R520 = Dissapointment

SugarCoat said:
broke 10k in 3Dmark05 with a single card, GPU overclock only. pretty sure its MSI based. that didnt take long.

Well I heard they got over 12k at the launch event thing.
 
AlphaWolf said:
Well I heard they got over 12k at the launch event thing.
Yeah, how come all that stuff has gone quiet? No press release about a new record or anything...

Even if it was with unofficial drivers, which means the results can't go on the ORB.

Jawed
 
AlphaWolf said:
Well I heard they got over 12k at the launch event thing.


LN2 does amazing things :). From the pics all i see are 3 120mm fans, 1 on the CPU, in addition to the stock cooling.

I wish they'd do more then the damn 3Dmark stuff though.
 
CMAN said:
I'm glad to see Monarch hasn't begun to charge crazy prices on the X1800 XL.

Has there been any word on whether AIBs vendors will be able to offer overclocked cards by ATI?

problably not, HIS/Sapphire have pages up with X1800XT specs and neither go beyond the stock, HIS definitly would if they could. Maybe Sapphire will bring back the liquid metal thing that poofed and do something but i doubt it.

http://www.hisdigital.com/html/x1k_promo.html
http://www.sapphiretech.com/en/products/graphics_specifications.php?gpid=122

But to be honost, with ATI giving Voltage control to the card in the bios/drivers (no more hard mods) as well as its own overclocking utility, once that gets support that is, it works out a bit better then some dismal OC to make one card "extra special" over the others. I always laughed at people paying 20 dollars more for the 20MHz OC back in the 6800 days.

something strange, Powercolor X1800 PDF has the XT as being available today..teasers
http://www.powercolor.com.tw/news_link/X1800-EN-20051003.pdf
 
Last edited by a moderator:
Hellbinder said:
Lots of stuff

You know what cracks me up the most about you Hellbinder. Is that this whole time you've acted like you knew what was coming, like you had an inside source. And now reviews come out, which oddly enough seem so freaking varied to me its hard to understand yet until we get the B3D one, that you had NO clue. None whatsoever, I find it funny, and I think you are getting what you deserved, its clear that you dont care about IQ, but only speed.

I like the X1k cards, a lot so far from what I can tell. I'll be fairly honest, I dont care about speed anymore, I see its importance though in the market. And if mid range numbers are true, ATi really shot themselves in the foot with the X1600 cards. Though from the Hexus review, ATi does have better speed. Sadly no one has gone into image quality yet.

I'm holding out judgement, so far it seems that if you bought a 7800GTX you're still happy unless you're a IQ whore (that's me). R520 is about what I expected, and it'll be interesting see how its staying power lasts.
 
Chalnoth said:
No, I'm really convinced that the reason that the R5xx sometimes tramples the NV4x is due to register pressure. If it were, enforcing FP16 for all operations should relieve this pressure and bring the NV4x back up....I doubt that it would. I'm willing to bet that it's due to pipeline stalls that the NV4x is experiencing that the R520 is not.

I'm not sure it's register pressure, I think your other explaination (scheduling of texture and special ops) is a much more plausible scenario. But moreover, taking full use of ILP in the split ALUs may also still be an issue for the driver compiler.

If you look at some papers on register file design, you'll see the average number of active *physical* registers in any basic block of code is quite small, about 4-7.

The major problem is that allocation of physical registers are automatic, and there is no "deallocation" instruction. Thus, either an architecture needs an extension so that a shader can specify "register Rn is no longer needed after this instruction", or, the compiler has to be very aggressive at eliminating temporaries. But eliminating temporaries interferes with other optimizations sometimes.

Ideally, the "deallocate" instruction could be completely hidden, and the compiler can use register use-def information "hint" to the driver when a register is dead.
 
AlphaWolf said:
Well I heard they got over 12k at the launch event thing.

Macci and Sampsa did 11150 actually; with only the X1800XT overclocked (the Athlon FX57 remained at stock speeds). So there is room for improvement :cool:
 
whats missing from the reviews is IQ comparisons like someone mentioned, as well as, overclockability(edit: i mean stock cooling) on both the XL and XT. I would love to see both.
 
I am quite impressed by this release. It definitely isn't a stinker, and it performs well in pretty much every situation. I like what they have done with the AA and AF, and hopefully availability won't be a problem for them.

Haha, I wonder how NV, MS, and ATI are now vying for 90 nm space these days (as well as all of TSMC's other clients)? It will be interesting to see how TSMC starts to fare here shortly with all of these 90 nm orders coming in droves.

I would say that this was a success for ATI, but people really were expecting a much better showing. That is understandable, as people would have been easier on ATI had it been released when they originally wanted to. Otherwise, it is really good competition to what NV has out.

I am somewhat disappointed in their midrange and low end parts so far, especially what I have seen in the reviews. They really don't offer much more than NV's current lineup with the 6200/6600/6800. I was expecting the X1600XT to whip the pants off of the 6800 and be fairly comparable to the 6800 GT, but that just doesn't look to be the case. This really makes you step back and wonder what NV will be pushing out here for their midrange 90 nm parts...
 
  • Like
Reactions: Geo
Skrying said:
. . .which oddly enough seem so freaking varied to me its hard to understand yet until we get the B3D one. . .

As much as that will help, and as much as I admire and appreciate Wavey's efforts, I think it is going to be longer than that. I forsee a couple months of poking/prodding at this beastie and different scenarios.

Someone asked an entirely legitimate question upstream --how much does the extra memory and faster memory speed play a part here in the eye-popping instances where X1800XT kicks butt? These are things that I think most of us are expecting NV to match in the short-to-mid-term. The fine, but-hardly-eye-popping, performance of X1800XL suggests an answer without quite nailing it down in my mind yet --not enuf scenarios yet, I think.

But I think we'll get there. I do think R580 is the one to moon over tho at this point.

It does seem to me that the "green team" has been a lot nastier about these cards on average (with exceptions, of course) than the "red team" was about G70 at release, for whatever reason. I certainly don't remember anyone suggesting that it was somehow inappropriate to benchmark 7800GTX against X850XTPE based on street price on release date of GTX. . . yet we've heard the analgous from more than one person the last couple of days.
 
geo said:
As much as that will help, and as much as I admire and appreciate Wavey's efforts, I think it is going to be longer than that. I forsee a couple months of poking/prodding at this beastie and different scenarios.


I certainly don't remember anyone suggesting that it was somehow inappropriate to benchmark 7800GTX against X850XTPE based on street price on release date of GTX. . . yet we've heard the analgous from more than one person the last couple of days.

Agreed and agreed.
 
geo said:
Someone asked an entirely legitimate question upstream --how much does the extra memory and faster memory speed play a part here in the eye-popping instances where X1800XT kicks butt? These are things that I think most of us are expecting NV to match in the short-to-mid-term.

Yeah that was me. I dont know why no one else is bringing this fact up.
 
DemoCoder said:
I'm not sure it's register pressure, I think your other explaination (scheduling of texture and special ops) is a much more plausible scenario. But moreover, taking full use of ILP in the split ALUs may also still be an issue for the driver compiler.

If you look at some papers on register file design, you'll see the average number of active *physical* registers in any basic block of code is quite small, about 4-7.

The major problem is that allocation of physical registers are automatic, and there is no "deallocation" instruction. Thus, either an architecture needs an extension so that a shader can specify "register Rn is no longer needed after this instruction", or, the compiler has to be very aggressive at eliminating temporaries. But eliminating temporaries interferes with other optimizations sometimes.
Well, I'm operating under the assumption that there's a "local" register file and a "global" register file. The local file need only store those few registers that are needed by the current execution cycle, while the total number is stored in the global file.
 
geo said:
Someone asked an entirely legitimate question upstream --how much does the extra memory and faster memory speed play a part here in the eye-popping instances where X1800XT kicks butt?

Well we can look to the x800xl 512mb scores to get an idea???
 
geo said:
It does seem to me that the "green team" has been a lot nastier about these cards on average (with exceptions, of course) than the "red team" was about G70 at release, for whatever reason. I certainly don't remember anyone suggesting that it was somehow inappropriate to benchmark 7800GTX against X850XTPE based on street price on release date of GTX. . . yet we've heard the analgous from more than one person the last couple of days.
I thought I pointed out the ridiculousness of pitting the 7800 GTX against anything in the R4xx lineup. The issue was that ATI just didn't have anything to compete with the 7800 GTX at its launch, and that was a far more damning issue for ATI than worrying about comparing products to the right competitors.

At least it has turned out that most reviews have been quite good about including the nearby competitors in their previews, as well as just the ATI-suggested competitors.
 
Chalnoth said:
At least it has turned out that most reviews have been quite good about including the nearby competitors in their previews, as well as just the ATI-suggested competitors.


But once the ATI cards are out and start to fall in price, would ATI's suggestions make more sense? I mean ATI used MSRPs to compare...
 
Back
Top