Official HD 2900 Discussion Thread

[koosjebram]

Ocworkbench got something to with 8.37 driver...

http://www.ocworkbench.com/2007/asus/EAH2900XT/g1.htm

When comparing Nvidia 's 16 x QAA mode vs Ati 's 8 x it's quite on par with GTX

[red]

We tested with the latest available testing driver from AMD which is 8.361 – and this is the shipping driver for Radeon HD 2900 XT. There is a new driver floating around but that is not entirely stable yet. We would not be surprised if AMD releases its Catalyst 7.5 monthly driver soon and that should probably see some performance improvements and fixes of known bugs. The driver that we are using now is not perfect and AMD has noted some known issues. For example, in some of our tests, the older Radeon X1950 XTX is actually faster than the HD 2900 XT – this is a driver issue and we hope AMD fixes the problems very soon.

Even though the TweakTown test showed that Nvidia owns the top end, it would've been nice to see the 8800 GTS in there.

[mascaras]

[IMG]We tested with the latest available testing driver from AMD which is 8.361 – and this is the shipping driver for Radeon HD 2900 XT. There is a new driver floating around but that is not entirely stable yet. We would not be surprised if AMD releases its Catalyst 7.5 monthly driver soon and that should probably see some performance improvements and fixes of known bugs. The driver that we are using now is not perfect and AMD has noted some known issues. For example, in some of our tests, the older Radeon X1950 XTX is actually faster than the HD 2900 XT – this is a driver issue and we hope AMD fixes the problems very soon.[/IMG]

.

drivers Avaible in ATi FTP :

Cat 8.361
CaT 8.374
Car 8.38 >>http://www.xtremesystems.org/forums/...&postcount=869

Even though the TweakTown test showed that Nvidia owns the top end, it would've been nice to see the 8800 GTS in there.

i have a BFG 8800GTS 320mb and i will receive a X2900 nxt week then i post some results to compare .

Ocworkbench got something to with 8.37 driver...

http://www.ocworkbench.com/2007/asus/EAH2900XT/g1.htm

When comparing Nvidia 's 16 x QAA mode vs Ati 's 8 x it's quite on par with GTX

nice scores in OC review

[bingo13]

im not talking of next drivers , 8.38 it was released few days ago , my friend have it for 4 days now


i dont want miracles , the actual driver its very good in 3dmarks , now i want a good driver for games , i understand its dificult and takes time ( nvidia takes 6 months to get a good performece driver (158.19/19) i just want a good game driver fast

The 8.38 drivers are an early beta 2 release, they were leaked. Those drivers have more issues with them than the 8.37.4 in several cases. CrossFire and OpenGL in Vista are messed up, IQ has been reduced in several cases also until they tune it. ATI does not want the review sites using those drivers until the final release on 5/23.

[Eastcoasthandle]

Can the red vreg (or mosfet I forget) HS from the x1900xt/xtx fit on the HD2900 XT. Because the HD 2900XT have that HS built right into the GPU HS. So if you take the GPU HS off you also take off the other HS. No Water cooling until a HS can be put on it IMO.

From looking at VR-Zone's review he was able to get 1 push down bolt on one side but had to use a blue plastic clip for the other side. That means 1 of 2 things:
-he broke the other push down bolt
-the other push down bolt doesn't fit the hole on the HD 2900XT

[mascaras]

The 8.38 drivers are an early beta 2 release, they were leaked. Those drivers have more issues with them than the 8.37.4 in several cases. CrossFire and OpenGL in Vista are messed up, IQ has been reduced in several cases also until they tune it. ATI does not want the review sites using those drivers until the final release on 5/23.

Ok thnks for advise

[deathman20]

So the "good offical" drivers will be released about 10 days after launch?

[insurgent]

Current DVi-HDMI adapters do not carry audio. This Active DVI to HDMI adaptor does. It allows full video and audio support via standard DVI output from Radeon HD 2000 series graphics cards

http://www.ocworkbench.com/2007/asus/EAH2900XT/g7.htm

[informal]

I really dislike the power consumption alot.

Oh really??
This: http://www.vr-zone.com/?i=4946&s=9 shows it consumes a "whooping" 17W than 8800GTX (not ultra!!! ).I guess it's gonna burn a hole in your wallet for a power bill if you go for a Radeon instead of a 8800GTX.

[red]

Why would you compare the power consumption of a slower part to that of a faster part? VR-Zone does have 8800 GTS numbers to compare to.

[Gorgul]

Ocworkbench got something to with 8.37 driver...

http://www.ocworkbench.com/2007/asus/EAH2900XT/g1.htm

When comparing Nvidia 's 16 x QAA mode vs Ati 's 8 x it's quite on par with GTX

They bencht the Nvidia Cards with very old drivers.

[koosjebram]

They bencht the Nvidia Cards with very old drivers.

Lol..guess they needed 5 months old drivers and 16 vs 8 X AA too make the HD2900XT look good.

I'm beginning to think we'll have to wait a month or so more to see what's the real deal with this card and mature drivers

[Noobie]

kinda off topic but when can we expect HD 2600 reviews?

Not for another 3 months or so I fear. I am worried about the amount of Processing Units (PU) they have put on it tough....

For easy calculations sake.....

PU SPEED - ATI
320 PU's * 742 MHz/PU => 475 GigaFlops (derived from VR-zone review)
120 PU's * 800 MHz/PU => 192 GigaFlops (calculated & matches VR-zone)

PU SPEED - NVIDIA
96 PU's * 1200 MHz/PU => 345.60 GigaFlops (derived from WikiPedia)
32 PU's * 1450 MHz/PU => 139.2 GigaFlops (calculated & matches WikiPedia)

If you compare the 2900XT to 2600XT calculation power, you see it matches nVidia's 8800GTS to 8600GTS almost spot on.

MEMORY SPEED - NVIDIA
8800 GTS - 320-bit & 1600 MHz -> 64 GB/sec
8600 GTS - 128-bit & 2000 MHz -> 32 GB/sec (again by calculation this is correct)

We see a 2.0x decrease in mem-power with a 2 1/2x decrease in shader power for the 8600 GTS over the 8800 GTS

Now image what ATi can do. They can put a 512-bit ring-bus on the 2600 XT, which is costly. Instead it is more likely they will put a 256-bit ring-bus on it. This decreases the mem-power by 1/2. If they slap faster memory on the 2600 XT, similar to what nVidia did with the 8600 GT we get the following....

We will see a 1.6x decrease in mem-power with a 2 1/2x decrease in shader power for the x2600XT.

The real question is then, how much difference faster mem will make. Luckily nVidia also spawned the 8600 GT. We again turn to maths:

139.20 / 113.28 => the 8600 GTS has ~1.25x as much shading power
32.00 / 22.40 => the 8600 GTS has ~1.43x as much memory power
675 / 540 => the 8600 GTS has 1.25x as much core power

Thus the 8600 GTS has more of a mem-increase then increase in other parts.

We turn to a review on http://www.guru3d.com/article/Videocards/426/16/ after which we run some number-crunching and we find out that : in the review the 8600 GTS is 1.1x faster than the 8600 GT. Why? Slack most likely. So we reach a dead end.

On Guru3D we find a comprehensive 8600GT/GTS roundup and we turn to a quote:

Yeah, you guessed it already. BTW make note of the performance from the Galaxy GTS card. Core and memory clocks are the lowest among all the GTS cards yet the score is really good (throughout most benchamarks actually). This is because Galaxy overclocks the Shader unitdomain really high. We'll get back on that later.

8600 GTS Galaxy 700/1000/1620/6029 (core/mem/shader/3D_2006)
8600 GTS BFG OC 710/1008/1512/5883

So an increase in 1.07 for the shader, results in 1.025 performance increase, this is not really linear.

We for now turn our attention to the GPU & Core OC test:
8600 GT Galaxy @ 600/800 - 58 FPS
8600 GT Galaxy @ 730/900 - 61 FPS
FPS increase : 1.05
mem increase : 1.125
core increase : 1.21
A mem & core increase does not scale so well either.

In short: the balance the videocard makers have made between mem, core and shader is so well thought out, that an increase in either shader, core or mem does not have a significant contribution to the end-performance.

Thus I conclude that with the ATI 2600 XT it is unlikely that ATI will use faster memory or core, but rather will keep the price low.

The question than becomes: how low will it be? With the 8600 GTS we have already seen that it sucks, cos the 8800 GTS 320 MB is far better value for money (less than 1.5x the price, but roughly 1.8x the performance). With the 8600 GT it is a little better.

How low can the price of the x2600 be? ATi will not get it low enough I fear because of the fixed costs with card manufacturing. The 65nm could help a lot tough.

Even worse the 8800 GTS has ~2.5x the shader power, whilst only having 2x the memory power of the 8600 GTS. The performance gap between 8800 GTS and 8600 GTS will improve further in future shader-intensive games. And with the 2900XT and 2600XT it is the same.

So why do the crappy cards sell? 260 euro for a 8800 GTS 320 MB (price here) is a lot of money, the 8800 is a power-hog and it is hot. Moreover the money you save from not buying a 8800 GTS 320 MB, can be put into use in your next videocard. It's a matter of whether your going to do a lot longer with the 8800 GTS 320 MB than with a 8600 GT/GTS.

ATI at least has the advantage of their x2600 being a 65 nm process. But they have hinted at a 65 nm x2800 XL which I guess is going to be a 256-bit bus with 160-240 PU's, and might be better value for money.

.....I wanted to get a x2600 XT, but now I write this it looks crap.....

Note of self-reflection: it is amazing to see the core-speed of the 8800 GTS vs the 8600 GTS. 500 vs 675, a factor 1.35. I didn't even use it in my calculations. I think the 8800 GTS has a much better core value than the 8600 GTS, which is simply needlessly fast seeing as how the heavy core OC of the 8600 GT didn't do much good.

[Shintai]

Oh really??
This: http://www.vr-zone.com/?i=4946&s=9 shows it consumes a "whooping" 17W than 8800GTX (not ultra!!! ).I guess it's gonna burn a hole in your wallet for a power bill if you go for a Radeon instead of a 8800GTX.

Uhm...burn a hole? yes! Would I buy the 8800GTX or Ultra? No.

I dont want a card that cost more on the powerbill for a year or 2 than the card itself cost.

So relax mr. Defender. I wouldn´t buy a 2 slot cooling GFX card.

And last time I checked, the x2900XT didnt beat the GTX but is on pair with the GTS. And the GTS is 65W lower in that review.

[Noobie]

Why would you compare the power consumption of a slower part to that of a faster part? VR-Zone does have 8800 GTS numbers to compare to.

Side note: Those power consumption numbers are useless to Vista Aero users. The 8800 can turn parts off, to save power. On one online test it was shown that in Aero the 8800 used less power than a videocard X, whilst vid-card X used less power in idle and less in full-load.

Then again I think Aero should be removed from Vista.

[Shintai]

Noobie...the 2600 series is 128bit...and you cant calc the flops scaling like that.

[informal]

As for the performance:this card has a lot of headroom trough the driver optimizations,it's fairly obvious thing with the latest set.Read Shamino's comment at the review about the game scores that went from 11% to 42% in different scenarios(games).
As for power bill, 65W is not that much for those who have the money for the card at the first place.And most of the time,card would be idling ,and the idle power is very good compared to other Nvidia cards.

[S1mon-]

Dunno if it has been posted yet but here is VR-Zone final review:

http://www.vr-zone.com/?i=4946

[ahmad]

That OC work bench review is the worst.

[Vice]

Few weeks and I'm going to pick one up. Hopefully by then some of the driver issues that are holding his card back will be fixed.

[MsB]

Overall I like this card for price point versus performance. I figured it would fall between gts and gtx in performance and with the OC ability I am not disappointed. However I want to see how these cards scale in CF as compared to SLI. If they scale better by 15% or better they are gonna be even better to me. I think SLI is a rip off for the cost versus extra performance.

[Noobie]

Noobie...the 2600 series is 128bit...and you cant calc the flops scaling like that.

I didn't calculate the FLOP but used ATI (trough a review) & nVidia's (trough wikipedia) information.

Here's the calculation nVidia used:
(MADD(2flops) + MUL(1flop)) * 1350MHz * 128 SPs ＝ 518.4 Gigaflops

Note that I am not comparing ATI vs nVidia based on Gigaflops (which would be stupid cos I would need to take slack, latencies etc into account).

I do not care about ATI vs nVidia, I care about the 8600-8800 relation and am using it to reflect on the possible 2600-2900 relation. 128-bit could work when they use faster memory. 512 = 3.2 * 1.25 (faster memory) * 128 (ring-bus). Card scaling would be 3.2 for mem vs 2.5 for shaders.

This would make the 2600XT-2900XTX more like a 8600GT - 8800GTS relation. That's good cos the 8600GTS:8600GT ratio is 1.25 for core, 1.25 for shader, but 1.43 for mem and the performance of the 8600 GTS is poor compared to the 8600GT (factor 1.1 mostly). But the 8600GT - 8800GTS is 2.85x, 3x for mem, shader. More shader power for the ATI compared to mem is nice, although this may be a bit out of proportion. It better be darn cheap to compete with the 2900 tough!

Note that it is really not wise to do such calculations and that results can vary greatly due to not taking a lot of things into account, but this is all I can do with the information & time I am given.

[fornowagain]

Part three

http://it-review.net/index.php?optio...=1335&Itemid=1

[fornowagain]

http://www.theinquirer.net/default.aspx?article=39578

Nvidia's anti ATi presentation

[naokaji]

performance of the hd2900xt card bounces around all over....
(according to the vr-zone review),

in some games nvidia does the happy dance...
in some other games its ati...
in some others they on par....

so.....

what should i think of that card?.... well... we will see what future drivers will bring... but then i might aswell stick with my 8800 gts and wait for r650....