| Monday, April 21 2008 |

ATI is moving ahead the launch of its next-generation video processors, as reported by TG Daily. Despite its previous faliures in the graphics business and $8 million loss in the first quarter of this year - AMD's president and chief operating officer Dirk Meyer stated that the company will roll out a significant number of products in May ahead of schedule, including the new RV770 and mobile (M88) graphics parts. RV770 will launch as Radeon HD 4800 and will make its way into the FireStream stream processor and FireGL workstation cards. Both GDDR3 and GDDR5 memories ranging from 256MB to 1024MB will be supported by the chip, but ATI itself will only be offering GDDR5 cards with 256-bit memory controller. The Radeon HD 4850 version is set to come to market with an 800+ MHz core (the final clock has not been specified yet and will not be available until the final qualification is completed), while the 4870 will be the first mass-production GPU with a clock speed higher than 1GHz. The graphics processor itself will integrate more texture memory units (TMUs) - 32 in RV770 against 16 in R6xx. Expect mass availability of RV770 cards soon after their announcement in May.
Source: TG Daily
Source: TG Daily
User comments
Ati .. to di worl'. Wonder if i should still get that 3850?
Sounds promising. I can't wait to se performance numbers, and more importantly pricing. Hopefully ATi can now compete on both fronts.
Man, I bet you could cut the tension with a knife at AMD. I would NOT want to be working there right now! :eek:
On the upside, their new products rolling out should put them back in the saddle for awhile..
On the upside, their new products rolling out should put them back in the saddle for awhile..
yes yes yes!
this is what i want to hear...cant wait to see the numbers!
this is what i want to hear...cant wait to see the numbers!
4870 clocked higher than 1GHz! dang..
This is good news, it looks like AMD is coming back in their graphics business. I'm really looking forward to their future solutions, especially the multi-cored models. What we need is developers to start programming for AMDs solution.
by: Scrizz;759535hells yea man :rockout:
4870 clocked higher than 1GHz! dang..
this seriously just made my day!
Those new cards sound delicious, so now i have to make the decision of going crossfire or buying a 4870 :P
i wish the 4xxx series would work with the 3xxx series....
Leaked benchmarks, where art thou?
I'm glad I haven't bought an upgrade yet.
2 x 4870x2 FTW !
2 x 4870x2 FTW !
Sooo tempting :D
hopefully this series gives decent performance compared to NVidia.
Wow, that was quick. The HD 3000 series is still pretty fresh.
hope they get a sample top teh folding fellaz as soon as possible.
So, can the 3870x2 be Crossfired with the 4870x2???
Curious if you can Crossfire a 3870 with a 4870...
I wouldn't want to if the R770 does have 32 TMUs. Compound that with GDDR5 and it would be a waste IMO.
WHAT?!?!?!??! GDDR5 already when Nvidia only has GDDR3. You think this thing can seriously own 9800 GX2?
by: entilza;759573Not likely..
Curious if you can Crossfire a 3870 with a 4870...
ATI may find itself with a little spending cash at the end of this year if they keep up their recent fortune.. These cards look sic!!:rockout:
I can't wait for a 4870x2!! Goodbye 3870's in crossfire.:D
Im waiting for the x2 aint gonna bother with its first offerings
OOooohh shiney!!!
May look to swapping my 2900Pro(flashed XT) to a 4870 :P should be a nice boost hehe.
Although I wounder what temps would be like when clocked over 1GHz!?
May look to swapping my 2900Pro(flashed XT) to a 4870 :P should be a nice boost hehe.
Although I wounder what temps would be like when clocked over 1GHz!?
Core clock of over 1Ghz means ZILCH without other statistics. It's like saying there's a new P4 Netburst Extra Extreme Edition P4EEE with 5Ghz. It's a worthless power-consuming hog. We need architecture changes, fab shrinks and multiparallelism.
If they said 500Mhz clock but with 512 memory bus and 1024 universal shader units, and 75W total consumption, then I'd be much more impressed.
If they said 500Mhz clock but with 512 memory bus and 1024 universal shader units, and 75W total consumption, then I'd be much more impressed.
Higher clocks are a cheaper way to get a little more performance.
One of these days, GPUs will be user-upgradeable, as will the ram.
"Yeah, I just dropped in the 35nm version of my old core, and my temps are way lower."
One of these days, GPUs will be user-upgradeable, as will the ram.
"Yeah, I just dropped in the 35nm version of my old core, and my temps are way lower."
douubt .. would cost more to put a socket on the gpu ... if it were on teh mobo now 

Well, if they find a way to do a die-shrink and have a larger yield on the chips, then the actual cores would be cheaper. So you would buy the card, and it would come with some core and ram, and you could change it and upgrade it with bigger and better stuff.
Cheaper cores + more expensive cards = roughly equally priced cards.
Cheaper cores + more expensive cards = roughly equally priced cards.
by: erocker;759581+1, Ill be doing the same thing!
I can't wait for a 4870x2!! Goodbye 3870's in crossfire.:D
hell yeah!! 4870x2 for me baby yeah!
this is definately worth the wait, glad i didn't upgrade :D
"While we expect the desktop boards to be announced soon, RV770-based FireGL boards will not be introduced until Siggraph 2008, which will open its doors on August 11."
FireGL August... retail 4870 some months later... ie. they are announcing now... to prop up the share price... but launch is still Q3 into retail channels.
FireGL August... retail 4870 some months later... ie. they are announcing now... to prop up the share price... but launch is still Q3 into retail channels.
by: lemonadesoda;759760Where are you getting that the 4870 will come out after FireGL?
"While we expect the desktop boards to be announced soon, RV770-based FireGL boards will not be introduced until Siggraph 2008, which will open its doors on August 11."
FireGL August... retail 4870 some months later... ie. they are announcing now... to prop up the share price... but launch is still Q3 into retail channels.
"RV770 will launch as Radeon 4800 and will make its way into the FireStream stream processor and FireGL workstation cards." should suggest otherwise... likewise with this line "We hear the company has struck gold with the RV770 and will be bring the chip to market earlier than expected."
The new desktop cards from what I remember were always slated for June/July.. lately there's been rumors about it being ready ahead of time and this just backs that up further so you suggesting that it won't come out till even after August doesn't make much sense unless you're just trying to stir up poop.
Grammatically the sentence is open to interpretation. I could be wrong. I could be right. Given recent profits situation, loss of staff, loss of market share, etc. they need to get a few "product launch" sound bites into the market. Otherwise that share price will tank. That seems to me to be the rationale of the "timely" press release. If consumer 4850/70 is in the retail channel in Q2 (ie BEFORE END JUNE) then I will be impressed, and I take back my cynicism. But I doubt it. I think this will be Q3 before you are running these things at home.
Awesome that will give nvidia a kick in the balls, but it sucks that i just bought an x1950pro
could have make use of all those 3xxx goin for sale
could have make use of all those 3xxx goin for sale
by: lemonadesoda;759760last i check the firegl models of the rv670 came out after the 3870...second these 4870 desktop cards are slated for a quarter 2 release...so i would think that the latest is june as it ends the second quarter...these cards have been touted as being more than twice as fast as the rv670 also articles floating around the web have these cards slated @ 300...way less than Nvidias next iteration the 9900s which is suppose to be twice as much ($600). Ati is playing it smart by planning to release earlier probably catching Nvidia off guard...unlike the whole 3870 fiasco..in which nvidia got wind of the 3800series and rushed the 8800gt.
"While we expect the desktop boards to be announced soon, RV770-based FireGL boards will not be introduced until Siggraph 2008, which will open its doors on August 11."
FireGL August... retail 4870 some months later... ie. they are announcing now... to prop up the share price... but launch is still Q3 into retail channels.
One word: Sweet!!!
by: snuif09;759901how long since you bought that 1950 pro...because you prolly' could return it...not sayin it is ethical but at the same time we wouldn't be given the opportunity to return if it wasn't now would we.
Awesome that will give nvidia a kick in the balls, but it sucks that i just bought an x1950pro
could have make use of all those 3xxx goin for sale
by: springs113;759921
how long since you bought that 1950 pro...because you prolly' could return it...not sayin it is ethical but at the same time we wouldn't be given the opportunity to return if it wasn't now would we.
i bought it second hand so i cant return it
Still, if we're comparing the 4870 specs to 3870 specs, we should expect AT LEAST a 50% boost, that doesn't sound too bad at all. If GDDR5 can also boost its performance noticeably, ATI might be able to hold the performance crown once again.
Im waiting for the x2 aint gonna bother with its first offerings
by: asb2106;759709
+1, Ill be doing the same thing!
Thats if they make the X2 , it may not be vialble with this new setup, only time will tell I guess , I hope they are going to make these pcie 3.0 complient,.
by: REVHEAD;760032This is true. Ill prob flip for 2 4870's. Im gonna scope it out first and see how they perform. I just spent quite some time volt moddin my 3870s to hit 1000core speed(which they do!!)
Thats if they make the X2 , it may not be vialble with this new setup, only time will tell I guess , I hope they are going to make these pcie 3.0 complient,.
So if the performance is not that much greater than what Im getting, Ill just wait for them to go cheap like the 3870s are now :)
by: lemonadesoda;759652well we can dream can't we?
Core clock of over 1Ghz means ZILCH without other statistics. It's like saying there's a new P4 Netburst Extra Extreme Edition P4EEE with 5Ghz. It's a worthless power-consuming hog. We need architecture changes, fab shrinks and multiparallelism.
If they said 500Mhz clock but with 512 memory bus and 1024 universal shader units, and 75W total consumption, then I'd be much more impressed.
The trouble though with those kind of posts is that it makes it seem like nothing is good enough for you, nvidia's at 128 shaders, why should ati have to be at 1024?
the 4870 should have 480stream processors through 96shaders. making the x2 a beast with 960sp's and 192 shaders. combine that with 32tmu's per core and gddr5 and it starts to look really nice. while again that is on paper we can only hope that this thing actually delivers.
Of course I wouldn't mind if ati increased the rop's to 32 and reduced the amount of simple shaders while increasing the amount of complex (something like a 4x4 setup would be sweet) I think the 4870 will be a decent performer.
cant wait :) i hope we can pay it specialy end this year :D
Still 16 ROP's? Pathetic. Even if these cards sport 1GHz core clocks, that would still translate into relatively poor raw pixel pushing power (2008-wise) and since there are no major architectural changes from the 29xx lineup, inclusion of a 256bit memory interface will only exacerbate filtering and/or AA issues AMD's GPU's suffer from.
(Note: AMD has more or less admitted that the 512bit memory interface on the 29xx lineup was really there to cover up for poor filtering/AA performance and architectural flaws than anything else.)
It seems everyone is "budgeting" their GPU's nowadays. Nvidia with cut-down G92 and its various offspring and ATI with their "tweaked" and "cooler" 38xx lineup. Both of which are nearly identical to their predecessors aside from being manufactured on smaller fabrication technology which allowed for somewhat higher core clocks.
I think the days of monster GPU's are long gone. Massive performance jumps and architectural improvements from one generation to another seem to be a thing of the past. Anyone who's expecting a monster GPU from AMD this summer, and then a monster GPU from nVidia in retaliation, is fooling themselves. AMD simply can't afford to make one and nVidia simply has no reason to make one if their competition can't do it to begin with.
Anytime a GPU developer pushes the the power-draw of their latest generation GPU as the foremost feature, you know for certain that performance has either suffered or hasn't improved much or at all vs. the previous generation.
(Note: AMD has more or less admitted that the 512bit memory interface on the 29xx lineup was really there to cover up for poor filtering/AA performance and architectural flaws than anything else.)
It seems everyone is "budgeting" their GPU's nowadays. Nvidia with cut-down G92 and its various offspring and ATI with their "tweaked" and "cooler" 38xx lineup. Both of which are nearly identical to their predecessors aside from being manufactured on smaller fabrication technology which allowed for somewhat higher core clocks.
I think the days of monster GPU's are long gone. Massive performance jumps and architectural improvements from one generation to another seem to be a thing of the past. Anyone who's expecting a monster GPU from AMD this summer, and then a monster GPU from nVidia in retaliation, is fooling themselves. AMD simply can't afford to make one and nVidia simply has no reason to make one if their competition can't do it to begin with.
Anytime a GPU developer pushes the the power-draw of their latest generation GPU as the foremost feature, you know for certain that performance has either suffered or hasn't improved much or at all vs. the previous generation.
ATi has been under so much pressure lately, I bet this thing will roxorz nvidia's soxorz
hey addsub ROPs are outdated everyone uses shader units now
hey addsub ROPs are outdated everyone uses shader units now
by: lemonadesoda;759849
Grammatically the sentence is open to interpretation. I could be wrong. I could be right. Given recent profits situation, loss of staff, loss of market share, etc. they need to get a few "product launch" sound bites into the market. Otherwise that share price will tank. That seems to me to be the rationale of the "timely" press release. If consumer 4850/70 is in the retail channel in Q2 (ie BEFORE END JUNE) then I will be impressed, and I take back my cynicism. But I doubt it. I think this will be Q3 before you are running these things at home.
I doubt it -- what you're saying seems to be based in bias more than rationale. There really seems to be absolutely no reason to think that it would come after the Radeon cards. Like someone mentioned above, even if you want to argue how the quotes can be interpreted recent history should point to the idea the the workstation cards would come after the fact.
Some of the other things you mention are also sounding a little off. While AMD is not in the green yet the last article I read on their profit situation pointed out that while it dropped from Q4 2007 it is up vs. where they were this time last year. Given the effects of holiday spending and the release of the 3000 series it probably should be expected for it to drop from Q4 last year by virtue of Q4 just being a particulary strong quarter.. the increase from this quarter last year should probably suggest that things are picking up for them (or if you prefer something that retains more pessimism, they're only in the shit up to the knees this year where they might have been up to their waists last year). Likewise, last I read they actually gained marketshare (from an article on VR-zone a couple of weeks back) with their 3000 series.
The AA problems from the 2xxx and 3xxx cards are supposed to be fix on these cards. So the major drops should be a thing of the past. Also, I personally think adding the extra TMU's will boost ATI's performace quite noticeably has well. 50% sounds nice. Cept in Crysis. 25 avg on a my 3850 to 40? But thats Crysis's fault. Although I have found that the 8.3/4 drivers have made it play nicer.
Still only 256bit ring bus though. I wreckon the 512 on the HD 2900 XT didnt go like they though or was over kill? for a card this caliber you would think they would up it. Also, one thing that bothers me, the quote about them failing miserably with their graphics section. I dont think that is true, taken from the fact that the x1900 on up has sold extremely well and even the uber ocers are taking ATI cards and reaching them high scores :D
damn - I've been holding off for a 3870x2, I guess I'll continue to wait until the 4870x2 is out!
Asides, though, I'm a little concerned by AMD saying they're going to be releasing stuff ahead of schedule - that could be either good or bad, depending on how "ready" the product is.
Little bummed about the 256-bit MEM BUS, but, IMO, GDDR5 partnered with a 1GHz GPU? I think she'll still be game. Although, 256b might be pushing it if a card gets packed with 1GB of GDDR5. I think that 1GB claim will prob be for the GDDR3 - on that note, why 3?! Why not use 4 instead? If a whole series was stouting GDDR4 and GDDR5, they'd be ahead of the curve over nVidia on a technological level.
GDDR4 must still be uber expensive, ATM.
Asides, though, I'm a little concerned by AMD saying they're going to be releasing stuff ahead of schedule - that could be either good or bad, depending on how "ready" the product is.
Little bummed about the 256-bit MEM BUS, but, IMO, GDDR5 partnered with a 1GHz GPU? I think she'll still be game. Although, 256b might be pushing it if a card gets packed with 1GB of GDDR5. I think that 1GB claim will prob be for the GDDR3 - on that note, why 3?! Why not use 4 instead? If a whole series was stouting GDDR4 and GDDR5, they'd be ahead of the curve over nVidia on a technological level.
GDDR4 must still be uber expensive, ATM.
by: brian.ca;7602101./ What bias? LOL, I'm an ATI fanboi... and ONLY got ATi Radeons and FireGLs. LOL
what you're saying seems to be based in bias...
I read on their profit situation pointed out that while it dropped from Q4 2007 it is up vs. where they were this time last year.
2./ You cant really compare the unaudited Q1 vs Q1 figures for 2007 and 2008. In both situations there is a significant loss. In 2007 there were significant costs associated with the Ati aquisition. To say profit is improved and everything is OK is falling into the apples vs. oranges trap. There are so many significant adjustments to the window-dressed accounts. Only last week the AMD board had a showdown with investment analysts and they were NOT convinced that the AMD board was delivering... (google for the transcript). AMD *must* do some +ve PR work at this time. If not, not only will the share price sink further... but there will be pressure to replace board members.
I'm NOT arguing about dates of which model gets released first, firegl or consumer. Thats not the issue. The point is the "release being brought forward"... " we are ready to release"... "revenues coming soon"... "promise"... is all *necessary* PR to keep the (investor)market happy.
by: lemonadesoda;759652We know it's meant to be based on R600, so an RV670 GPU at 1GHz would be pretty impressive itself, plus the extra shaders, it's looking to be a very nice line up.
Core clock of over 1Ghz means ZILCH without other statistics. It's like saying there's a new P4 Netburst Extra Extreme Edition P4EEE with 5Ghz. It's a worthless power-consuming hog. We need architecture changes, fab shrinks and multiparallelism.
If they said 500Mhz clock but with 512 memory bus and 1024 universal shader units, and 75W total consumption, then I'd be much more impressed.
by: suraswami;759553Now you know that's not true, sure they're not the fastest, but they're not that far behind NV, especially when it comes to price/performance ratio.
hopefully this series gives decent performance compared to NVidia.
The 9800gtx really isn't that much faster than the 8800gtx. Nvidia must be quaking in it's boots after seeing that its brand new gpu will be squashed so soon.
I personally doubt that AMD will make the same mistake twice. They're not going to release a product before it's entirely ready again like the 2900xt.
I personally doubt that AMD will make the same mistake twice. They're not going to release a product before it's entirely ready again like the 2900xt.
by: flashstar;760327
The 9800gtx really isn't that much faster than the 8800gtx. Nvidia must be quaking in it's boots after seeing that its brand new gpu will be squashed so soon.
I personally doubt that AMD will make the same mistake twice. They're not going to release a product before it's entirely ready again like the 2900xt.
Even funnier, rumor has been that the HD5000 series is slated for early to mid 09, and will be ushering in the R700 dual-core GPU.
We all need to go grab some popcorn for the bitch-slapping fest that's about to kick back up between ATI and nVidia! :rockout:
next year will be an impressive year for GPU's, Ati will make nvidia step up their game by the end of this game and nvidia will have to release a new gpu.
very nice
very nice
by: KainXS;760390nvidia will be releasing a new card that supposedly built from the ground up... something after the release of these ati cards...
next year will be an impressive year for GPU's, Ati will make nvidia step up their game by the end of this game and nvidia will have to release a new gpu.
very nice
so on the graphics front things will be really nice...another note...on the cpu side of things amd demoed a feature in their upcoming chipset that will allow higher clocks(in windows) of their processors...and their 45nm have been demoed to hit 3.2ghz.
by: lemonadesoda;760289/shrug.. you don't seem it, but one way or the other I said bias rather than rationale -- ie: the rationale seemed missing (particularly in the original quoated statement) so I chalked it up to bias, if you're not biased feel free to attribute the faulty logic to something else.
1./ What bias? LOL, I'm an ATI fanboi... and ONLY got ATi Radeons and FireGLs. LOL
2./ You cant really compare the unaudited Q1 vs Q1 figures for 2007 and 2008. In both situations there is a significant loss. In 2007 there were significant costs associated with the Ati aquisition. To say profit is improved and everything is OK is falling into the apples vs. oranges trap. There are so many significant adjustments to the window-dressed accounts. Only last week the AMD board had a showdown with investment analysts and they were NOT convinced that the AMD board was delivering... (google for the transcript). AMD *must* do some +ve PR work at this time. If not, not only will the share price sink further... but there will be pressure to replace board members.
I'm NOT arguing about dates of which model gets released first, firegl or consumer. Thats not the issue. The point is the "release being brought forward"... " we are ready to release"... "revenues coming soon"... "promise"... is all *necessary* PR to keep the (investor)market happy.
Looking at an article attached to google fiance's page for AMD real quick there's this, "Sunnyvale, Calif.-based Advanced Micro Devices Inc. said that during the quarter it lost $358 million, or 59 cents per share, compared with a loss of $611 million, or $1.11 per share, in the same period a year earlier. The latest quarter's results include charges of 8 cents a share for the acquisition of graphics chip maker ATI Technologies." Looking at their website (2006 numbers are also noted as unaudited for reference) the ATI aquisiton charges for Q1 07 were $113 million, this year $50 million... they narrowed their losses by $253m this year -- going by the numbers above the difference in ATI aquisition charges only accounts for $63m out of that $253m... so they've still narrowed their loss a fair bit even exluding the ATI acquisition charges.
But keep in mind, I'm not aruging that everything is peachy keen and right as rain... the knee deep analogy should have pointed to that. But I don't buy that they're in that much worse of a position than they've been for the last year and a half or two that they would have to be putting out a bunch of PR that may not hold too much water to compensate. From everything I've heard to date before this article, the 4000 series was slated for release for June/July and was well on course to that end. So, to me, that they might be able to push it out a month or so ahead of schedule doesn't sound implausible / probably shouldn't be written off as PR bs.
And I understand your main point, but don't forget the original statement I replied to was "FireGL August... retail 4870 some months later... ie. they are announcing now... to prop up the share price... but launch is still Q3 into retail channels." You seemed to be arguing this was mostly PR (necessary in your opinion) but release would still be later, and seemed to be basing that off the idea of the 4000s coming out after the FireGLs which were coming out in August... that was the only reason it became any bit of an issue. B/c that was not a reason to think that the release would actually be later (or rather remain the original slated release date) which would make all this empty PR.
by: WarEagleAU;760254Don't quote me on this one but from what I read and understand using the larger bus was overkill & at the same time added a significant cost to producing the cards. The reason you saw a lot of those cards being used for OCing records was b/c that overkill equated to head room when doing some of that extreme OCing mumbo jumbo.
Still only 256bit ring bus though. I wreckon the 512 on the HD 2900 XT didnt go like they though or was over kill? for a card this caliber you would think they would up it. Also, one thing that bothers me, the quote about them failing miserably with their graphics section. I dont think that is true, taken from the fact that the x1900 on up has sold extremely well and even the uber ocers are taking ATI cards and reaching them high scores :D
But from what I understand a larger bus isn't really needed when you have faster memory. The orginal article linked above seems to point to this saying, "With a 256-bit memory controller, we're talking about 115 to 141 GB/s of bandwidth. This number equals the memory bandwidth record set by the 2900XT 1GB GDDR4 (512-bit interface with GDDR4 at 1.1 GHz DDR)."
Going by the full article above it sounds like the GDDR3 variants are aimed more at OEMs and GDDR5s at the retail level. So I guess the 3s are to help OEMs keep costs down.
Little bummed about the 256-bit MEM BUS, but, IMO, GDDR5 partnered with a 1GHz GPU? I think she'll still be game. Although, 256b might be pushing it if a card gets packed with 1GB of GDDR5. I think that 1GB claim will prob be for the GDDR3 - on that note, why 3?! Why not use 4 instead? If a whole series was stouting GDDR4 and GDDR5, they'd be ahead of the curve over nVidia on a technological level.
by: brian.ca;760499
Don't quote me on this one but from what I read and understand using the larger bus was overkill & at the same time added a significant cost to producing the cards. The reason you saw a lot of those cards being used for OCing records was b/c that overkill equated to head room when doing some of that extreme OCing mumbo jumbo.
But from what I understand a larger bus isn't really needed when you have faster memory. The orginal article linked above seems to point to this saying, "With a 256-bit memory controller, we're talking about 115 to 141 GB/s of bandwidth. This number equals the memory bandwidth record set by the 2900XT 1GB GDDR4 (512-bit interface with GDDR4 at 1.1 GHz DDR)."
Going by the full article above it sounds like the GDDR3 variants are aimed more at OEMs and GDDR5s at the retail level. So I guess the 3s are to help OEMs keep costs down.
I hope so. Really, though, I'd love to see ATI stouting only DDR4/5 with this series, it would give them a slight edge on nVidia as far as spec sheets go.
I think you've got it right about the MEM BUS as well, partly why I mentioned 256bit isn't that big a deal if the GPU is clocked at 1GHz with DDR5. The bandwidth of the MEM itself will make up for it. But, as I also pointed out, if they're packing 1GB of high bandwidth MEM, a 256b BUS could prove to be a limitation - we'll have to see, the upgrading to 32 TMU might work out just nicely.
Either way, the next year and a half is stacking up to be quite competitive between red and green - which is what we all really want to see more than one camp leading the pack. We benefit more from close competition more than we do one leading and one trailing.
Wonder what Nvidia is going to usher against these ones. 9800 series seem to get old before it even got started, then again it got starte about 1.5y ago :D
gg nvidia? The fact that AMD immediately saw that the R600 series wasn't going to be stellar and decided only to improve in a way that wouldn't really have implications on the R&D effort for the RV700, which they have been working on for ages.
Um FYI, the 512bit/384bit memory bus is really redundant at this stage as the architecture and the GPUs don't, and cant use the 512 memory bit bus addressing to its fullest potential.
Most of you guys are thinking way too "zomg 256bit memory bus suxxors". Its the raw calculating power of the GPU itself thats important as well as the efficiency of it. The bit width of the memory bus isn't important if the GPU architecture is poor.
Okay in this case not poor but weaker, say for example G92 vs RV670. GDDR4 evidently has way more memory bandwidth, however the RV670 is slower than the G92! Now Compare RV670 to R600. R600 has the 512 bit bus... any performance increases? Little to none. The GPU isnt fast enough/can't proccess that much to use the 512 bit width to its max potential, same reason why Nvidia took a step back as well.
Another thing, it costs more to make a card with a wider memory bus. Why you may ask? Because it is required to have more memory chips. Each chip is 32 bits. Therefore 32bits x 8 chips = 256; 256bit, 32bits x 12chips = 384bit, and finally, 32 bits x 16 chips = 512; 512 bit.... may seem obvious to some but that's why the G80/R600s were priced so damn high versus current 256 bit cards of equivalent. More memory chips, more components needed onboard and finally a requirement for a longer PCB (Usually) due to increased power consumption from the extra chips as well as the core (larger memory controller).
[sarcasm]Hey look!!!! Its awesome that ATI Ripped out their ROPs... now I cant even game in 3D AWESOME!!![/sarcasm].
ROPs are needed FYI.
I'm guessing the reasons why the numbers of components are in a core are because:
1. Core balancing, as with multi GPU technologies, I've noticed the linear decrease in performance as you add more GPUs. This means that GPU R&D HAVE to balance out the core; more doesn't equal better a lot of the times, and i think the same applies for GPUs. Within an architecture, you probably can only have a specific amount of parts for the GPU before you start getting decreases in efficiencies.
2. Another reason is the fact that the numbers make it modular to manufacture
3. Cost/performance feasibility.
Its easy for you guys to go "HEY LETS CHUCK 1024 SHADERS AND 32 TMUs AS WELL AS A 512 BIT BUS!!!111", but wouldn't they have done it if it was THAT bloody easy?
Anyway guys, please stop arguing "you're a fanboy/you're biased!" with each other...
OT: I'm wondering if Intel's larabee will be even decent, the fact that its just a really powerful CPU thats not really designed to be dedicated to rendering somewhat worries me. However since one of their Xeons do ray-tracing at liek 60 fps or something I might be wrong (then again, games NEVER use ray-tracing...nor do GPUs have the ability.
by: lemonadesoda;759652
Core clock of over 1Ghz means ZILCH without other statistics. It's like saying there's a new P4 Netburst Extra Extreme Edition P4EEE with 5Ghz. It's a worthless power-consuming hog. We need architecture changes, fab shrinks and multiparallelism.
If they said 500Mhz clock but with 512 memory bus and 1024 universal shader units, and 75W total consumption, then I'd be much more impressed.
by: imperialreign;760515
I hope so. Really, though, I'd love to see ATI stouting only DDR4/5 with this series, it would give them a slight edge on nVidia as far as spec sheets go.
I think you've got it right about the MEM BUS as well, partly why I mentioned 256bit isn't that big a deal if the GPU is clocked at 1GHz with DDR5. The bandwidth of the MEM itself will make up for it. But, as I also pointed out, if they're packing 1GB of high bandwidth MEM, a 256b BUS could prove to be a limitation - we'll have to see, the upgrading to 32 TMU might work out just nicely.
Either way, the next year and a half is stacking up to be quite competitive between red and green - which is what we all really want to see more than one camp leading the pack. We benefit more from close competition more than we do one leading and one trailing.
Um FYI, the 512bit/384bit memory bus is really redundant at this stage as the architecture and the GPUs don't, and cant use the 512 memory bit bus addressing to its fullest potential.
Most of you guys are thinking way too "zomg 256bit memory bus suxxors". Its the raw calculating power of the GPU itself thats important as well as the efficiency of it. The bit width of the memory bus isn't important if the GPU architecture is poor.
Okay in this case not poor but weaker, say for example G92 vs RV670. GDDR4 evidently has way more memory bandwidth, however the RV670 is slower than the G92! Now Compare RV670 to R600. R600 has the 512 bit bus... any performance increases? Little to none. The GPU isnt fast enough/can't proccess that much to use the 512 bit width to its max potential, same reason why Nvidia took a step back as well.
Another thing, it costs more to make a card with a wider memory bus. Why you may ask? Because it is required to have more memory chips. Each chip is 32 bits. Therefore 32bits x 8 chips = 256; 256bit, 32bits x 12chips = 384bit, and finally, 32 bits x 16 chips = 512; 512 bit.... may seem obvious to some but that's why the G80/R600s were priced so damn high versus current 256 bit cards of equivalent. More memory chips, more components needed onboard and finally a requirement for a longer PCB (Usually) due to increased power consumption from the extra chips as well as the core (larger memory controller).
by: hat;760120
ATi has been under so much pressure lately, I bet this thing will roxorz nvidia's soxorz
hey addsub ROPs are outdated everyone uses shader units now
[sarcasm]Hey look!!!! Its awesome that ATI Ripped out their ROPs... now I cant even game in 3D AWESOME!!![/sarcasm].
ROPs are needed FYI.
I'm guessing the reasons why the numbers of components are in a core are because:
1. Core balancing, as with multi GPU technologies, I've noticed the linear decrease in performance as you add more GPUs. This means that GPU R&D HAVE to balance out the core; more doesn't equal better a lot of the times, and i think the same applies for GPUs. Within an architecture, you probably can only have a specific amount of parts for the GPU before you start getting decreases in efficiencies.
2. Another reason is the fact that the numbers make it modular to manufacture
3. Cost/performance feasibility.
Its easy for you guys to go "HEY LETS CHUCK 1024 SHADERS AND 32 TMUs AS WELL AS A 512 BIT BUS!!!111", but wouldn't they have done it if it was THAT bloody easy?
Anyway guys, please stop arguing "you're a fanboy/you're biased!" with each other...
OT: I'm wondering if Intel's larabee will be even decent, the fact that its just a really powerful CPU thats not really designed to be dedicated to rendering somewhat worries me. However since one of their Xeons do ray-tracing at liek 60 fps or something I might be wrong (then again, games NEVER use ray-tracing...nor do GPUs have the ability.
hehe the illogical fanboi thread!
Seriously, 'omg nvidia is quaking' 'nv will crush this' - stop. really. The last real major breakthrough in the video market was the 8800GTX, and it hasnt changed much since then - slapping two cards onto one PCI-E slot is nice, but its not new.
In the end, all it comes down to is who has the fastest card (to gather more fanbois/investors) and who has the most popular card - the price to performance ratio. The 9600GT and 3850 are filling that segment right now, but they dont match the 8800GTX or 3870x2 in performance do they?
Who has the fastest doesnt really matter, its who has the most cost effective.
Seriously, 'omg nvidia is quaking' 'nv will crush this' - stop. really. The last real major breakthrough in the video market was the 8800GTX, and it hasnt changed much since then - slapping two cards onto one PCI-E slot is nice, but its not new.
In the end, all it comes down to is who has the fastest card (to gather more fanbois/investors) and who has the most popular card - the price to performance ratio. The 9600GT and 3850 are filling that segment right now, but they dont match the 8800GTX or 3870x2 in performance do they?
Who has the fastest doesnt really matter, its who has the most cost effective.
I love cheap performers as much as the next guy. If I had to buy a card I would go for the pp champs rather than the uber cards that fall from grace in 2 months. I actually bought a 3850 for my sis's pc the other day so she could use her br-drive & now she's sitting up there playing assassin's creed :confused:
However, I still love to play with the uber cards & will happily beat the crap out of a 4870x2 when it comes out :D
However, I still love to play with the uber cards & will happily beat the crap out of a 4870x2 when it comes out :D
ROP'S, TMU's and SP's ALL required :D
by: Mussels;760665your quite right, i dont think these cards can even stay on the shelves.
hehe the illogical fanboi thread!
Seriously, 'omg nvidia is quaking' 'nv will crush this' - stop. really. The last real major breakthrough in the video market was the 8800GTX, and it hasnt changed much since then - slapping two cards onto one PCI-E slot is nice, but its not new.
In the end, all it comes down to is who has the fastest card (to gather more fanbois/investors) and who has the most popular card - the price to performance ratio. The 9600GT and 3850 are filling that segment right now, but they dont match the 8800GTX or 3870x2 in performance do they?
Who has the fastest doesnt really matter, its who has the most cost effective.
by: Scrizz;759535
4870 clocked higher than 1GHz! dang..
Yup, that's really needed if the shaders on the RV770 don't have clock-generators of their own. Maybe part of the reason behind ATI's approach to shaders not paying off well so far is because the shaders use the core's clock.
To think I was going to upgrade the cooler and start overclocking, I might as well just get 2 4870s and xfire those bad boys up hopefully when they come out in may!
looks promising hope we can get a good driver released..this time
il be getting a 4870x2..hopefully...and maybe a 4870 as well...:P depends on my funding
i'll wait for a higher res screen before doing that... lol.
Just to add some sanity here: do you NEED that much performance for current/upcoming titles? i'm seeing a lot of games lately that work fine on max details on an 8800GT, so are you sure you want to go that far?
Just to add some sanity here: do you NEED that much performance for current/upcoming titles? i'm seeing a lot of games lately that work fine on max details on an 8800GT, so are you sure you want to go that far?
by: Mussels;769835I am just being greedy really :) no reason :p
i'll wait for a higher res screen before doing that... lol.
Just to add some sanity here: do you NEED that much performance for current/upcoming titles? i'm seeing a lot of games lately that work fine on max details on an 8800GT, so are you sure you want to go that far?
by: azazel;769831if ya do u need to sell me your X2 ;)
il be getting a 4870x2..hopefully...and maybe a 4870 as well...:P depends on my funding
i wont get a 4xxx card unless the performance is just amazing compared to current 3xxx cards
