Flamey my boy, I never doubted you and nVidia when it came to drivers heheh, but I always had doubts of the FSAA improvement, as it seemed it was inherent to the architecture.
All I can say though, is...amazing.
No one has yet to comment in this thread, in detail, but I will.
First of all, the cooler is somewhat different from the pics other sites had in preview, not sure if I am wrong on that. Second, I wish THG would release GPU temps and power consumption charts. The card is likely to still be overly hot and consuming.
Glad to know it isn't a dustbuster anymore, though the fan itself looks awefully familiar to the FX Flow. At least in the end, nVidia had the guts to admit it wasn't succesful, the NV30. Though not that much props, since you can easily interpret that since they will release a product that will shadow the FX5800 and even replace its market availability, then they can make fun with no problems.
We've yet to hear of the FP performance which was supposedly doubled, but I am excited.
The Intellisample advancements show once more how nVidia can truly improve. I thought the NV3x serie in itself will have utterly horrible FSAA, but the drivers have done their job. If you ask me though, there is no architectural enhancement in this domain though, the entire extra performance comes from the bandwidth which I am willing to bet, is not even entirely used at all (consider having 10GB more, and you only achieve a moderate 30% maximum over your brother).
Detonator-XP: Now this is what power is all about. There is no denying it, nVidia has probably rehired the real programmers, and yanked out a surprise boost that is significant, enough to put the FX5800 Ultra in a totally different light. One has to wonder why they did not try the FX5600 yet, though I am hoping it does well. They also mention 44.05 as the version, what about the 50.xx. Do you think there is so much more to be expected?
What surprised me majorly is that drivers actually removed the misery the NV30 had in FSAA and Aniso, and in such a good way that it actually rapes sometimes by nearly 50% the competition. I just don't know why it was so bad back then, but just a few months of programming, yeilded so much in return. The Pentium 4 could do that if you use SSE 2 so well, so I'm glad to see some real programming.
Regardless, hope the Detonator FX do some justice to even the NV2x serie.
Benchmark-wise, I'll go through in order:
Medium Quality: Ok, so who actually claimed all that time that ATi was the best suited for Doom III? It doesn't make sense at all, the NV30s are so well coded here, they rape.
High Quality: Ok something got a bit weird here and the FX5900 Ultra lost. It is a bit odd it loses this badly performance, compared to Medium.
The FX5600 Ultra even has a lovely time against the 9600PRO, who btw was playing on a 3.06GHZ with i850, the second highest-performing chipset for the Pentium 4 under Medium Detail. Can you imagine you users under slower comps? I hope THG tests CPU scaling, FAST, because that leaves us with a dry mouth if we tried Doom III on 1.4GHZ-2GHZ comps, without any notice of CPU scaling. One thing is clear though, the DX9 is doing its job for the FX5600 Ultra, didn't expect it to shine this much over the Ti4200.
As for the FX5200 Ultra, DX9 for 79 works, only for Dells and gamers who prefer Medium to Low quality, otherwise imagine High Quality crap.
Still, overall, the DOOM III results could've been better IMO. They're definitely optimized though, much more than the Alpha ever was. It just worries me sick what CPU will be the bottleneck limit.
I would, however, like to direct the viewer's attention to these two odd results: <A HREF="http://www.tomshardware.com/graphic/20030512/geforce_fx_5900-12.html" target="_new"> EXHIBIT A
</A>
In Exhibit A, we see the regular performance of the FX5900 Ultra under High Quality.
<A HREF="http://www.tomshardware.com/graphic/20030512/geforce_fx_5900-13.html" target="_new"> EXHIBIT B</A>
In Exhibit B, we see the FSAA results for HIGH QUALITY as denoted by THG.
The prosecution rests, the defendants must now answer this: How do you achieve higher performance with FSAA than regular, if you claim High Quality setting was used?
UT2003: Nothing new, except that the high res results of the FX5800 Ultra are simply jaw-dropping.
Serious Sam 2: Under Aniso 8X, the FX5800 Ultra rises to the point of unbelievable driver strengh, it manages to outdo the highest competition by a clear 50%. Nothing short of amazing, I simply have a hard time believing it.
Under full FSAA+ANISO, it and the FX5900 Ultra continue to enjoy a rather healthy and loveable dinnertime eating the competition away.
Splinter Cell: And now we move on to what I like to call: The hall of crap.
In comes the 3rd proof I can provide you now, once more, as to why Splinter Cell is badly optimized, that it doesn't use Pipe awareness, or even IPC enhancement. It simply relies on clock speed, PERIOD. The FX5900 Ultra loses, it is clear. This is demonstrated ocne more by the game's lack of optimization, and it proves that not everyone has learned from the Commanche 4 disaster. There are still monkeys out there, FIRE THEM.
While at first I was surprised with the 3dMark 01 results, I was later convinced of the overall result, due to the Shader speed, which was once again rather odd. There is still also no explanation to the abnormally high Poly 8 Light test result. Also another test which does not use card architecture but sole clock speed.
<b>And here comes what I am most skeptical about, and what may rise some eyes. 3d Mark 03, a few days ago was previewed with the FX5900 Ultra, and had performance way over 6000 pts. That was with the Detonator 50. What does this mean? Does this mean there is an extra 15% to be expected from the Detonator FX, on all FX cards, especially the FX5900 Ultra who so far has NOT proven itself to its family?</b>
Now to analyze and view a few things -OBSERVATIONS-: (this list could get updated hehe)
-The FX5900 Ultra, is a disappointment to its siblings. Why am I saying that?
At the moment, the FX5800 Ultra stands most of the time at 5-10% behind it. How does having much better clock performance and 10GBs more per second not help?
Which leads me to be skeptical and awaiting the Detonator 50s. BTW, Catalyst 3.4 on the new R9800 256MB is a failure, so I must ask myself what about the rest of the family? How can the 9700s edge more than this, if they would trample the 9800s which barely have a lead over them?
-The card's clock is still too high. This proves that while nVidia's drivers are excellent and the card is performing solidly and way beyond my expectations, if the card was downclocked to 380MHZ and 320MHZ memory, it's likely not be as competitive. It may have arrived at the point of having equal FSAA and Aniso performance per clock though, if not better. But standard perf? Doubtful.
-One has to wonder, just how less performing is the FX5900 Value? And how consuming is it? I'd love a contender to the 9700PRO, with performance rivaling the FX5800 or slightly less, at such price. But I am worried about the power consumption and heat above all, so THG better try to get some info on this.
-Will we ever hear the new fan? Sound is subjective as we know.
-The 9800PRO 256MB is new the wanker of the century, costing us Canucks a horrid 750$ CDN for an extra NON-ENHANCING 128MB which often even lowers performance due to higher search time in the memory (latency).
-The price above all will make this NV35. I am hoping to god retailers will opt for lower prices or that nVidia will release value oriented versions, which maintain a CLEAR lead over the 9700 non-PRO at least. After seeing the Doom III results, I am convinced that I have to go for the 9700, or else I'll suffer like Ape will, who had thought otherwise... :frown:
-The FX5800 Ultra uses 16GB/sec of bandwidth and the same color compression, yet as we have witnessed, it raped the competition despite the usual claim that bandwidth makes everything in image quality enhancement. In contrast, the 9800 uses 21.8 and is probably not inching close to using it all effectively, and the 5900 is 11GBs above and still not more than 50% better, but rather around 20% at most, against its younger brother. That is a bit sad, but like I said, drivers might change even more the situation, as if this wasn't good enough!
-From what I just read at [H]OCP, the FX5600 for the very first time, is raping away the 9500PRO, killing the 9600PRO (VERY DECEIVING UNDERPERFORMANCE) and contending very well, far beyond what I ever imagined, especially since it boasts supposedly crappier DX9. Again I am confused at the late performance improvements and hope a future article solely on the new drivers will clarify all new FX cards' performance.
Well, that ends -or supposedly in my opinion-, this inevitably long opinionated post, which I know many view it as I.
This should represent more than enough of the opinions who were to have equal length posts, like Grape Ape would do or Twitchy-boy. :tongue:
--
This post is brought to you by Eden, on a Via Eden, in the garden of Eden.
<P ID="edit"><FONT SIZE=-1><EM>Edited by Eden on 05/12/03 09:00 PM.</EM></FONT></P>
Facebook
Twitter
Instagram