ATI Radeon 9600 PRO Review

Discussion in 'Beyond3D News' started by Dave Baumann, Apr 16, 2003.

  1. Dave Baumann

    Dave Baumann Gamerscore Wh...
    Moderator Legend

    Joined:
    Jan 29, 2002
    Messages:
    14,079
    Likes Received:
    648
    Location:
    O Canada!
    [​IMG]
    Having announced the Radeon 9600 along with Radeon 9800 and 9200, ATI are finally ready to unveil the performance of Radeon 9600 powered boards. Today we review ATI's Radeon 9600 PRO 128MB and also we look at an early version of Sapphire's Atlantis Radeon 9600 board. <blockquote>"Whilst the R300 core features eight DirectX9 pixel pipes and four Vertex Shader units, despite the smaller silicon process not everything made it across to RV350. Essentially RV350 is half an R300 core, with only four pixel pipes and two Vertex Shading units. The reductions in size, though, means that this is able to scale to higher speeds than the R300 could, with the 9600 PRO product shipping at 400MHz core speed, thus offsetting some of the fill-rate and triangle rate advantage that R300 has in its 9500 PRO guise."</blockquote>Read the full review here
     
  2. Arun

    Arun Unknown.
    Moderator Legend Veteran

    Joined:
    Aug 28, 2002
    Messages:
    5,023
    Likes Received:
    299
    Location:
    UK
    Good review, Dave!

    Sounds like the 5600 Ultra, and maybe even the regular, will do just fine against the Radeon 9600 Pro, hmm?

    The 5200 Ultra is in big trouble though, if the 9600 Regular really beats a 9500 and costs $129...
    Sounds like ATI really caught them off guard, I guess...


    Uttar

    EDIT: After rechecking some numbers...
    The 9600 Pro is still about 5% faster than the 5600 Ultra, it would seem.
    So it's still a problem to the 5600 Ultra, but it doesn't completely own it as the 9500 Pro did.
     
  3. BenSkywalker

    Regular

    Joined:
    Feb 6, 2002
    Messages:
    823
    Likes Received:
    5
    Very nice review Dave :)

    The most shocking thing to me about the entire review was that you have the first review I've seen anywhere without losing any of the typical quality of B3D's reviews. Most impressive :)

    Now, how did you manage to pull it off...? :shock:
     
  4. jandar

    Newcomer

    Joined:
    May 27, 2002
    Messages:
    225
    Likes Received:
    1
    Location:
    JVille, FLA
    ok, the 9600 Pro gets released on my Birthday

    I already have a 9700Pro.


    anyone feel like giving me some inside info on when I can pickup a M9600 in a laptop?
    Or any Ati employees feel like being nice to a guy?

    BTW, kick ass review Dave. B3D rocks as usual.
     
  5. Tagrineth

    Tagrineth SNAKES... ON A PLANE
    Veteran

    Joined:
    Feb 14, 2002
    Messages:
    2,512
    Likes Received:
    9
    Location:
    Sunny (boring) Florida
    I'm happy I got my 9500 Pro then ^_^

    And... why does 9600 take such a brutal hit from AA at 1600x1200, despite needing LESS memory thanks to no hierarchal Z?
     
  6. Evildeus

    Veteran

    Joined:
    May 24, 2002
    Messages:
    2,657
    Likes Received:
    2
    Nice review.

    Seems that 9600 pro has just a bit more performance than a 5600 ultra (hopefully due to the last increse in the chip frequency) and a 9600 a bit less than a 5600 :?

    BTW, where did you get the conclusion that :
    I've only found that statement over here :?:

    http://www.hardware.fr/articles/461/page1.html
     
  7. mr

    mr
    Newcomer

    Joined:
    Oct 7, 2002
    Messages:
    143
    Likes Received:
    0
    Very nice review as usual, Dave.

    Just two questions:
    1) Where's that Splinter Cell benchmark ? (I assume you didn't have enough time)
    2) Did you mention anywhere in your review that both the 9600 and the 9600 Pro "lack" an external power connector ? (I'm sorry if I've missed it)
     
  8. martrox

    martrox Old Fart
    Veteran

    Joined:
    Feb 6, 2002
    Messages:
    2,065
    Likes Received:
    16
    Location:
    Jacksonville, Florida USA
    Nice review, Dave. It's pretty much what I expected, and just goes to show what really great deal the 9500Pro's are. While the GFFX5600's are fairly competitve with the 9600Pro, looks like the standard 9600 should clean clock on the 5200's.

    EDIT:ACK! After looking at [H]OCP's just published review, where the 9600Pro is tested against both a 9500Pro AND a 5600Ultra, nVidia's in BIG TROUBLE, again!
     
  9. demalion

    Veteran

    Joined:
    Feb 7, 2002
    Messages:
    2,024
    Likes Received:
    1
    Location:
    CT
    Nice mention of the AGP issue, that's a good heads up to readers. First time I recall such a detail being covered.

    The 1600x1200 performance drop seems similar to what ATI has done in the past by having memory utilization issues with the initial driver release for a card. I'd be interested in the fill rate graphs across resolutions if the drivers were fooled into thinking it was a 9500, to see if the issue was an untweaked driver path.

    It will be interesting to see how the 5600 non Ultra/5200 Ultra pricings react to the 9600 non pro pricing.
     
  10. Ostsol

    Veteran

    Joined:
    Nov 19, 2002
    Messages:
    1,765
    Likes Received:
    0
    Location:
    Edmonton, Alberta, Canada
    Nitpick: shouldn't the size be 9.5x9.5 mm? 95 mm is 9.5 cm. . . :shock: That'd be quite a big chip. . . :wink:
     
  11. MuFu

    MuFu Chief Spastic Baboon
    Veteran

    Joined:
    Jun 12, 2002
    Messages:
    2,258
    Likes Received:
    51
    Location:
    Location, Location with Kirstie Allsopp
    Excellent review as usual, Dave. 8)

    60M transistors is about the same as an R200, I think.

    http://www.beyond3d.com/reviews/ati/rv350/index.php?p=19

    Those results make it pretty obvious (?).

    Consider the very nature of how a Hi-Z buffer works (being effectively a low-res Z-buffer that needs to be on-die for "free" Z-checking) and it's quite easy to understand why they have most likely ommited it due to cost considerations. The necessary transistor allocation must be pretty high.

    MuFu.
     
  12. Dave Baumann

    Dave Baumann Gamerscore Wh...
    Moderator Legend

    Joined:
    Jan 29, 2002
    Messages:
    14,079
    Likes Received:
    648
    Location:
    O Canada!
    Hi Guys, Glad you enjoyed the review.

    OK, I've done some very very quick analysis on from the NV31 and RV350 Excel benchmark sheets, the NV31 numbers are those that were in the NV31 Preview. That used application settings so IQ is comparable, and the same system settings - the 5600U was on the 42.69 drivers though. Here's a %'age performance difference of the 9600 PRO and 9600 from the 5600U.

    Code:
    Max Payn640x480 800x600 1024x7681280x9601600x1200
    9600 PRO     10%     30%     46%     50%     38%
        9600      2%     -1%      0%      3%     -7%
    
    UT Flyby640x480 800x600 1024x7681280x9601600x1200
    9600 PRO     22%     46%     53%     50%     39%
        9600     11%     10%      3%      1%     -7%
    
    UT Bot  640x480 800x600 1024x7681280x9601600x1200
    9600 PRO      6%      8%     20%     48%     50%
        9600      6%      7%      9%      9%      2%
    
    RTCW    640x480 800x600 1024x7681280x9601600x1200
    9600 PRO     -5%     -2%     14%     56%     65%
        9600     -5%     -3%      7%     19%     15%
    
    SS:SE   640x480 800x600 1024x7681280x9601600x1200
    9600 PRO    -10%     -5%      9%     21%     16%
        9600    -12%    -13%    -14%    -13%    -20%
    Looking at those numbers it seem that the normal 9600 is still pretty competetive with the 5600U with similar IQ settings. These are quick numbers, so correct me if I'm wrong. I've not looked at AF/AA differences either.

    I'm still scratching my head here given the 5600U has such a bandwidth difference to both the 9600 and 9600 PRO, and RV350 looses its Hier-Z!

    With no sleep as usual! ;)

    Of course, the review method has to remain sort of formulaic, which is why things change gradually. However, on this occasion I have to thank ATI as I got over a week with this board before the NDA date, rather then a few days that I've had previously. I also have to apologise to the board vendors who are waiting for their reviews! :oops:

    I wanna know where I can pick one up for review!! :D

    I think the explaination is, at the moment, internal cache sizes. these are still the semi old 7.84's though, so there are likely to be a few opptimisations here and in 6X (although I wouldn't expect much, because this part just isn't built for these resolutions).

    There were a number of errors in the marketting docs that I highlighted and after a discussion with Sireric he corrected PR on a nuber of points, so I don't know if other reviews have gone by the marketting docs or not.

    Anyway, Sireric confirmed to me that there is no Hier-Z (not HyperZ, since all the other parts remain in tact, including early Z reject, just Hier-Z is gone), and this is evidenced in the Overdraw Reduction tests. The Hier-Z is actually an onchip process on R300/R350 and it eats up a LOT of transistors (this is why its disabled on 9500, because its likely to be one of the failure points) and so if you are making a smaller die part then it makes sence (from a cost point of view) to remove this.


    1.) V Busy with other reviews at the moment, so I've not been able to reasearch new benchmarks at the moment. Rev has been in contact with the developers and has been following the benchmark up, but he's got the US version wheras I have the European version which never seems to work with the benchmark patches so far - I've got to try again with the new Europ 1.2 patch yet though.

    2.) See the Package, Installation & Drivers section. :)

    Nice spot, good call! :)

    FYI - this board was clock locked, however Ichy has recently provided me with a new version of Rage3D overclocker so I might gie that a whirl tonight. I'm hearing core overclocks to 500MHz and I want to know if thats ture or not. If it is then this product may still have some headroom. It also looks like ATI are really concerned about cost and margins as well - if they can clock it further and stick 350MHz RAM on it then there's probably still a lot more to come, but the fact is they purely appear to be looking competetively.
     
  13. Evildeus

    Veteran

    Joined:
    May 24, 2002
    Messages:
    2,657
    Likes Received:
    2
    Well, i think i didn't put the right quote :D I did find a lot of review saying the same thing, but i dislike the way it was handle on B3D, or i missed somthing :?. The point is Ati is saying that the card have H-Z 3:

    Code:
    HYPER Zâ„¢ III+
    • Hierarchical Z-Buffer and Early Z Test
    reduce overdraw by detecting and
    discarding hidden pixels
    • Z cache optimized for real-time
    shadow rendering
    Lossless Z-Buffer Compression and Fast
    • Z-Buffer Clear reduce memory bandwidth
    consumption by over 50%
    But in B3D review we have the feeling that Ati doesn't claim that, and that's the result showing no HZ-3 is normal. Sure Sieric is right, but the marketing guys have more impact ;)

    I think you should have focused on that contradictory result with the marketing buzz. That's all. Good to hear that on the board nevertheless.
     
  14. mczak

    Veteran

    Joined:
    Oct 24, 2002
    Messages:
    3,015
    Likes Received:
    112
    the R9600pro is IMHO slightly disappointing. Not that I expected it to beat the R9500pro, but it's a pity ATI didn't include Hier-Z - in that case it would beat the R9500pro at least sometimes probably. About how many transistors would that feature cost? 5M?
    Though it is not really surprising (the R9000 hasn't Hier-Z, neither has the R9500), but still...
    I'm wondering what the street price of the R9600 non-pro is going to be. If it's going to be priced about the same as that "99$ entry level DX9 card" it should be a very interesting part for OEMs.
     
  15. Dave Baumann

    Dave Baumann Gamerscore Wh...
    Moderator Legend

    Joined:
    Jan 29, 2002
    Messages:
    14,079
    Likes Received:
    648
    Location:
    O Canada!
    I just got the impression that there was nothing in there to intentionally mislead people (at least, I hope not) but that there was insufficient communication from engineering to PR. I strikes me that the docs were rather hastily done and they had just converted them from 9800 PRO (for instance, it also stated that it had a 4 crossbar memory controller, which just didn’t make any sense). When I communicated this to sireric I later got a reply from Canada PR stating that these were incorrect, the docs had been reworked and that other reviewers had been contacted – however I don’t know if Canada PR subsequently disseminated that to Euro PR (which I assume hardware.fr goes via) to give to the other European reviewers. I don’t know if these docs are public yet (I haven’t looked) but given that they openly said this was a mistake and others had been contacted with the corrections I don’t see that as an issue.

    Edit: Having said that, its still listed on their site. http://mirror.ati.com/products/pc/radeon9600pro/specs.html
     
  16. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    Good job with the review, Dave!

    I have to say that the regular Radeon 9600 looks very interesting with that 325 MHz core to boot.

    At $129 at seems like a good mainstream DX9-product that might end up at around $110 fairly soon with bargain OEM's.

    OTOH since the core speed is decent the OEM's can also pack them with higher memory speed than 200(400) MHz to gain quite a bit for performance.

    Nice little chip that can almost cover NV31 Ultra to NV34 regular.
     
  17. LeStoffer

    Veteran

    Joined:
    Feb 6, 2002
    Messages:
    1,253
    Likes Received:
    13
    Location:
    Land of the 25% VAT
    Ack, that was me. Bloody auto login doesn't work in news!
     
  18. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    Thanks for the additional info, Dave.

    Yep, that's why I said 5% and you said about 40% :)
    I was comparing the RV350 4x AA & 16x Perf AF to the NV31 4x AA & 8x Balanced Aniso - it's a fair comparaison IMO, beside the AA IQ, but you can't do anything about that...

    Look at the AA scores. You'll still be scratching your head as to why the RV350 is still slightly faster, though!
    And I am scratching my head too, about it. My guess is the 64x2 memory bus: nVidia isn't used at all to having 64 bit buses. ATI is, they've done it for generations and generations now.
    So, AFAICS, nVidia's 64x2 memory bus isn't very optimized yet. Now, with the original GF3 memory bus, we've seen driver improvements - it remains to be seen if we'll get such improvements with the NV31 too, or if it's hardware-related.

    Another explanation is below.

    Hmm, might be. I'd rather guess it got something to do with AGP Texturing.

    Wouldn't surprise me if ATI emulated the Hierarchical Z buffer in memory, costing them memory but saving them memory bandwidth. Heck, I'm sure you could implement it that way: you'd sometimes only have to read 1 Z value from memory, instead of 4. Of course, it shouldn't have a HUGE influence on memory - but heck, doing 5MB of AGP Texturing/frame instead of 0 could, among other things such as cache, cause such a slowdown.

    Heck, Hierarchical Z done in memory ( instead of caches ) could even justify part of the higher performance of the 9600P over a 5600U!

    Of course, maybe I'm just plain wrong and it's only cache size - but it would be nice having a sure "no" on my theory, at least


    Uttar
     
  19. Dave Baumann

    Dave Baumann Gamerscore Wh...
    Moderator Legend

    Joined:
    Jan 29, 2002
    Messages:
    14,079
    Likes Received:
    648
    Location:
    O Canada!
    Well, most of the time you will fail the hier-z test, so with a memory solution you can end up using 3 times the z check bandwidth! the point of having it on chip is that it doesn't require any extra memory bandwith, so when its a fail (most of the time) its no loss, but when its a pass its a big gain.
     
  20. Mephisto

    Newcomer

    Joined:
    Feb 7, 2002
    Messages:
    200
    Likes Received:
    0
    You can't have "the fünfer und's weggli". ;-)

    HierZ is expensive to include, as Dave pointed out. 5 M transistors is almost 10% on a 60 M part. The Radeon 9500 PRO is just "too good". Margains are horrible, ATI never should have launched it from a business point of view. I find it odd reading all the complaints about the Radeon 9600 PRO beeing slower than a Radeon 9500 PRO, 'cause you can't buy 9500 PRO most of the time and the R9600 is going to be cheaper than that part quite soon after the launch.
     
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...