Was the 9500 rushed?

[Lezmaka]

Lol, i assume they must have at least a few, but they seem to have wanted to rush this thing out because they didn't even have any boards designed for the 9500 ready to give out to (p)reviewers.

Anand said the board he got was a 9500 Pro stuck on a 9700 board with 1/2 the memory disabled.

HardOCP said it was an underclocked 9700 Pro with one memory channel removed.

Either way, they either didn't have any boards to send out, or didn't want to (which would be really odd).

What do you guys think?

[StealthHawk]

edit: it is too late at night to read clearly

hmm, i would say that the R9500Pro needs some driver work. it does extremely poor in some situations. 5.4fps in Commanche with 16x AF and 4x FSAA when the GF4 is churning out fps in the 20s?

but it was expected that the 9500 wouldn't be available in large quantities, i think.

[Lezmaka]

Both Anandtech and HardOCP have posted their articles on it.

http://www.hardocp.com/article.html?art=Mzc0

http://www.anandtech.com/video/showdoc.html?i=1735

[jbirney]

Well Rushed? No I dont think so. The 9500 was on the road map way back when we saw leaks of it back in May. Why they did not have any final samples of it then? I dont know really. Also keep in mind that the 9500 only has 64mb so some of those benchmarks use more texture memory then will fit on a 64Mb board. I dont trust reviews when the downclock or use hacked version. Not saying they did anything wrong, but I am gonna wait till I see final shipping samples reviewed.


The nice thing is that it will push DX9 into the main stream even though its not a GF4 killer. with its MSR of aroun $179 I am sure you will see it priced on-line for less and it could be a good thing to get DX9 games here sooner.

[kmolazz]

well, i don´t get it


i´ve read both articles, and they give the opposite idea. The [H] shows the 9500pro being as fast as a ti4600 and a lot faster with AA and AF ( and the tested card only as 64MB ). The Anands article shows that the 9500pro can barely compete with an ti4200, although it is faster with AA and AF. Well wich one is it??


and why don´t the benchmarks scores match?? .. i cound´t find the system description on the anands article though..



As to if the card as been rushed out, i don´t think it really was, although it should have been done better. None of the articles is about the 9500 non-pro card, which is the one that should start shipping as of now... like the 9700 non-pro. The 9500 pro, will only start shipping in a month ( acording to ATI's official press release ), so i think there is still time for them to show the real card..

[Joe DeFuria]

Oddly,

Anand and Hard OCP have very different impressions / benchmar scores.

Basically, HardOCP has the 9500 pro beating the 4600 in most cases with Aniso and AA.

And anand has the 9500 Pro in between the 4200 and 4400?

Something ain't right. Though it appears to me that most of the benchmarks Anand used (except for UT2003 Flyby) are CPU limited. And in UT 2003 flyby with AA and aniso, Radeon 9500 Pro does beat the 4600.

I was expecting the results in non CPU limited cases to be more in line with what Hard OCP showed: that the 9500 Pro would meet or exceed the 4600 in AA and especially aniso cases.

EDIT: Looks like we have another review with numbers at Toms Hardware:

http://www.tomshardware.com/graphic/02q4/021024/index.html

And it looks like Tom's agrees with Hard OCP: 9500 Pro is a clear winner compared to the GeForce4 Ti. About the same non AA / non Aniso performance as Ti 4400, but clearly better than even the 4600 with AA / Aniso.

Where did all the Quake3 benchmarks suddenly go? That at least would be a good fill rate test....

[MtViewGuy]

I think I know the primary market for the Radeon 9500 Pro: OEM's.

This would be the perfect card for large system builders like Dell and Gateway to put together high-performance gaming systems without paying a serious premium for the Radeon 9700 series, since the card will likely cost around US$200 more or less. It would not be appropriate for the very high-end gaming computer market (e.g., Alienware, Falcon Northwest, etc.) because the high-end market will want the fastest cards regardless of price.

It's akin to the market for the GeForce4 Ti4200--the primary market is for large OEM's.

[Lezmaka]

HardOCP ran both cards using 2XAA and 8XAF, while Anand used 4XAA for both, and the highest AF level for both (8X for GF4's and 16X for 9x00's)

Could that be enough to kill the performance of the 9500's?

[Mono]

go look again. Hard ran some test with 4x aa and others with 2x. Anyways, i'll wait tell they have some retail boards instead of these underclocked/feature hacked 9700's. Non-pro 9700 looks pretty sweet eh? Assuming it overclocks good and for a hundred less :O

[StealthHawk]

so would someone like to explain why we don't have actual cards tested at the "launch?" seems a little fishy any two ways you look at it.

[Bigus Dickus]

I have one possible explanation: The cards tested were the "9500 Pro" type cards. We've been told by the inq (or somewhere else maybe) that the 9700 and 9500 will be released first, followed by the 9500 Pro a while later. My guess is that, referencing the above, there simply aren't 9500 Pro boards yet to demo, and ATi didn't want to let reviewers use the 9500 because it is only a 4 x 1 card, and the performance relative to the GF4 Ti series wouldn't have been that impressive.

I haven't read the reviews yet either, so I could be way off! :)

[StealthHawk]

Originally posted by Lezmaka
Anand said the board he got was a 9500 Pro stuck on a 9700 board with 1/2 the memory disabled.

HardOCP said it was an underclocked 9700 Pro with one memory channel removed.

which is exactly the same thing and exactly what happened.

it's time to put this issue to rest now that i scrounged a few mins to look at the non-benchmark part of the review.

straight from AnandThese aren't the boards you'll get once the Radeon 9500 Pro starts shipping since ATI actually didn't have Radeon 9500 PCBs ready in time. Instead ATI mounted a Radeon 9500 Pro GPU on a Radeon 9700 Pro board; as you can guess, this doesn't work perfectly since the Radeon 9700 Pros are wired for a 256-bit path from the GPU to main memory.

The result of this is that only half of the 128MB of memory can be used (only 128-bits out of the 256-bits wired are used) so all of our tests are with a 64MB Radeon 9500 Pro, not the 128MB version that will be shipping.

Tom's Hardware said a simliar thing, and i didn't look at [H]'s review besides the benchies.

[Nv40]

well Jb and Mr biggus did you have any more doubts
about the benefits of a second TMU ? :)

http://www.tomshardware.com/graphic/02q4/021024/ati-11.html

In the single-texturing tests, the ATI 9700 boards are able to use their 8 pixel pipelines to their full advantage. It's a somewhat different picture with multi-texturing, because only one texture per pipe and cycle can be calculated, while each of the four pipes of the GeForce 4 Ti can process two textures per cycle. What's bewildering is that the 9500 PRO falls behind in this test, because, like its bigger siblings, it also has 8 pixel pipes. So the results are in stark contrast to the official specs, which claim that it has double the fill rate of the GF4 Ti 4600! Here, it seems that the memory bus, reduced to 128-bit, makes itself felt, and this leads one to make some interesting assumptions about the architecture of the R300 memory interface. On the other hand, this could be due to the drivers not having been optimized for the 9500 yet.



Believeme ,you will see higher diferences in
multitexturing games and in 3dmark in highers resolutions .
The Nv30 should not have problems to beat the radeon9700pro
in multitextures benchmarks ,even if it has exactly the same
hardware specs of R300 but with a second 8x2 texture pipeline. :)


i think the best bang for buck still is the Geforce4 ti200-8x
as low as $130 .. some cards in the market ships withs memory
easily overclockeable to Geforce4 ti4600 levels and even surpass it a bit in when no overcloking the expensive one .. :)


and finally Toms say in a review the true about Anisotropic Filtering between Ati and Nvdia ,, is comparing apple vs oranges..

http://www.tomshardware.com/graphic/02q4/021024/ati-08.html

ATI plays its trump card in anisotropic filtering. As opposed to NVIDIA, ATI does not filter all textures with the set filter levels, and instead, decides on how thoroughly each texture is filtered depending on the user's viewing angle to the texture. NVIDIA claims that ATI gains in performance at the cost of quality. In practice, however, hardly any difference can be noticed. Interestingly enough, ATI has recently begun calling its filtering technique "adaptive," or "adaptive anisotropic filtering."


There is no fair way to compare in benchamrks
ATI vs Nvidia anisoF methods.. i agree is not fair to
compair 8x in Nvidia vs 16x in ATI ,because ATI do more aniso
in some angles . (the irony here is that is the radeon 9700 switch from 16x to 4x from ~10 to 35 angles .more or less.. )

yes Ati surely have the Radeon9700 and 9500 has the
BEST IMAGE QUALITY/practical method for games ,i agree .
even i hope Nvidia employ something like this in their future
cards .but for GOD sake .. LABEL that adaptive method
as highquality/performance ... or something like that ..
and let gamers choose to use a full 16x/16x anisof without
ANY decrease in any angle in IQ .

i dont think is Impossible for ATI engineers to design a card
that do an aniso all the time ,like Nvidia ..
yes ,its takes huge! performance , but aleast you have the choice to not use it , a and choose better
and better performance aniso the in the control panel .

again Ati have the overal -all around better imagequalityl in anisof ,the diference from the drop in image quality
probably is not noticeable , (in the same way for some people is not noticeable to play beetween 16bits vs 32bits ) .

but , does that means because it is noticeable to most people
in actual gaming is is FAir to benchmarks card A in 16bits vs card B in 32 bits ?

just remember that before radeon9700pro
it was fair to 95% of the hardware sites to compare
anisoF benchmarks between radeon2 and
Geforce4 .. the first one switch
from On to off in not flat angles. :rolleyes:
and ATi takes takes all the flowers and congratulations
from all hardware sites..


So as i said before ,i really think that while the Nv30 will
be faster than the radeon9700pro ,Nvidia will have
a bit trouble in in benchamrks if they deliver Anisof at 16x .
because it will be again Apple vs oranges ,
one card will lower the quality in some angles and the other not.

i really hope that this "adaptives" algoritms from ATi dont switch
to Antialiasing too . because it will make imposible to fairly
compare boths cards in the future . ;)

my 2 cents..

[Mono]

Well, ati does let you use both performance an quality anisotropic filtering. Truthfully, from first hand expierence of course, i've found that 16x af on my 8500/9700 have looked far better then some of my friends gf4's @ 8x af. Now, screenshots sometimes tell a diffrent story at diffrent angles etc etc etc. But I really feel the performance increase is worth the marginal quality loss i'll never even notice in game anways. Just my 2 cents though :D

[Nv40]

update!
a year later still sites benchmarks apple vs oranges

look by yourselves ..

http://firingsquad.gamers.com/hardware/9500/page14.asp

i can hear many rage3d members ..
"WOW!! the radeon2 is faster than the the RAdeon9700Pro my God ... " :rolleyes:

millions of future customers ,who doesnt have time or know too much about hardware like me ,will eat that ;)

http://www.digit-life.com/articles/gf4/index6.html

i like that site ,dont take me wrong ,but usually
Hardware reviewers expect that what their are reviewing
doesnt do any magic or tricks ,to be ahead of the competition.

it is an embarrassment ,that the help of Russians
was needed to teach western side of the world
How to review a video card . Most sites only
cares above numbers and never care to
closely to experiement if there is really! a diference
in image quality.

thanks God ,now more reviewers are paying alittle more
attention to the image quality ,good for gamers .

the above REview ->Clearly shows what i have been
saying since years !



1)It is not that ATI ANisotropic FIltering looks bad while playing
in games ... in Fact near nobody noticed anything wrong
with Ati anisoF method in the old Radeon2..
even 3dfx gamers believed glide was superior quality
than 32bits.. :rolleyes:

if its looks good for you ,better for you ..

2) its is not that it is not a "smart" way of doing ANiso
without losing performance .. its is a good adaptive method
for getting the best of quality /without too much "noticeable"
image quality drop .. as we see above
the radeon2 kicks the radeon9700pro without noticeable
decrease in image quality in gaming .. ;)

3) it is that it confuse reviewers and tells
not all the truth for harcore gamers who kill anyone
for the "best" Antialiasing or Aniso card .

because most casual gamers dont notice the diference
between anti aliasing 4x to 0x in a quake3 deathmatch
doesnt means its is a good thing to label a video card
as capable to do in hardware 4x antialiasin ,when
the truth is not that .

im fine with the Radeon9700pro ANisoF method today ,its looks
great ,i have seen it , but "great" is relative , to the point of the observer . and while its looks better in most scenarios than the 8X aniso of Nvidia today ,it will surely will look no so great compared with an 16xaniso that is irrevelant of the angle you are looking .. ;)

and i agree that there is not a right way to do Aniso
or antialiasing or pixel shaders in the future .
Every company decide which way is the right for them .
However what it is not right is to market something
like a hardware feature that works in "X" way ,when in reality
it works in "y" way ,and the above links prove that .

I think it is Far better , when you buy something and
you dont have to be nervous for any half-true specs
advertised by the video card .

Far better ,no matter if you cant notice the diference between
green or black ,agp4x vs agp8x ,16bits vs 32bits , or 60hz vs 100 hz or AF/AA 16x vs 8X or 128bits of the Nv30 vs 96bits radeon9700 . because you deserve your new hardware to work
exactly as advertized ,you have paid for that ..

[MuFu]

It don't think it is suprising that ATI can't supply 9500 and 9700 cards since they aren't actually making them. I heard that the 9500 Pro was due to be launched slightly later, so it's understandable that they don't have a production board at this stage.

MuFu.

P.S. Those PCBs they are using aren't absolutely the same as the 9700 Pro ones, BTW. They are a variation of some kind - different PN suffix. Maybe an ES designed for experimentation with different controller/framebuffer configs.

[jbirney]

NV40,

for once I agree with you about apple to oranges. Its been this way in all AA benchmarks on the GF3/4 and R9700. Reviews never seem to noticed the differnece in IQ quality in SS vrs MS methods. It was perfectly fine for them to compare only the numbers with out really seeing what each was doing. But they compared the 8500 to the GF3/4 cards in AA and always decalared the GF4 the winner simple because it was faster even though it was doing less to enhance IQ. Oh well....

Note: I am not saying that the GF4 AA was worse as its all subjective. I am just stating that the two methods provided for different results and those were NOT factored into most of the reviews.

[StealthHawk]

Originally posted by MuFu
It don't think it is suprising that ATI can't supply 9500 and 9700 cards since they aren't actually making them. I heard that the 9500 Pro was due to be launched slightly later, so it's understandable that they don't have a production board at this stage.

shouldn't sites have been testing the cards at R9500 speeds and not R9500Pro speeds then?

[Bigus Dickus]

Originally posted by StealthHawk
shouldn't sites have been testing the cards at R9500 speeds and not R9500Pro speeds then?

9500 only has 4 pipelines, Pro has 8. They tested the cards in the 9500 Pro configuration, for whatever reasons may be.

[StealthHawk]

Originally posted by Bigus Dickus
9500 only has 4 pipelines, Pro has 8. They tested the cards in the 9500 Pro configuration, for whatever reasons may be.

well, yes. i meant that, if regular 9500s would appear first, wouldn't it make more sense to test those than the 9500Pros that appear later(if that is indeed the case)?

well if the R9500Pro isn't ready i guess that means there are no working 9500s either. i think we all know what this launch was about :D

[Mono]

uhhh.. didn't the press release say they should be in stores by next month?

[StealthHawk]

Originally posted by Mono
uhhh.. didn't the press release say they should be in stores by next month?

and that still has nothing to do with the fact that ATI launched the cards when they didn't even have final silicon they could give to reviewers to test.

just another smart move to get people more insterested in their products versus nvidia's.

[ErrorS]

Originally posted by StealthHawk
and that still has nothing to do with the fact that ATI launched the cards when they didn't even have final silicon they could give to reviewers to test.

just another smart move to get people more insterested in their products versus nvidia's.

well they had the R9500Pro GPU(? whatever) ready .. ATi is doing this 30 day before on shelves launch thing with all their cards (except R9000 ).. its working

i think PCB design.. or even just finalizing things is a last minute thing .. im sure ATi just wanted to keep with the 30 days thing (although I dont understand the pro PReviews first) .. They didnt wanna go to their manufactuars to tell them "Make me 4-5 R9500 boards for previews" ..

dunno what happened.. but this gen of cards ATi has done some insane **** with marketting .. its working too ..

[StealthHawk]

Originally posted by ErrorS
well they had the R9500Pro GPU(? whatever) ready .. ATi is doing this 30 day before on shelves launch thing with all their cards (except R9000 ).. its working
just how do they have it ready if there are no boards :confused:

conside that most launches of high end products are nothing but paper launches, and reviewers have hardware samples of those.

IMO(and i don't see how there's any evidence to contradict this) the R9500Pro launch was premature. if there are no boards ready, then it obviously means we won't be seeing the chips in large volumes OR that the chips won't be seen for a little bit.

[ErrorS]

Originally posted by StealthHawk
just how do they have it ready if there are no boards :confused:

conside that most launches of high end products are nothing but paper launches, and reviewers have hardware samples of those.

IMO(and i don't see how there's any evidence to contradict this) the R9500Pro launch was premature. if there are no boards ready, then it obviously means we won't be seeing the chips in large volumes OR that the chips won't be seen for a little bit.

im guessing it's ready .. all they gotta do is slap the GPU on a 8(?) layer PCB add some memory and they'll be good to go .. could be wrong about this, but to me the GPU seems to be the most important and time consuming thing when it comes to developing video cards.. design for the actual board is probably already done, they just dont have any in production yet (not even 4-5 for previewers) ..

So.. in short.. I'm guessing they just started the R9500 (Pro?) production a few days ago.. or if they havent yet, they are soon .. see what I'm saying?

Could be anything, I really think them not having the R9500's board for previewers isnt important at all... They just werent in production yet .. and if they arent making them themselves, then there is no way to get a finished product to previewers unless they wait a couple of weeks.. and waiting is never good..

Don't think we'll see R9500s for a month ..

I dunno if i got to my point.. dont care.. been playing NFS:HP2 all night and need some sleep :p