance you'd better but another cooler for it. You are right at this point, infact personally i also would buy another cooler because when i get the 4870 i am going to OC the crap out of it :p:
But when you look from the perspective of the normal custumer... it isnt really a problem :)
high temps are due to poor fan profiles and probably poor heatsink mounting (4850's get a lot cooler by remounting).
True, i saw an 5C improvement by remounting(and tighten the screws some more. because stock the screws werent at its tightest)when i ran a stresstest after 10 minutes remounting My HD3870 was 5C cooler. In LOAD The weeks after that it went up to 7C cooler :)
In idle it was only like 2C to 3C
Still a good difference :)
my apologies for the thread revival..
nice to see R7xx live up to the code name
I just read Anand review. I must admit that I'm impressed with ATI. As for future rampage user, 4870 will be a good way to begin :)
More than 100% CF scaling? :p:
http://images.anandtech.com/graphs/a...5208/17137.png
Posted by DerekWilson in comments:
Quote:
greater than 100% scaling is due to margin of error combination for both single card and dual card tests in the vast majority of cases.
we also tested single card performance on an nvidia system and crossfire performance on an intel system, so the different computers will also add margin of error.
two card solutions generally don't scale at greater than 100% except in extraordinarily odd situations (where rebalancing loads might help with scaling on both individual cards -- but that's odd and rare).
Same error for both 4850 and 4870 and with 5%+ difference. More their testing method or drivers thats bugged.
And 2 cards will never scale more than 100%, never. Well with the exception of nVidia I guess if 1 card is also doing physics?
But besides that, 4850/9800GTX(+) looks like the cards to get. Or 4870 if higher, but its still hard to pay 50$ more for so little. And SLI/CF shows its uselessness as usual.
Its also another round on how little improvement we get. 4850 is basicly 3870 and with same wattage. Same story with nVidia more or less. And their bigger cards scoring 20% more or so. Ultimo 2006 was the year of the big wall :p:
>100% scaling could be that the 512MB with a single card isn't enough and the CF setup does act like 1GB instead of mirrored 2*512MB.
Margin of error sounds more plausible, though.
http://img295.imageshack.us/img295/9718/52427626ad1.jpg
http://www.anandtech.com/video/showdoc.aspx?i=3341&p=9Quote:
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.
One thing is for sure, anything AMD can do to assist in providing more reliable consistent scaling with CrossFire will go a long way to help them move past some of the road blocks they currently have with respect to competing in the high end space. We're excited to see if this really makes a difference, as currently CrossFire is performed the same way it always has been: by combining the output of the rendered framebuffer of two cards. Adding some sort of real GPU-to-GPU communication might help sort out some of their issues.
Stop those astrologic preditions. Lets wait and see. You also said 100 times that HD 4800 suck, but sorry they don´t, actualy is the opose.
So before you say 100 times that R700 is like R680 wait those astrological power and lets see what R700 bring to the table.
There are new things that may come to be very interesting ;)
First of all, this aint X2 card is it? No thats right, its 2 card CF. No magic there.
I never said 4800 series as such sucked. Its just not the divine reborn of chirst that you and a few others think. There is a reason its 199 and 299$ and not 299 and 499$. Unless you think AMD think its funny to lose money. nVidia instantly countered the 4850 card as expected in price. Since most of us non fanboys knows nVidia milked us for their record after record profits.
4850 is basicly=3870 and 4870 is some 20% faster. Its not exactly spectacular performance wise. But luckly enough, GT200 aint either.
And I sure hope it has a corssfire sideport. What do you think goes through the ribbon? I´m sure you can find a hype that can go crazy enough until its utter failure before the R700 release. Plus it might only fix some of the issues. As always the R700 will jump from excelent performance and down to 4870 performance or even below.
Before you guys start talking about errors, shouldn't we see results of one card, then two in CF on the same platform? Plus for many of us, I'd wish they'd do nVidia's Intel systems as an after thought LOL!Quote:
we also tested single card performance on an nvidia system and crossfire performance on an intel system, so the different computers will also add margin of error.
It´s a diferent crossfire because R700 will use a new feature that 2x RV770 in one mainboard can not use.
:rolleyes::rolleyes::rolleyes:Quote:
I never said 4800 series as such sucked. Its just not the divine reborn of chirst that you and a few others think.
You are the first trashing HD4800 everywhere. And if HD 4800 is not the reborn of chirst it is very close for the reactions we have seen :D
The reason is writen everywere. First AMD target more volume (200-300$ market), and second HD 4800 cards are less expensive to produce. It´s writen everywere I don´t need to guess.Quote:
There is a reason its 199 and 299$ and not 299 and 499$. Unless you think AMD think its funny to lose money. nVidia instantly countered the 4850 card as expected in price. Since most of us non fanboys knows nVidia milked us.
And AMD want to get back lost market share in the next 2 quarters and that´s the way to go -> strong performance and less expensive then competiton.
I am not sure of nothing. I simply wait knowing that is there something new in R700. I just don´t start trashing the card.Quote:
And I sure hope it has a corssfire sideport. What do you think goes through the ribbon?
No, that's not true - the whole point of redesigning crossfire for the 4xxx series was to share the total pool of vram among the cores (think of it as using each card as a stick of dram, it's all the same pool to the processor).
Keep in mind this was a design goal, we won't know if they succeeded until the both the 4850 and 4870's have been put through the paces for a couple of weeks.
Your knowledge or ability to keep up with the tests is not exactly spectacular either, but that does not suit the purpose of threadcrapping on each and every thread about an AMD/ Ati product, does it? The 4850 is more than 50% faster than the 3870 in quite a few test. Read up or shut up.
http://anandtech.com/video/showdoc.aspx?i=3341&p=15
The shared memory makes it not quite the same as having two different PCBs by default, right? 2 x 512 MB = 1024 MB with the shared memory, but with SLI and CF, 2 x 512 MB = 512 MB. I am not saying it will yield better performance, time will tell about that, just that there is a significant difference. VRAM is of concern when going to XHD resolution and maxed settings, and 512 MB isn't always enough.
But concerning Shintai's previous statements that CF/SLI is a gimmick because it can't scale well, or that "magic" >100% scaling isn't possible.
Here are some screenshots with and without SLI...
SLI ON:
http://img374.imageshack.us/my.php?image=me1iw8.jpg
SLI OFF:
http://img108.imageshack.us/my.php?image=me2mn7.jpg
20 vs 9 fps... Yeah, both are not playable but it was an easy and extreme example to do in those few minutes since my last post.
I can do more reasonable and playable ones...
I'm saying 4850/70 Crossfire is absolutely no different from 3850/70, nothing about perfomance. You still need to mirror the data in both cards. 4870X2 is the only card that has something new with its shared memory pool. And if you do Crossfire with two 4870X2, internally you have 1GB (2x512), but you will have to mirror the same 1GB of data in both 4870X2's.
Check the 9600GT SLI reviews, I saw the same 100%+ SLI Scaling in some games. It's more of an application thing than a driver or bug thing.
Perkam
You realize that you just contradicted every single Radeon 4850/70 review known to living human? :p:Quote:
Originally Posted by Shintai
You're getting desperate lately...
Lets see,AMD managed to get to 88% of gaming performance of GTX280 which is 570mm2,1.4B transistors,0.93TF monster chip with a lil RV770 that packs in under 260mm2,has 960bil. trans. and 1.2TF.
Talk about failure some more,you entertain us lately,very funny dude :p:
some of u guys sound just like the guys 10 years ago about voodoo5.. this multi gpu/card discussion never ends..
1x512 vs 2x512 or 4x512 is not all the same.. yes 512 max @ all how can you have more than 512??
but for goodness sake.. the same 512 only has to render 50% @ dual 25% @ quad
if this wouldnt be the case than whats the point of multi gpu/card?
OMG "luck" lol... Give credit where credit is due. You mention objectivity but you're no different from any other fanboys. You're just downplaying this because, I dunno, you're a freakin' amd-hater ("wtf amd/ati did something right that benefits consumers.. oh noes! it's just the grfx division anyway, they still half-suck").:shakes:
But seriously, mid-range products that can catch up to high range ones and you act as if that's "meh". Nv did it with 8800gt and now amd brought something similar, if not better. Oh and 4850 @stock can match up to 9800gtx+ for less money. Amd/ati doesn't have the fastest card but they offer the best value right now, unless you're a fanboy.. yes you.
OK I will put it so NapalmV5 understands:
Crossfire of 3850/70 512MB --> 512MB available.
Crossfire of 4850/70 512MB --> 512MB available.
4870X2 (512x2, shared) --> 1024MB available.
Crossfire of 4870X2 --> 1024MB available.
I'm not talking about what GPU renders what. Only about memory size and availability. Got it already? ;)
Well then I think you will be seriously disappointed...
Although I do hope that you are right, there is absolutely no evidence that R700 will have shared memory, none whatsoever. Some are making that "CrossFireX Sideport" into some kind of alien technology, but this sideport is just the part that connects two cards together (through the ribbon cable on normal CFX). There probably are some more tricks in R700 to boost scaling but I don't think there will be shared memory. But, just like you, I'm also hoping it will be there....
the argument goes out the window @ multi gpu/card no matter how you look at/put it
some of you kids here were in your mothers womb while many of us were running multi core/card a decade ago.. and youre lecturing others on how it all works.. doing great so far..
but if you guys want to believe that 512 is the same @ multi by all means knock yourself out
LOL, OK, can you really define mid-range?
-4850 can catch up to GTX260 in some cases, even sneak up to beat GTX280 in minimum frames in CoH (ars technica's review). That's 200usd vs. 650usd, that was unimaginable.
-4870, for less than half the price of a GTX280 doesn't trail too far behind. If some cards are 50 bucks, and the 4870 is 300, and the 280 is 650 bucks, the 4870 is right there at the middle. But hey, if you disagree with this logic let's go back to 4850.
The point is, both offer great value right now.
Thinking about ATI Fury MAXX (AFR) and/or 3DFX Voodoo solutions (SLI - Scan Line Interleave, not Scalable Link Interface)?
Because if you are thinking about MGPU solutions dating before these, then please do state them and your experience :)
Double the memory effectively in total, using MGPU is not very hard to do in effect, but the performance issues are quite obvious, as MGPU has been implemented to this date (for desktops solutions and games at least).
MCPU / SMT is very old in comparison, and may even predate yourself and you being in a womb (I have not checked your age, hence the "may").
Nah of course, the shared memory I'm talking about is just to explain my point, there is no memory sharing in 4850/70 CF. All regarding R700 is pure speculation. Let's hope they can do it. But the main problem about Crossfire is not scaling, the problem is that you have graphic glitches, less perfomance than single card in some cases, crap min FPS, etc.
Seriously dude, if you need to start talking about how old are you... you should probably stop right there because you don't know how old are I, maybe you'll be surprised. Another thing, I'm having a hard time trying to figure WTF are you talking about with your percentages and 512 of memory. What you say makes no sense at all. I guess this is another win for you (and the last one, welcome to the ignore list), because I don't want to talk with someone with his mind 15 years back in the past. The only thing you do well is to call everybody wrong (which is wrong itself), but you don't tell us the right thing. Way to go tech granpa http://img299.imageshack.us/img299/6350/oldrg8.gif :ROTF:
Thanks to dinos for the icon :D
I'm considering selling my 9800GX2 and going 4870 CF.
Or wait for 4870x2, then CF that.
But would 4870 CF be a good idea to out-perform the 9800GX2?
You're thinking about ancient split-screen sli i guess.
In modern crossfire with 2 cards, AFAIK both 512 MB must hold the exact same library of textures. Each gpu builds a sequential frame (gpu 1 frame 1, gpu 2 frame 2, gpu 1 frame 3 etc) having direct access to its own 512 MB (which holds 100% the same as the other).
But i'm no expert. :shrug:
Yep that is why I think it is called Alternate Frame Rendering,(AFR) basically card 1 renders Frame 1, Card 2 renders frame 2 then Card 1 renders Frame 3 card 2 renders frame 4..and so on...
That is why we get microstutter, ocasionaly the synching between frames 3 and 4 are a ms out of sync.
John
Scan line was horrible too. The reason that doesnt work anymore is due to shader processing etc. If we returned to scan like you would basicly need 1600SPs on each card instead of 800. The world was just so much simpler back then.
I think they only way AFR could become micro-stutter free was if a timing system were used in the proccess, ie vsync or something to limit the frame rate to a constant value so that the gpu knows to render xx frames per second. That or some clever person needs to invent a new system all togeather :P
it's entirely possible that ATI did that, however I think like was stated before the bandwith increase between the cards get rid of the 3870x2's issues will help alot.
Swore ATI was takeing a new approach with the R700/4870X2 that I believe even if they don't get rid of it entirely will atleast make it alot less frequent and noticable.
Bandwidth and latency is 2 different things. You could do the sync with no shuttering on a 28.8K modem..it would just need to have a low enough latency.
People tend to wish alot for magical things these days. First mini cores, then shared memory and so on. Its like watching a desperate race to keep the hype going.
I miss Pong and Tetris.
:cool:
http://www.techpowerup.com/64301/Dia...k_Edition.html
oh, overclocked edition ;)
Well everyone want's innovation and technical advancement if they didn't we'd never get any new or better tech.
Thats funny... :shrug:
I bought a factory OC'd 8800GTS (640mb) the day they were available and have played games unfettered for just $390US. Vanguard and Crysis were the only two games that I had to tweak the settings on. Comming up on almost 2 years for this card and I will be moving to the HD4870 or HD4870x2.
My other rigs all have a Radeon 850XTs. Which will be rebuilt with the new HD4850s.
Have no idea why you just didn't buy an 8800gtx, you'd saved yourself a lot of time and money. Not quite sure about your "longevity" comment...!
.
can anyone tell me why gtx280 got less flops then 4850 if gtx280 got 1.5bil transistors?
Take a look at many of the indepth architecture writeups found in the 4850 reviews
An easy less complicated answer is that the 4850 will hit 1 Teraflop under best case scenario. Depending on the work done, it may or may not come close to that number....but for advertising/marketing purposes its a 1Teraflop card because its capable of so.
*Dreaming HD 4870 with GDDR5 1GB* :D
Quite a pity that people think that they need 1GB. It's like people think that 8GHz Pentium 4 is 100% faster than 4GHz Core2Duo. :p:
Well, ignorance has it's downsides..
It's very easy to exceed the milestone of >512MB VRAM usage in games these days... Even old games like STALKER - SoC takes 582.65MB (I tested it just for this). That's immediately after launching the game and loading a save. I hear Oblivion eats way, way more with high res textures. So I really don't know what you're ranting about.
:stick::confused:Quote:
It's like people think that 8GHz Pentium 4 is 100% faster than 4GHz Core2Duo.
Err... Umm...
What?
Naturally, the amount of available VRAM does not affect FPS. It eliminates video memory swapping which manifests as stuttering.
twwen2,
Indeed.
This is a subject that is hardly ever discussed, but i did notice stuttering when i had my 256mb 7950GT which i thought should have played everything great at 16*10. Low VRAM was probably why it wasn't so smooth :shrug:
Just check out the 8800GT 256mb vs 512mb vs 1GB review.. CookerJC did it iirc.
256mb card sucks
512mb card works great
1024mb card kicks the 512mb card to the curb when large texture amounts come into play.
Every card has to load all the assets to its memory. Even if it was still using scanline, it'd still be true with todays shaders. You need to have the info for the whole scene for shaders and shadows to work properly :)
still I think the cards alternate in rendering whole frames these days :)
yes; it will be interesting to see.
hopefully 5 million more shaders plus some decent bandwidth will be enuf for ati to do its rap dancing with 512MB vram :eh: y waste $ on ram you dont need :hehe:
but i guess someone will find the limits soon enuf.
lots of talk, no hard proof.
@ http://www.tgdaily.com/content/view/38194/135/Quote:
ATI readies OEM special: SuperRV770 to challenge GeForce GTX 260, 280
The ATI Radeon 4870 ships with two six-pin power connectors, which support a maximum thermal design power (TDP) 225 watts (75 watts +75 watts and an additional 75 watts from the motherboard), while the board will never consume more than 160 – 170 watts at stock speed, we are told. That means that there is at least 55 watts of room for overclocking and enough space to find out what these GPUs are capable of.
AMD GPG (ex-ATI) is binning the parts to a lowest denominator required for good yields and a level of performance that reaches or sometimes overtakes Nvidia’s GTX 260. But this time around, the company developed an AIB/OEM-only product codenamed "Super RV770", which will be much more powerful.
The "Super RV770" will arrive with pre-installed water-cooling and features unlocked BIOS, which enables the GPU to be pushed all the way to 950 MHz, while the memory can be pushed to 4.8 GT/s (1.2 GHz QDR). According to our sources, you may be able to push the GPU even beyond 1 GHz, if you use TEC elements, and keep the temperature of GPU low.
At 4.8 GT/s, the "Super RV770" can achieve a bandwidth of 150 GB/s, taking the crown from the GeForce GTX 280. The RV770 chip has a 256-bit (32-byte) interface, which means we can calculate the bandwidth in the following way:
ATI RV770: 32-byte x 4.8 Billion Transfers/sec = 150 GB/s
Nvidia GT200: 64-byte x 2.2 Billion Transfers/sec = 138.37 GB/s
Not surprisingly, the first "Super RV770" product is already out. Diamond Multimedia teamed up with Danger Den and Smooth Creations to launch the "ATI Radeon HD 4870 XOC Black Edition".
Diamond clocked the core of its Super RV770 at 800 MHz and the memory at 4.4 GT/s (1100 MHz QDR). There is some room for overclocking left, as Diamond its own version of an unlocked BIOS and the board can be propelled up to 950 MHz for the GPU and 1200 MHz for the memory.
Mario Gastelum, director of product development & engineering stated that the “The Diamond Radeon HD 4870 XOC Black Edition was clocked to kick some ass. We wanted the fastest card that could kick the living daylights and bust some performance records".
Diamond's version of the Super RV770 is available in systems from Smoothcreations. Expect more announcements from other AMD/ATI partners, including Asus, Sapphire and GeCube.
:up:
You and me both :D I see texture load stuttering in a couple of games that I play, it would be really smooth without that. Oblivion and Age of Conan are the two games I'm thinking of, AoC has texture stuttering on high settings in cities (from player armor textures I think) even though the game is rendering at 30-50fps.
Come on, 512MB frame buffer has been here for five generations now (X800XL 512MB was the first ATI gaming card I belive). Bet this will mimic 7800GTX 256. I think I'm going to get one just the same, because in contrast to the uber expensive 7800GTX 256 the 4870 is super cheap at launch!
So, with the fan speed increased to say, 60-100%, does that improve the overclockability of the 4870? To 850MHz on stock voltage, perhaps?