Must be really upsetting ending up being another marketing tool used to generate hype.
GF100 evaluation thread
- Thread starter rpg.314
- Start date
Thank you. And if i could find my original post from Dec. Jan I think it was, The info I was given isn't to far off from what they settled for in launching products. Just woulda been nice had they went with the full 512 instead of short changing it to 480.
Funny, you found Cypress to be MEH and this one to be YAY. Just stating the obvious here.
Don't see what's so funny about it. Cypress was a straight doubling of RV770 so the expectation there was clear as day. Fermi managed to beat GT200 by at least as much as Cypress beat RV770 while at the same time delivering significant architectural changes and a robust DX11 implementation. And all with much less texturing power to boot...
According to computerbase at 1920x1200 4xAA the 5870 is running 55% ahead of the 4890 while the 480 is 63% ahead of the 285. So 63% faster, massive tessellation performance and global memory caching. So aside from power consumption where's the MEH?
Very true. On the other hand Fermi somewhat justifies its girth in terms of extra goodies.
Well, yeah it is a fair bit more exciting architecturally, but it is pretty disappointing from a 3D performance/watt/mm^2/dollar point of view. This is further compounded by the fact that Nvidia appears to be artificially limiting DP performance in its "consumer" devices, when it likely cost them a not insignificant amount of die area to implement.
I'm not the one in the minority here saying that both cards rock here.
Why pick only 1920x1200 4xAA? I'll do a "trini" by cherry picking 2560x1600 and I arrive at 480 59% faster than 285 and 5870 a whopping 76% faster than 4890. Funny you were the one throwing the pre-mature cherry picking accusations around.
"whoops"jimmyjames123
Regular
The GTX 470 seems pretty good, while GTX 480 is good too. Neither are truly exceptional with respect to performance. Certainly not worth the upgrade for anyone with a high end DX11 card from ATI/AMD, as they would be better off by adding a second card. That said, for anyone new to the DX11 market, and/or for anyone coming from a G80-based or GT200-based card looking to play DX11 games and PhysX games, then GTX470/480 would be a worthwhile upgrade.
GTX 470 seems to be the better buy to me, and I think some reviewers got this completely wrong based on a very small sample size of games tested (hint: [H]OCP). GTX 480 costs ~40% more money, for only ~25% more performance than GTX 470, with the exact same feature set. And there is no doubt that GTX 470 SLI will have far higher framerate than GTX 480. Power consumption is reasonable too on the 470. According to PCPer review, idle power consumption on GTX 470 is only ~ 6% more than HD 5870 (load power consumption is ~25% more than HD 5870, but still within reason). Looking at a wide variety of games, it seems that performance of GTX 470 is closer to HD 5870 than it is to HD 5850, and in many games the 470 is as fast or faster than 5870.
In the Anandtech review, the author mentioned that there is a problem with the new Transparency Supersampling (TrSS) mode being too aggressive with supersampling, which causes the framerates to tank when enabling this setting on GTX 470/480. So that would explain some of the oddly low framerates in two of the four games tested by [H]OCP. Apparently this issue will be fixed with new drivers, release 256.xx.
One area where GTX 470/480 really seem to excel is in keeping a relatively high minimum framerate at graphically intensive settings, so real world gaming performance should be quite good. And of course these cards are relatively strong with extreme tesselation settings.
There is no doubt that GTX 480 is louder than most of the other cards in the test suite, but I think that some claims of 70dB are highly exagerrated due to having the sound level meter placed very close to the graphics card. Hilbert @ Guru3D measured 37dBA at idle and 45dBA at load with the sound level meter about 2.5 feet away. Keep in mind that sound pressure level increases by 6db for each halving in distance between sound level meter and source. So Hilbert's 45dBA at 30 inches away would result in 69dBA at 1.875" away from the card. Pretty big difference in sound pressure level there. And of course having 30 inches of space between ear and PC seems far more realistic than having only 2"!
Looks like GTX 470 and GTX 480 are about 15-20% short on performance to really catch people's attention. These cards give NVIDIA a competitive product in the $300-$500 price range, but certainly not a dominant product by any means.
GTX 470 seems to be the better buy to me, and I think some reviewers got this completely wrong based on a very small sample size of games tested (hint: [H]OCP). GTX 480 costs ~40% more money, for only ~25% more performance than GTX 470, with the exact same feature set. And there is no doubt that GTX 470 SLI will have far higher framerate than GTX 480. Power consumption is reasonable too on the 470. According to PCPer review, idle power consumption on GTX 470 is only ~ 6% more than HD 5870 (load power consumption is ~25% more than HD 5870, but still within reason). Looking at a wide variety of games, it seems that performance of GTX 470 is closer to HD 5870 than it is to HD 5850, and in many games the 470 is as fast or faster than 5870.
In the Anandtech review, the author mentioned that there is a problem with the new Transparency Supersampling (TrSS) mode being too aggressive with supersampling, which causes the framerates to tank when enabling this setting on GTX 470/480. So that would explain some of the oddly low framerates in two of the four games tested by [H]OCP. Apparently this issue will be fixed with new drivers, release 256.xx.
One area where GTX 470/480 really seem to excel is in keeping a relatively high minimum framerate at graphically intensive settings, so real world gaming performance should be quite good. And of course these cards are relatively strong with extreme tesselation settings.
There is no doubt that GTX 480 is louder than most of the other cards in the test suite, but I think that some claims of 70dB are highly exagerrated due to having the sound level meter placed very close to the graphics card. Hilbert @ Guru3D measured 37dBA at idle and 45dBA at load with the sound level meter about 2.5 feet away. Keep in mind that sound pressure level increases by 6db for each halving in distance between sound level meter and source. So Hilbert's 45dBA at 30 inches away would result in 69dBA at 1.875" away from the card. Pretty big difference in sound pressure level there. And of course having 30 inches of space between ear and PC seems far more realistic than having only 2"!
Looks like GTX 470 and GTX 480 are about 15-20% short on performance to really catch people's attention. These cards give NVIDIA a competitive product in the $300-$500 price range, but certainly not a dominant product by any means.
Last edited by a moderator:
Archaeolept
Newcomer
people aren't pointing out xman because he was wrong about the SP count - most likely nvidia itself wasn't sure on its final decision for the card until very recently - but about how adamant he was about it, and how certain he was about the reliability of his "source"
I had to vote double WTF, although I likely would have gone WTF 480 meh 470 if that had been an option. Performance is fine, really, overall, (say 10-15% for the 480 using a wide variety of not just nvidia suggested games and the trustworthy sites) but the power -> heat -> noise equation is just brutal. Much worse than I thought would be the case.
Along with high idle power, even higher 2d power, and mind-bending power-use if you have 2 monitors (which, I suspect, a majority of the potential customers would have), and that most sites (as per standard practice) tested the rigs outside of a case... that can only add up to a failure in my eyes. If nvidia is unable to solve the 2 monitor issue I'd go so far as to classify the hardware as broken.
or, not quite. certainly even so there will be some enthusiasts for whom it is the best card, really, and not just if through fanboy glasses. With a very good case for airflow, and a fat enough wallet, and an allergy to AFR, the 480 could be the right buy; but only for really a small number of potential targets.
I had to vote double WTF, although I likely would have gone WTF 480 meh 470 if that had been an option. Performance is fine, really, overall, (say 10-15% for the 480 using a wide variety of not just nvidia suggested games and the trustworthy sites) but the power -> heat -> noise equation is just brutal. Much worse than I thought would be the case.
Along with high idle power, even higher 2d power, and mind-bending power-use if you have 2 monitors (which, I suspect, a majority of the potential customers would have), and that most sites (as per standard practice) tested the rigs outside of a case... that can only add up to a failure in my eyes. If nvidia is unable to solve the 2 monitor issue I'd go so far as to classify the hardware as broken.
or, not quite. certainly even so there will be some enthusiasts for whom it is the best card, really, and not just if through fanboy glasses. With a very good case for airflow, and a fat enough wallet, and an allergy to AFR, the 480 could be the right buy; but only for really a small number of potential targets.
Last edited by a moderator:
Why pick only 1920x1200 4xAA? I'll do a "trini" by cherry picking 2560x1600 and I arrive at 480 59% faster than 285 and 5870 a whopping 76% faster than 4890. Funny you were the one throwing the pre-mature cherry picking accusations around.
Heh, "only" 1920x1200 4xAA? Feel free to scrounge around for whatever settings you like. Maybe even check out those 8xAA tests while you're at it.
No matter where you look you're going to come back with the same conclusion - big gains over GT200 and strong DX11 performance.[edit] Of the 9 settings used in that review you picked the only one where Cypress shows a bigger gain? Lolz, yes surely I'm cherry picking when 8 of 9 settings are in line with my comments. Sigh....
I dunno, there's definitely something impressive going on in the tests where it beats the 5870 by 50%. What that is, and whether it's also a factor in its occasional min framerate wins, I'd love to know. Is it its monstrous z-rate (looking at hw.fr's review)? Its newfangled setup? Its cache? Simply more memory channels? Why is it so much better at culling tris than 5870 (again, hw.fr)?
Obviously the heat is worrying and the power draw is off-putting, but that only makes GF104 that much harder to wait for.
Obviously the heat is worrying and the power draw is off-putting, but that only makes GF104 that much harder to wait for.
That's an easy one isn't it? 4 tris culled per clock versus 1? In terms of min framerates who knows, maybe the cache helps smooth out intra-frame bandwidth usage or it better manages high instantaneous demands on the geometry pipeline. Can't tell for sure without profiling the app.
brain_stew
Regular
Its 90C at idle when using dual monitors.
http://www.legitreviews.com/article/1258/15/
Having a higher power draw than ATI's dual GPU solution all for ~10% better performance than a 5870 makes it a spectacular failure in my book. Its a huge disappointment as far as I'm concerned, here in the UK the GTX 470 costs the same as a 5870 and yet its slower, hotter, louder and consumes a lot more power. Its impossible to recommend.
thatdude90210
Regular
I saw someone calling it a dual screen oven.
jimmyjames123
Regular
So what is the deal with the wildly varying numbers on power consumption and temperature in these GTX 470/480 reviews?
Here is what Hardware Canucks showed regarding idle/load power consumption:
In that chart, the GTX 470 is within 5% of the HD 5870 with respect to idle power consumption, and GTX 470 is within 10% of HD 5870 with respect to load power consumption. And in turn, the GTX 480 is within 13% of the GTX 470 with respect to idle power consumption, and GTX 480 is within 11% of the GTX 470 with respect to load power consumption.
And here is what Hardware Canucks showed regarding idle/load temperatures:
In that chart, the GTX 470 is within 4% of the HD 5870 with respect to idle temperature, and GTX 470 is within 11% of HD 5870 with respect to load temperature. And in turn, the GTX 480 is within 12% of the GTX 470 with respect to idle temperature, and GTX 480 is within 3% of the GTX 470 with respect to load temperature.
Here is what Hardware Canucks showed regarding idle/load power consumption:
In that chart, the GTX 470 is within 5% of the HD 5870 with respect to idle power consumption, and GTX 470 is within 10% of HD 5870 with respect to load power consumption. And in turn, the GTX 480 is within 13% of the GTX 470 with respect to idle power consumption, and GTX 480 is within 11% of the GTX 470 with respect to load power consumption.
And here is what Hardware Canucks showed regarding idle/load temperatures:
In that chart, the GTX 470 is within 4% of the HD 5870 with respect to idle temperature, and GTX 470 is within 11% of HD 5870 with respect to load temperature. And in turn, the GTX 480 is within 12% of the GTX 470 with respect to idle temperature, and GTX 480 is within 3% of the GTX 470 with respect to load temperature.
Sample variability? Let us know when you managed to put one in your rig and have experimented with it yourself
The leakage variation is probably very wide.
The percentage differences for the graphics card alone are much wider than that as these look to be system load powers.
jimmyjames123
Regular
That's true, good point. Now, that said, what should we be looking for with respect to power consumption? It seems to me that total system power consumption is the important metric for the end user, right?
BTW, It appears that the ALU:TEX in fermi is 8:1. Wouldn't that be too high.
Yup. B1, if it arrives, will be a better target.
Yes and no. Big changes are due, but they need not arrive all at once. EG adds r/w caches (albeit at L2 level), 2 rasterizers for a limited support for distributed ff processing, and most importantly DX11 so that you don't lose a fight by not turning up.
Hecaton could conceivably add a unified cache hierarchy with multiple tessellators without too invasive changes. Larger LDS might be a convenient way of solving the bank conflicts we saw in tessellation. Distributed setup/raster might be doable too. NI seems more reasonable time for it though. After all, you already have Cypress split down the middle into two SIMD. And who's knows, distributed geometry might have been wrecked by 40nm too and the 2 rasterizers survived to chew the smaller triangles generated by tessellation.
It is due, but shorter refresh cycles mean that there is no need to throw the entire kitchen sink into one chip.
aaronspink
Veteran
ouch, that's really bad. Though honestly all the card should be able to handle multiple displays with less power/heat but the 480 is ridiculous.
aaronspink
Veteran
actually the tests kyle did are pretty much best case. Inside a case the ambient temperature would be significantly higher in all but effectively mesh cases and the vast majority of cases provide very little sound dampening.
Sure if you put it inside my always on machine it will quite it down, but my always on machine has $100+ in mcmaster-carr aftermarket additions (love me mass loaded vinyl!) plus a significant cost in the best fans money can buy.
Similar threads
