Moore's law is not what it used to be, but demanding higher performance and cursing at the wind is not going to change that.
Moore's law is not what it used to be, but demanding higher performance and cursing at the wind is not going to change that.
I take it you're blaming TSMC for not innovating fact enough? What do you suggest AMD and Nvidia should do about that? Any suggestions where they can find low hanging fruit?
50%-100% faster <> no difference in unplayability? I agree that 500 vs 300 fps is not noticeable. Are you running Doom 3 by any chance? Wolfstein? Commander Keen?
Didn't you know that Charlie is SEMI-accurate ?
I prefer to call it inaccurate..at most occasions
Are you saying that TSMC is directly responsible for two companies who use varying designs and perf ethos? Are you saying that AMD couldn't run the 7970 @ 1GHz from the word go, when everyone else seems to be able to anyway? Or that TMSC are responsible for AMD having put a load of compute stuff into their gaming part, and we got 32 ROPs instead of 48/64. Or NV releasing a part thats *just* enough to compete, when BigK seems to be ready, but not needed. The last couple of pages of this thread have been about the supposed NV supply issues being nonsense, so lets agree that TSMC doenst have any major crippling issues that could stop either company releasing a more powerful card.
No, I'm blaming AMD/NVidia for doing "just enough". Which seems to work, granted, cos people buy their cards.
Or, you know, Crysis and Metro2033, which run at ~15fps 2560x1600 with AA on the 4870x2 and ~30fps on the 670. 100% improvement. Still fairly unplayable.
Isn't the 7970 quite a bit larger than the GTX 680? Because that alone could put a limit on clock speeds, everything else equal, just due to the higher power requirements of having more transistors.
It's BS, but I wish I could give you an up-vote for 'perf ethos'.
Dave has addressed this: they were unsure about the bin spread of the 28nm process and decided to be relatively conservative during silicon qual. That's completely defendable. Especially since, at the time, there were probably justified in assuming that a 256-bit 300mm2 GK104 would not beat them in gaming performance.
By how much do you think, average performance will go up if they had used 48 instead of 32 ROPs? 5%? Don't think it's going to be much more than that, but let's put it generously at 10%. So you're whining that a GTX670 is only 100% faster than a power hogging 4870X2, but suddenly a 10% difference is a big deal?
I don't know where you get your info, but I haven't seen anything about it being ready. All I've seen is an announcement. And a release target sometime Q4.
Kindly do the effort and reread: the last couple of page were about the yield issues being nonsense. The lack of wafer allocation is not in dispute, quite on the contrary.
You obviously missed my point: in this generation, it's clear that both of them have similar perf/mm2 and perf/W, especially after discounting the compute related overhead of Tahiti. This should tell you something: at the very core, AMD and Nvidia depend on process technology to dramatically advance performance. Not the 10% stuff, but the major steps forward. Given enough time, I'm sure Nvidia and AMD could improve perf/W and perf/mm2 by some amount for the same process, but the very fact that they are so similar shows that they're both scraping at the bottom of the barrel. Conclusion: from a bird-eye point of view, they are restricted by the speed at which TSMC innovates and introduces new technology.
Once you accept this point, the only thing that matters how both companies decide to sprinkle their roadmap with silicon versions. That's marketing at work for you. A delicate balance of trading of features (perf) vs. cost (die size) vs. price points.
With constraints above, all you're complaining about is that both Nvidia and AMD dared to create a die that wasn't die enough to your liking. Well, tough.
For all the examples you can give of games running at 15fps now running at 30pfs, I can give many more that went from 25fps to 50fps. IOW: from unplayable to very playable. But you knew that, right?
And yet there are plenty of standard users (and some AIBs) that are pushing the clock past 1Ghz without significantly increasing wattage. And I said back when the 7970 was first released and I had my hands on one for a week or so, the cooler feels like a solution for a hotter (?) card. Given all the easy overclocks of this card, I'm convinced that it's been "artificially" lowered in performance. It's easily capable of so much more, without breaking heat or TDP levels.
I'm going with what I've read and actually used, rather than anything coming from AMD or NV, cos they would fudge the numbers to put out the least they could get away with, would they? Less work(load) on this gen means less work on the next gen. Hopefully its wrong, but that report of AMD bringing out a new gen and pretty much all thats happened is a minor (and already achieveable) clock increase? How is that OK with you?
UniversalTruth
Veteran
Yeah,
How easy it is to simply bash somebody who you most probably don't even respect, than instead give a proposal what other reputable sources to read...
Are there any others?
Yield.
Last edited by a moderator:
Factor in driver improvements from 11.12 to 12.6 and they could probably claim 20% over the original 7970 performance.
Yeah, for the production run in november.
Actually, thats not really what I said at all.
I think that depends upon what you mean by "significantly". And given that it seems consumers are considering power requirements more and more these days, it makes total sense to me to lower clocks a bit to keep the power consumption down.
Meh, I don't care all that much. I got my GTX 670. I'm happy for a while.
Blazkowicz
Legend
radeon 4870X2 is precisely at the end of an exponential increase in bandwith and power use that was taking place since the TNT / Voodoo2 days. exponential progress starting from < 1GB/s bandwith and chips with no heatsink and ended somewhat with being stuck with gddr5 and very high powered cards such as 4870X2 or GTX 470.
you're whining because they did not make a GPU with a 1024bit bus and 1KW TDP, too bad.
you're whining because they did not make a GPU with a 1024bit bus and 1KW TDP, too bad.
i read "understanding things with the new process" as "process variation", because in the hypothetical case of no process variation, you can understand everything with just 1 engineering sample.Dave Baumann said:
A1xLLcqAgt0qc2RyMz0y
Veteran
All this talk about the updated higher clocked Tahiti has me thinking. With the lower power of the GTX680 (190 watt, 170 typical) vs Tahiti what prevents Nvidia from doing the same and releasing a GTX685 that is a higher clocked GTX680 and has 6 & 8 pin power and a 225 watt max (210 typical) power usage.
Supply?
HD 7970 and the GTX 680 are actually very close in power consumption. Here a couple of power readings that take the reading from the graphics card alone and over the course of the benchmark rather than the max power at the wall.
http://www.hardware.fr/articles/857-8/consommation-performances-watt.html
http://www.pcgameshardware.de/aid,873907/Test-Geforce-GTX-680-Kepler-GK104/Grafikkarte/Test/?page=19
It really depend of the game used.. i remember someone have post one review with power consumption over 10 games... difference between each games was sometimes night and day.
Anyway the difference with real usage is even bigger, most test dont use v-sync on, or the gpu usage will change a lot when it is enabled. Its even bigger with CFX or SLI, by limiting the framerate, the gpu usage is really decreasing.
On my system, in BF3, the gpu usage with v-sync is just 50-60% on both cards. ( my min framerate is a lot over 60fps )
Anyway the difference with real usage is even bigger, most test dont use v-sync on, or the gpu usage will change a lot when it is enabled. Its even bigger with CFX or SLI, by limiting the framerate, the gpu usage is really decreasing.
On my system, in BF3, the gpu usage with v-sync is just 50-60% on both cards. ( my min framerate is a lot over 60fps )
Last edited by a moderator:
