PDA

View Full Version : Your take on what happened to the R400


Pages : [1] 2 3

lokeshray
02-13-04, 05:57 PM
first post here

last time i heard about the R400, it was supposed to be released in June-july 2003.

and then came the rumors saying R400 has been cancelled and that there is now supposed to be an R420.

the way ATI has gone about dealing with the competition seems pretty lame-ass. ATI had nvidia at their knees with the release of the radeon 9700 pro and the subsequent lacklustre peformance of the NV30.
ATI just had to deliver the killer blow by releasing the R400. even if they would have released the existing R300 on a 0.13 micron core, they would have outclassed all the nv3x based processors.

instead, ATI decided to play it safe, and just went on relasing higher clocked variations of the R300 ..thus they let go of the 6 month advantage they had over nvidia with the release of the radeon 9700 pro.

as far as speculation is concerned, i think there are 2 reasons why The R400 was cancelled.

1]
the x-box2
when ati got the x-box 2 project, it resulted in a change in the entire roadmap. ati also has to make the graphics controller for the next generation nintendo console. with the existing roadmap it would be impossible to deal with both gaming consoles

2] ATI realised that a heavily modified r300 core on a 0.13 micron process would be able to beat the R400. the R300 was an 8x1 pipeline design but , for multi-texturing with 2 textures per pixex it behaved like a 4x2 system. thats why the performance of the geforce4 ti-4600 was so close to that of the radeon 9700 pro in multi-texturing. The R400 must have been a 16x1 design and ati realised that a modified (HEAVILY MODIFIED) r300 core on 0.13 micron process with an 8x2 design could beat the R400. also,
its pretty obvious that it costs much less to modify an existing architecture rather than design an entirely new architecture. there is less risk involved in using the r300 architecture(modified) on a 0.13 micron process.

there were 3 major risk factors while developing the R400.:-
............1]high end product on 0.13 micron process. lots of problems with 0.13 micron process
............2]new architecture
............3]low yields on 0.13 micron process


so, it seems ATI did a cost benefit analysis and decided to go for the major modification approach. i believe the R400 has not been cancelled, it has ben rescheduled for a year later. This allows ATI to work out all the kinks in designing on a 0.13 micron process, i believe the R400 will be the last chip ATI makes on a 0.13 micron process (or 0.11 process??)

seems that ATI has been working for a very long time on the R400 AND R500, so, we can expect those chips to be released slightly earlier than expected. ATI is working on the R500 for including it in the x-box2. remember how soon nvidia released the NV25 after releasing x-box 1?(GEFORCE 4TI SERIES)


Anybody got more information on the R400?

PS:::this scenario REMINDS ME OF WORLD WAR 2, the way ATI had nvidia on their knees and nvidia had no answer reminds me of Great Britain after the retreat from DUNKIRK. most of the british munitions were left back at Dunkirk. As a result, Great Britain had a severe shortage of weapons. If Germany had attacked Great Britain at that time, they would have won the war. Instead, the Germans went and sat up in the forts along the maginot line along the border with france and nothing happened.......

we knw what happened later..........
Great Britain won the war

-=DVS=-
02-13-04, 06:08 PM
Actually R400 is now a R500
And R420 is Improved R350 with some of the original R400 stuff...

Atleast that what people said a while ago. Doesn't realy matter what happend to R400 , as long as new Rxxx release is faster and better. :D

nobie
02-13-04, 11:15 PM
Originally posted by lokeshray

PS:::this scenario REMINDS ME OF WORLD WAR 2

let's not overdramatize it :p

Hellbinder
02-14-04, 12:07 AM
The R400 had some very inventive ways of doing things that simply were taking to long to engineer to fit into the product Cycle origionally intended.

So basically what happened is they (ATi) reached a point about a year ago where they realized that the R400 was simply an impossability in the Given time frame. It caused Quite a Stir from what i understand.... Something akin to a mild heart attack as you might imagine...

What was decided upon after much discussion and debate is that the Core Technologies of the R400 project would get perfected and increased in order to make their Debute as the R500. Which of course origionally would have had much of the R400 concepts in it. Now theough the R500 is going to me much more than that. Litterally having an aditional 18 months of Development time considering the R400 project was rolled into it.

Things like, Unified independant Pipelines, Emmbedded Memory, Massivley improved Occlusion Culling etc...

After the passing of the Mild heart attack ATi had to decide on an immidiate course of action for the next Gen part that they would need now. That project is known as Loki or R420. Which from what I understand targeted at nothing less than at *least* doubbling the Power of the R300 Series in every area and in some casses Trippling it. There is not a lot of Finesse going on in there like the R500 will have. The R420 has one goal and one goal only,,, having the Raw rendering power to put ATi in the performance crown for the next year. It is seriously that simple. in fact ATi is this go round willing to sacrifice Margins for Performance. Something they are not normally given to. In other words the R420 *WILL*, and there can be no doubt, be faster than the Nv40. For the Simple reason that They are only at about 160M Transistors with Low-K and a new Transistor process that doubbles transistor state speed. Thus... They will be able to Clock the Core through the Roof just like the RV360 can now. Except the process is will be in its second generation for ATi and they have improved on it. On top of that they are using some seriously fast ram and an improved memory controler.

Floating point performance will be a HUGE leap over the current crop of cards. As will Fill rate, and a considerable Jump in Bandwidth. The Pipelines of the R420 are massively powerful as it is, each one more like Two pipelines than one.

Now you know and that is the abbridged version. I am sure that the actual details of what transpired are even more interesting.

Rytr
02-14-04, 12:31 AM
Thanks for the abbridged version HB, interesting. :)

EMunEeE
02-14-04, 12:40 AM
Originally posted by Rytr
Thanks for the abbridged version HB, interesting. :)

Definately, makes, the R420 sound like the next BIG thing.

MuFu
02-14-04, 11:03 AM
This has been discussed to death, but I'd say the three main contributers were:

1) Over-ambitious targets for the projected TTM and R&D timeframe.
2) Longhorn/DX Next and API "stagnation" with DX9.
3) Xbox2.

The second point is critical. It's because there were only minor additional featureset deliverables (PS/VS 3.0 support) that basing the design on R300 was feasible. It shouldn't be overlooked that there's still a lot of the original R400 project in the mix as well (its VS architecture, for example).

MuFu.

lokeshray
02-15-04, 12:30 AM
Originally posted by -=DVS=-
Actually R400 is now a R500
And R420 is Improved R350 with some of the original R400 stuff...

Atleast that what people said a while ago. Doesn't realy matter what happend to R400 , as long as new Rxxx release is faster and better. :D


r400 is supposed to be 0.13 micron,
and r500 is 0.09 micron.
thus, i dont see how original R400 can be called R500,

ok, about the loss of money caused by cancellation of R400, how much would be the loss? is it enough to drown ATI?

ok about R420 performance,,,,,it will be using 1 GHZ memory ..
i dont see how r420 can beat nv40 especially since nv40 gonna have 1.4 ghz memory..bandwidth does matter..
and about ATI's approach to memory speeds..they seem sooo scared to use fast memory....unlike nvidia...
we all knw how good the r360 would have been if it would have used memory as fast as geforce fx 5950.
how can ATI think of beating the nv40 with slower memory? :confused:

-=DVS=-
02-15-04, 01:39 AM
Originally posted by lokeshray
r400 is supposed to be 0.13 micron,
and r500 is 0.09 micron.
thus, i dont see how original R400 can be called R500,

ok, about the loss of money caused by cancellation of R400, how much would be the loss? is it enough to drown ATI?

ok about R420 performance,,,,,it will be using 1 GHZ memory ..
i dont see how r420 can beat nv40 especially since nv40 gonna have 1.4 ghz memory..bandwidth does matter..
and about ATI's approach to memory speeds..they seem sooo scared to use fast memory....unlike nvidia...
we all knw how good the r360 would have been if it would have used memory as fast as geforce fx 5950.
how can ATI think of beating the nv40 with slower memory? :confused:

First of "r400 is supposed to be 0.13 micron,and r500 is 0.09 micron" is just a rumor and 0.09 micron tech is far away so.....its just a rumor.

Well obviouly you do not follow simple facts how for example AMD with less megaherz beat up Pentium 4,
same for Radeon it does more with less , example Radeon 9700Pro core and memory speed are nearly twice as small compared to NV30 , and yet it R9700Pro wins !

To make it simple megahertz and gigahertz mean squat because apple not equal to orange , even deeper spec like pipelines doesn't show who is supperior , there are to many factors , only actual product benchmarked in same systems under real world various games, can be indication of speed.

Cheers :cool:
P.S Hellbinder already said R400 went to R500 project :o just as i said.

AthlonXP1800
02-15-04, 03:33 AM
It already explained why R400 was cancelled back to November 2003, it not because of Xbox 2 project, it a long way before that, actually it all because of NV40.

There are 3 reasons why R400 cancelled:

1) ATI knew they cannot deliver it in the required timeframe. If they did, they will reproduce NV30 disaster.

2) ATI had some information about NV40 and realised NV40 is more powerful than R400.

3) R400 was very costly and high risk project. If ATI did launched R400 to compete with NV40 before the required timeframe, then R400 will be a huge financial losses for ATI.

With R400 cancelled, ATI moved the R400 and R500 projects, original R400 renamed as R500 and original R500 renamed as R600. And then ATI began low cost and low risk R420 project from scratch based on heavily modified R300 core. ATI hope R420 will be good as fast or slight fast than NV40.

jimbob0i0
02-15-04, 08:07 AM
Originally posted by AthlonXP1800
It already explained why R400 was cancelled back to November 2003, it not because of Xbox 2 project, it a long way before that, actually it all because of NV40.

There are 3 reasons why R400 cancelled:

1) ATI knew they cannot deliver it in the required timeframe. If they did, they will reproduce NV30 disaster.

2) ATI had some information about NV40 and realised NV40 is more powerful than R400.

3) R400 was very costly and high risk project. If ATI did launched R400 to compete with NV40 before the required timeframe, then R400 will be a huge financial losses for ATI.

With R400 cancelled, ATI moved the R400 and R500 projects, original R400 renamed as R500 and original R500 renamed as R600. And then ATI began low cost and low risk R420 project from scratch based on heavily modified R300 core. ATI hope R420 will be good as fast or slight fast than NV40.

Your point 2 was VERY different to what I had heard and kinda contradicts itself. Firstly *if* NV40 was supposed to be so much more powerful than R400 why turn R400 into the next next gen chip? If that was the case then the R500 will be totally and hopelessly underpowered... makes NO sense whatsoever.

I heard that the R400 targets were totally overkill compared to what they heard NV40 would be doing. As such they moved R400 to R500 and kept a few of the features to add to a smaller die size beefed up R300 codenamed Loki to face NV40.

In the light of which product is targeting the other for competition this version does make more sense.. however at teh end of teh day it's all hearsay and rumour and until CeBIT that's all it will be really.

AthlonXP1800
02-15-04, 08:47 AM
Originally posted by jimbob0i0
Your point 2 was VERY different to what I had heard and kinda contradicts itself. Firstly *if* NV40 was supposed to be so much more powerful than R400 why turn R400 into the next next gen chip? If that was the case then the R500 will be totally and hopelessly underpowered... makes NO sense whatsoever.

I heard that the R400 targets were totally overkill compared to what they heard NV40 would be doing. As such they moved R400 to R500 and kept a few of the features to add to a smaller die size beefed up R300 codenamed Loki to face NV40.

In the light of which product is targeting the other for competition this version does make more sense.. however at teh end of teh day it's all hearsay and rumour and until CeBIT that's all it will be really.

Yes R400 was packed with lots of new features just like NV30 did. When ATI got some info about NV40 and realised R400 is not powerful enough to face to face NV40 and decided it will not good to compete with NV40. ATI do not want to make mistakes with R400 the same way as Nvidia made mistakes with NV30. Everybody thought NV30 is very fast chip but in the end, it not. Same thing with R400, ATI think it not fast enough and renamed it to R500 and give it extra 18 months to make it go faster and put in more features. Now ATI have R420 based on R300 core is faster than original R400 at low price.

Mariner
02-15-04, 02:31 PM
Well the rumour that came out at the time was that the original R400 would have been far beyond the capabilities of NV40 that it was not necessary to release that technology - an updated R300 would do.

This only really makes sense if you are a rabid ATI admirer, however!

Personally, what would make more sense to me is that ATI found out some information about the expected performance of NV40. We've heard that the planned R400 would have been a major new architecture and hence it would be more risky to produce. If ATI believed they could easily produce an updated version of R300 technology (R420) which could match or even surpass NV40, they could do so, giving them more time to work on the new "R500" (i.e. originally R400) architecture. This would also give them more time to solve any problems encountered with the new technology (such as those NV encountered with NV30, for example).

AthlonXP1800
02-16-04, 04:39 AM
Originally posted by Mariner
Well the rumour that came out at the time was that the original R400 would have been far beyond the capabilities of NV40 that it was not necessary to release that technology - an updated R300 would do.


Yes R400 had more features than NV40 but it not going to be faster than NV40. This remind the same way as NV30 had more features than R300 but NV30 not faster than R300.

I got something very interesting to talk about the original projects you may not know about the same way why original R400 project cancelled.

Remember original 3dfx Rampage project that lots probably never heard and surprised. A few years ago I downloaded a huge 3dfx zip file contained the Glide source as well as all 3dfx projects.

The original 3dfx Rampage was scheduled to launched after Voodoo 2, Rampage had both 2D and 3D engine and 3dfx was not happy that Rampage is not that much fast 3dfx thought in between Voodoo 2 and Voodoo 3. 3dfx cancelled Rampage and began Banshee project, Banshee is just a Voodoo 2 with 2D engine added into the core. When 3dfx Voodoo 4 and 5 projects began, 3dfx decided to resumed Rampage project and wanted Rampage to launch after Voodoo 5, it added lots of features, increased core and memory speeds on Rampage card.

Then the hype started when 3dfx told the world about next generation Rampage after Voodoo 5 scheduled to launched in Q1 2001, it will be the most powerful chip that will last 2 generations, it will saved you money to buy 2 cards in 2 generations, it will be faster than NV20 and NV25 and it will have T&L co-processor called Sage that will floored everybody with very stunning graphics. Everybody is excited and looked forward to buy 3dfx Ramage in next few months.

Then in a suddenly everybody was shocked and mysterious at 3dfx winded up the company and sold the assets to Nvidia and many felt betrayed by 3dfx and Nvidia killed 3dfx to stopped the most powerful Rampage to launched and slaughtered NV25. But that the opposite, at 3dfx board meeting, received information about future NV25 and realised NV25 is much more powerful and it no way Rampage can compete NV25. 3dfx decided the due course of action, it either go ahead to launch Rampage and it will make 3dfx go out of business or the alternative option to get the money is to wind up the company and sell the assets to Nvidia.

After Nvidia bought 3dfx assets, Nvidia cancelled original NV30 design and then send 2 teams of Nvidia and ex 3dfx engineers to redesigned NV30 and incorporated some of 3dfx technologies into NV30. People hoped Nvidia will launched Rampage someday but it never see the light of the day.

Until more than a year ago somebody bought a mystery 3dfx prototype card from Ebay and assumed it a Voodoo 4 card but the core is much bigger than Voodoo 4 core. The guy installed the card and then boot the PC, the guy was surprised at the BIOS screen showed "3dfx Rampage", it managed to run Quake 3 ok with FSAA and AF on but it is buggy but it is as fast as Voodoo 5.

Mystery of the the most powerful 3dfx Rampage was solved. It never going to be powerful than Geforce 3 and Geforce 4 Ti, it never had a Sage T&L co-processor, Sage is non-existence and it not in any 3dfx projects and the Rampage card do not have a co-processor socket either. 3dfx Rampage is turned to be actually much improved Voodoo 5 5500 version with single processor instead of dual VSA-100 processors and sport extra features badly needed like anisotropic filtering but it still dont have hardware T&L unit.

-=DVS=-
02-16-04, 05:33 AM
Good story AthlonXP1800 , but its incomplete

Rampage project was never finished and 3Dfx died that why it had no Sage T&L unit :rolleyes: , there were suppose to come out several versions of Rampage , A. without T&L unit , B. with one Sage and C. with two sage as pro card , but obviously none of them made out......

Wish i had kept some data PDF's of it :( , poor 3Dfx it could have lived longer.

Hanners
02-16-04, 03:08 PM
Originally posted by AthlonXP1800
2) ATI had some information about NV40 and realised NV40 is more powerful than R400.

Wrong. It was in fact the exact opposite - R400 would have been overkill against anything NV40 could muster, so coupled with the fact that R400 would most likely have ended up being delayed, R420 was born.

AthlonXP1800
02-16-04, 03:58 PM
Originally posted by Hanners
Wrong. It was in fact the exact opposite - R400 would have been overkill against anything NV40 could muster, so coupled with the fact that R400 would most likely have ended up being delayed, R420 was born.

Well I dont think R400 was overkill, same thing Nvidia scientist David Kirk said 256 bit memory bus is overkill, so he turned to be wrong. NV30 need it badly.

randsom
02-16-04, 05:56 PM
AthlonXP1800, get a grip.

Paul
02-16-04, 06:51 PM
Originally posted by Hanners
Wrong. It was in fact the exact opposite - R400 would have been overkill against anything NV40 could muster

AFAIK, it had nothing to do with how it was going to compare to NV40. It was mostly a combination of your second point (delays, ambition) and the XBox2 project that did it.

Skuzzy
02-16-04, 07:20 PM
There was just no reason for ATI to spend the money on putting out another new piece of hardware. They had the lead with what they currently were shipping and nothing on the horizon from NVidia would figure to beat them.

Good business decision.

MuFu
02-16-04, 07:27 PM
Originally posted by Hanners
It was in fact the exact opposite - R400 would have been overkill against anything NV40 could muster...

Ha, you say that but wait till R420 comes out. I have absolutely no idea what the specs are (neither does anybody else that's speculated publically, AFAIK - including Anandtech) but believe it will be faster than the original R400 would have been in most situations.

Over-overkill, perhaps? Like microwaving a hamster for 30 minutes then running it over with a truck.

MuFu.

ClyssaN
02-16-04, 08:50 PM
I think the two main reasons were:

1: No dx10 for a good time.
2: The timeframe for the r400 was not achievable, that with the dx10 delay, the better choice was aiming for the maximing dx9 performance they can get...

Just my opinion...

lokeshray
02-16-04, 11:51 PM
Originally posted by -=DVS=-


Well obviouly you do not follow simple facts how for example AMD with less megaherz beat up Pentium 4,
same for Radeon it does more with less , example Radeon 9700Pro core and memory speed are nearly twice as small compared to NV30 , and yet it R9700Pro wins !



Cheers :cool:
P.S Hellbinder already said R400 went to R500 project :o just as i said. [/B]


i clearly knw why athlon xp is faster than pentium 4, and why sometimes pure meaghertz doesnt matter, but in case of the nv30, it lost out to radeon 9700 pro inspite of faster memory because of the 128 bit bus it had. this bus was peanuts as compared to the massive 256 bit bus the radeon 9700 pro had.

but this changed when the nvidia engineers worked on the nv30,gave it a 256 bit bus( and lots more) and named it the nv-35 and then we saw how the R3XX's finally found their match.
the nv-35 had a 256 bit bus , nv35 has much faster memory and hence lots of bandwidth.

more bandwidth translates into better performance at higher resolutions and with anti-aliasing .

the nvidia cards geforce fx 5700 and geforce fx 59xx have always sported faster memory and hence more bandwidth.

the radeon 9600 pro and radeon 9600 xt have memory clocked at the same speeds.
dats why the radeon 9600 xt gets its a$$ whipped by the geforce fx 5700 ultra.

in fact there was a radeon 9800 pro 256 mb from sapphire that had 2.5ns(or is it 2.8ns) hynix memory, but when the reviewers tried to overclock this memory , there were artifacts indicating that the PCB couldnt support high memory speeds.

besides, the overdrive featture is totallly lame ass and a waste of time for the driver team who spent writing drivers to enable it...

looks like history is repeating itself.
both the nv40 as well as r420 wil have 256 bit buses.
the r420 will have memory clocked at 1ghz
nv40 will have memory clocked at 1.4 ghz
the one with the higher bandwidth will win.
its been tough for ATI, half life 2 getting cancelled, as a result of which they missed out the opportunity to make HUGE sales during the holiday season.
maybe ATI is trying to appease its OEM partners. well maybe thats where the money is.
after all, even if an ATI card peforms slightly slower than an nvidia card, ATI will still a lot more money if they sell a more of cards than nvidia. Thats why it seems like ATI is trying to appease its OEM partners like DELL for example.
a majority of the population cannot fix a graphics card by themseleves into their PC.

now nv40 gonna be packaged with doom3(rumored).
i really admire ATI, but cant help admitting that the year 2004 may be very tough for them as far as hanging on (or regaining) to the performance crown is concerned.

i agree that the delay of directx 10 may have been a factor.

ellroy80
02-16-04, 11:59 PM
Originally posted by MuFu
Over-overkill, perhaps? Like microwaving a hamster for 30 minutes then running it over with a truck.

MuFu.

Nice Mufu, nice. Can I use that for my sig? (if I ever get around to setting one up)

EMunEeE
02-17-04, 02:53 AM
Originally posted by ClyssaN
I think the two main reasons were:

1: No dx10 for a good time.
2: The timeframe for the r400 was not achievable, that with the dx10 delay, the better choice was aiming for the maximing dx9 performance they can get...

Just my opinion...

You can best believe the original R400 was going to put a 1st round KO on the NV40...heck the R420 will do it too. :p