ATi R420 & R423 specifications 19 April 2004/from Leonidas/page 1 of 1 we had themselves already once in February with that at that time admitted information to the coming ATi chips R420, R423, R480, RV370, RV380 and RV410 employed. Whereby the information specified in this article still applies to the chips R480, RV370, RV380 and RV410 - just as also that ATi will set 2,0 with its coming NextGen Riege further on the pixels Shader.  
 
However at the number of pixel pipelines of the R420 and R423-Chips in the meantime some did, the number of "8", specified in this older article, tunes now - well-known-measured - by far no more. Since now however in the last days ever more information to R420 and R423 to the light came, we as well as want to use the opportunity, in order the picture of the R420 and R423-Chips hereby almost-moved to specify. ATi's R420-Chip will now to be inferred as from the News of the past days begun in three variants, whose main difference will be the number of active pixel pipelines: Radeon X800SE with 8 active pipelines, Radeon X800 pro with their 12 and Radeon X800XT with their 16.  
 
With all these versions comes the same R420-Chip with original 16 pixel pipelines to the employment, ATi deactivated thus for the variants Radeon X800SE and Radeon X800 pro in each case a part of the pipelines. However it will give also still another clock difference, so that the following specification table can be provided at present compared with the GeForce 6800/Ultra maps of nVidia:  
 
Radeon X800SE Radeon X800 pro Radeon X800XT Radeon X880XT GeForce 6800 GeForce 6800 Ultra code name ATi R420 ATi R423 nVidia NV40 production 180 Mill. Tr. in 130nm 180 Mill. Tr. in 130nm 222 Mill. Tr. in 130nm technology class DirectX9 Shader 2,0 DirectX9 Shader 2,0 DirectX9 Shader 3,0 geometry pipelines? (probably ? (probably ? (probably ? (probably ? (probably 6) 6 pixel pipelines 8 12 16 16 12 16 Shading units per pixel pipeline? (at present completely unknown, the R3x0-Chips has a TEX unit + an arithmetic general-purpose unit) a SFU/MUL/TEX unit + a WAD/CDot unit Splitfaehigkeit for arithmetic units?  
 
(at present completely unknown, the R3x0-Chips has 4:0 (no Split) and 3:1) 4:0 (no Split), 3:1 and 2:2 chip clock 450 MHz 475 MHz 500 MHz 500 MHz? (estimated 350 MHz) 400 MHz storing act 400 MHz 475 MHz 500 MHz 500 MHz? (estimated 400 MHz) 550 MHz memory binding 128 bits 256 bits 256 bits 256 bits 256 bits 256 bits memory equipment 128 MT GDR 256 MT GDDR3 256 MT GDDR3 256 MT GDDR3 128 MT GDR 256 MT GDDR3 chip interface AGPx8 AGPx8 AGPx8 PCI express x16 AGPx8 AGPx8 Launch in the middle of June 4 May 31 May 14 June? 14. April market entrance? (probably July) May? (probably June)? (probably July)? (probably June) May starting price 299 $ 399 $ 499 $ 499 $ 299 $ and/or 349? 499 $ and/or 549?  
 
With the GeForce 6800 some estimations came to carrying, which we expressed here, since their exact clock rates are not certain at present yet. Also with the Radeon X8x0 maps had to become estimated some at present still missing data - which was marked on our part however in each case. The clock rates of these maps specified by us come of with the fact the all-newest information whereby it is always possible that up to the respective Launch in these in orders of magnitude of ± 25 MHz still changes something.  
 
Largest problem for achievement before evaluation is - beside the point that these data are to come from ATi, but nothing the despite it is not official - the absence of information for the organization of the pixel pipelines with the R420/R423-Chips and/or the Shader units which are used there. The NV40-Chip sets here on a SFU/MUL/TEX unit as well as a WAD/CDot unit per pixel pipeline (detailed view of the NV40-Pipeline), with the R420 is dependent we on speculations.  
 
Naturally one can assume that ATi - there the R420 a R3x0-Chip in 130nm is in the long run seen in principle - took over here the R3x0-Technik and so that also with the R420 on a TEX unit as well as an arithmetic general-purpose unit per pixel pipeline sets. If this should apply in such a way, the NV40-Chip would have more power when pure counting. If however constantly texture instructions come, the NV40-Chip would have disadvantages - zumindestens as said, if one assumes ATi took over the R3x0-Logik of the Shader units with the R420.  
 
Unfortunately the supposed Shader performance cannot be compared yet with the existing data, but zumindestens one can do this concerning the conventional theoretical filling rates and ranges. To mark however that in particular the filling rate lost at present by the Shader performance ever more their once enormous meaning, it would be in no more case than exclusive characteristic of the raw achievement of a diagram chip to thus apply can. Nothing the despite here the overview: Theoretical filling rates and ranges NEXT gene chip filling rate Multitexturing (GT/sec) memory range (GB/sec) Radeon X880XT (499 $) 8.0 GT/sec 30.5 GB/sec Radeon X800XT (499 $) 8.0 GT/sec 30.5 GB/sec Radeon X800 pro (399 $) 5.7 GT/sec 29.0 GB/sec Radeon X800SE (299 $) 3.6 GT/sec 12.2 GB/sec GeForce 6800 Ultra (499 $) 6.4 GT/sec 33.6 GB/sec GeForce 6800 (299 $) 4.2 GT/sec (estimated) 24.4 GB/sec (estimated) 1.2 GT 5 GB 2.4 GT 10 GB 3.6 GT 15 GB 4.8 GT 20 GB 6.0 GT 25 GB 7.2 GT 30 GB 8.4 GT 35 GB 9.6 GT 40 GB a surely high-interesting picture - out which one should derive rash conclusions however in no case over the material efficiency of the 3D-Beschleuniger releasten yet (the material achievement results like said not only from the filling rate & range, but also the Shader performance as well as the efficiency of the total thing).  
 
Rather there one can be inferred with a partly clearly higher raw power than this start from this picture rather only that ATi has quite the chance to exist against the NV40-Chip. In the apron it had finally given doubts whether ATi with warming the past R3x0-Technologie up would have a chance with the R420-Chip opposite nVidia's NV40, which now however is obviously unfounded.  
 
However only once only the Radeon X800 pro will rise against the GeForce 6800 Ultra into the ring, since this represents that variant from the R420-Chipserie, which first (at the beginning of May) will be releast. With one for the nVidia chip somewhat the ATi chip would have to put better raw achievement a higher efficiency like the NV40 to the day, in order to be able to exist against the GeForce 6800 Ultra.  
 
We expect therefore that the Radeon X800 pro the new nVidia leading product not yet to strike to be able - which does not have the Radeon X800 pro in addition, finally is it concerning the price around 100 dollar/euro lower settled.
 
The "genuine" duel between ATi and nVidia arises thus only with release of the Radeon X800XT, speaks only for the end of May then begins - completely roughly regarded - same basic architectures (with in each case 16 pixel pipelines and 256 bits memory interface) against each other, with which on the one hand the accurate clock rates will give and on the other hand the efficiency of the respective pipelines the excursion.  
 
Of the raw achievement some speaks for the Radeon X800XT in this comparison, but by those points Shader performance and total efficiency at present did not admit as well as by the higher memory range of the GeForce 6800 Ultra can this nominal advantage problem-free also in the opposite turn itself. Well possible that the two respective leading products differ then again once only around few per cent; -).  
 
For most users however less GeForce will be 6800 Ultra, Radeon X800 pro, Radeon X800XT as well as Radeon X880XT interesting, but rather the more favorable NEXT gene chips GeForce 6800 and Radeon X800SE. Due to many missing data to the GeForce 6800 however at present is this no really safe comparison. However our acceptance of 350/400 MHz if Takraten should be halfway confirmed with this map, we see the nVidia solution in this market of the 350-Euro-Beschleuniger somewhat before the ATi solution. The reason lies in their 12 pixel pipelines opposite their 8 with the Radeon X800SE, which should probably give the excursion for the nVidia solution with approximately same clock rates.  
 
Thereby the comparison of GeForce 6800 and Radeon X800SE with the HighEnd remarks of the respective predecessor series is interesting. The GeForce 6800 is past thereby from the raw achievement only scarcely behind the GeForceFX 5950 Ultra, due to the clearly better Shader performance of the NV40-Chips might it the GeForceFX 5950 Ultra however probably from the material achievement even over wings. The Radeon X800SE might be braked something by their only 128bittigem memory interface, but with optimal process also here material achievements are possible zumindestens in the proximity of the Radeon 9800XT.  
 
These two more favorable NEXT gene chips GeForce 6800 and Radeon X800SE could quite replace thus the past HighEnd Grafikkarten GeForceFX 5950 Ultra and Radeon 9800XT. It can recapitulatory be said that ATi has warmed up "technology its chances very probably with one only" to reach or strike with the R420-Chip the achievements of the NV40-Chips of nVidia. Due to some missing data R420 can be only courage-measured at present over the exit of combat NV40 vs., but even with complete data there could only be a final dissolution of this question in the material performance comparison of these two NEXT gene chips. Off its the two more favorable NEXT gene chips seem to become in the form of GeForce 6800 and Radeon X800SE quite interesting products with good price/achievement relationship, these should thus further attention be given

ils ont même été validés par futuremark.