Yup, that sounds about right. Very impressive for a 2009 GPU. Only 3 fps slower than GTX580 with no AA!
I also think 2 things are making HD5870 look better than normally would have been the case:
1) Most games are still made with consoles in mind. Even FC3, while pretty, is not really a next generation game in terms of graphics (just my opinion). That means even if you crank everything to the max on GTX680/7970, FC3 won't look much better than it does on HD5870, as most games look very similar now between High an Ultra. That makes it a big harder to see the value of newer $500 cards.
2) I don't view GTX500/HD6900 series as a real new generation in performance. They are more of a refresh in my eyes. That means since Sept 2009 when HD5870 launched, we have really only gone through 1 major new generation (GTX600/HD7900), but it's been about 3 years and 3 months. That means the pace of innovation/performance improvements has slowed.
Just to give you a point of reference, we went from 9800GTX in just 2.5 years:
March 31, 2008 = 9800GTX launched
November 9, 2010 = GTX580 launched
GTS250 is actually slightly faster than 9800GTX but you can see GTX580 is 143% faster than GTS250 is with AA/AF:
http://www.computerbase.de/artikel/grafikkarten/2011/bericht-grafikkarten-evolution/3/
HD7970Ghz is about 80% faster than 5870 is with AA/AF:
http://www.computerbase.de/artikel/grafikkarten/2012/test-amd-radeon-hd-7970-ghz-edition/4/
So we have 143% increase from NV in 2.5 years vs. 80% increase in 3.25 years when comparing HD5870 to where we are today. Games aren't looking much better than they did compared to Metro 2033/Crysis/Warhead and GPU speed has increased at a slower pace. The end result is HD5870 still looks great.
Good read! :thumbsup:
HD 5870 was awesome, yeah for a 3.25-year old card considering that there's not yet a single GPU card that's 2x as fast OVERALL.
But I did wish that AMD also put that 512-bit bus on the 5870 like they did with the 2900XT from a few years earlier. In that link
http://www.computerbase.de/artikel/grafikkarten/2011/bericht-grafikkarten-evolution/3/
(although it tests only 7 games)
it shows HD 5870 to beat HD 4890 by ONLY 39-44%, despite the fact that HD 5870 had pretty much 2x the GPU muscle (apart from the memory clock/bus), with identical core clock, 2x TMUs, 2x shaders, 2x ROPs, plus roughly 2% improvement from Evergreen architecture optimizations. Even during the article's publication later on in 2011 with newer, more shader-heavy games (although still DX9 only for comparison purposes), they still did not make the 5870's 1600sp appear to shine much at all compared against the 4890's 800sp part. If the 5870 did sport a 512-bit bus, it would easily have been a full 15% faster than what it were. GTX 480, being only 15% faster overall, would have been a complete LOSS due to being insanely inefficient - perhaps NV would have fully delayed GF100 until GF110 (GTX 580) was ready. Why, oh why, didn't AMD do it aggressively enough? They were extremely aggressive with the X1900XTX and the HD 2900XT. I guess they just couldn't cope with the defeat very well, when Nvidia was even more aggressive with their godly 8800GTX and 8800 Ultra. The backpedaling to multiple, tiny GPUs (HD 3870X2) proved to be unsuccessful. They wanted to do a quad-3870 card, but it was never quite so feasible with the drivers not always having ideal scaling in games, plus insane power consumption. The 4870 should have been done in place of 3870 (about 7 months earlier), in order to completely tip the scales over before Nvidia even got to their G92 architecture. It would have even forced NV to go ahead with their G90 architecture (I think I know what the specs are) rather than taking the luxury of moving on to the slightly delayed GT200 arch.
Sometimes, in a fight, aggressiveness is the main ingredient.
Losing with the HD 2900XT did get them down, but they should have kept at it using the same fierce strategy. They were working on a 640sp R700 (rather than 800sp part) for 4870 months earlier right when the 3870 was launched, but they should have unleashed it instead of waiting all the way until summer 2008. Then they could do yet bigger version of it, perhaps with more than 800sp and even 32 ROPs to give the over-sized GTX 280 a really, really good fight until Nvidia was finally ready with their late 55nm GT200b.
True, the 5870 was finally their 'win'. It was great, alright. Nvidia's continued aggressiveness caused them to stumble early, while AMD kicked them in the sweet ribs with the 5970.
The 5970 was what they originally wanted to do with their 3870X2. Sheer dominance with grace and ease in the majority of popular modern games.
However, I would still have liked to see them do a 512-bit version of HD 5870, even if it was not until after GTX 480 came out, just to have a single GPU card keep Nvidia down and humble.
Then AMD would have been able to store and recycle that dominant energy over into Cayman, by also designing them to be at least 512-bit cards. A 512-bit 6970 would have been rather neck-to-neck with GTX 580.
Also, it would have been nice to see a souped-up version of Tahiti sporting 48 ROPs without these limited crossbar access shortcuts to the 384-bit memory, for more efficient bandwidth utilization (and greater theoretical peak pixel fill-rate). Ever seen a 7870 actually beat a 7950 in a few games, despite the 7950 having greater Gflops capability and 384-bit bus? Hopefully, 48 ROPs is what AMD will be doing with their Sea Islands HD 8870. It's just strange seeing the 960 VLIW5 sp HD 6850 having the same 32 ROPs as the much bigger, vastly more muscular 7970. AMD actually said that Barts XT with 32 ROPs and 1120 sp was 2% faster overall than the prototype with 16 ROPs but with 1280 sp (with the 16 additional ROPs making it yet 2% faster than adding 160 shaders). That was one of the most insightful things an engineer has ever shared, apart from refusing to ever explain WHY a VLIW5 Barts performed so well against the older Evergreen in virtually all games, like as if it were actually behaving like a VLIW4-optimized architecture like its bigger Cayman brother, SPECS-wise. The engineers just refused to explain the front-end magic in that one.
Here's to the future! HD 8870, probably beating Big Kepler to the clock, finally giving us more than 2x the single-GPU performance of 5870! I don't expect Sea Islands to be faster than the BigK, but I will never know for sure until..!! What if there's just some sick magic in it that allows a 410mm^2 GPU to beat a 550mm^2 goliath during this difficult 28nm round?!?
Heck, AMD could have easily done it with a beefed-up RV790XT"X" that had 960sp and 32 ROPs, slinging a rock in between the eyes of the 576mm^2 GTX 280 goliath, the biggest GPU ever made! Why, oh why ONLY a re-designed RV770 into a RV790 with millions more trannies, but nothing except for 100Mhz increase?!? Sometimes, all a knockout takes is a 100% fully-powered punch, not one done with 90% of the energy.
Facebook
Twitter