AMD: R9xx Speculation

10% might be more accurate.

 

Im not basing this on one tech site. Im basing this on 10+ reviews which average out to those numbers.

 

It all depends on the resolution of choice, 1680x1050 could be that 15%, even 1920x1200 might be close to it, but 2560x1600 on the other hand is a lot stronger for the Radeon, balancing the average to around 10%, IMO

 

Radeon HD6970 is a ultra high resolution (2560 X 1600 and beyond to Eyefinity resolutions) performer (where GTX 580 is only slighly faster, even at the same level in some popular applications like Crysis) and those averaged tests including resolutions like 1024X768 and similar are just dust and have to be ignored. So in this respect, I think 6970 is a solid product, and with new drivers it should shine even more.

Now @newegg Radeon HD5850 is already set at the 170 USD price tag, so with the launch of GTX 560 we should see even lower prices and also for some other products, like the 6950 and 6970. The competition is so good thing. Only if there are no agreements between the competitors to keep the prices high, which is rather sad and unfair for the customers.

 

Well I got my 6950 today.

Not sure if there are any reviews up with the 10.12a but there seems to be a definite big performance difference in Crysis at least.
Note: My Crysis isn't stock, running some 'prettier visuals' mod that I installed ages ago & couldn't even name anymore so others results may vary but I chucked in my 5770 results from earlier in the day too.

Benchmark_GPU
10.11 from driver CD

Spoiler

Code:

==============================================================
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
    Play Time: 67.03s, Average FPS: 29.84
    Min FPS: 25.88 at frame 140, Max FPS: 46.98 at frame 1013
    Average Tri/Sec: -23912640, Tri/Frame: -801473
    Recorded/Played Tris ratio: -1.14
!TimeDemo Run 1 Finished.
    Play Time: 53.63s, Average FPS: 37.29
    Min FPS: 25.88 at frame 140, Max FPS: 47.56 at frame 1004
    Average Tri/Sec: -29422840, Tri/Frame: -788982
    Recorded/Played Tris ratio: -1.16
!TimeDemo Run 2 Finished.
    Play Time: 53.61s, Average FPS: 37.31
    Min FPS: 25.88 at frame 140, Max FPS: 47.56 at frame 1004
    Average Tri/Sec: -29446210, Tri/Frame: -789316
    Recorded/Played Tris ratio: -1.16
!TimeDemo Run 3 Finished.
    Play Time: 53.60s, Average FPS: 37.32
    Min FPS: 25.88 at frame 140, Max FPS: 47.56 at frame 1004
    Average Tri/Sec: -29429798, Tri/Frame: -788670
    Recorded/Played Tris ratio: -1.16
TimeDemo Play Ended, (4 Runs Performed)
==============================================================

10.12a

Spoiler

Code:

==============================================================
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
    Play Time: 62.38s, Average FPS: 32.06
    Min FPS: 27.54 at frame 164, Max FPS: 61.45 at frame 1635
    Average Tri/Sec: -25710632, Tri/Frame: -801969
    Recorded/Played Tris ratio: -1.14
!TimeDemo Run 1 Finished.
    Play Time: 43.17s, Average FPS: 46.33
    Min FPS: 27.54 at frame 164, Max FPS: 63.48 at frame 1637
    Average Tri/Sec: -36595932, Tri/Frame: -789850
    Recorded/Played Tris ratio: -1.16
!TimeDemo Run 2 Finished.
    Play Time: 42.96s, Average FPS: 46.56
    Min FPS: 27.54 at frame 164, Max FPS: 64.87 at frame 1659
    Average Tri/Sec: -36749028, Tri/Frame: -789318
    Recorded/Played Tris ratio: -1.16
!TimeDemo Run 3 Finished.
    Play Time: 42.92s, Average FPS: 46.60
    Min FPS: 27.54 at frame 164, Max FPS: 64.87 at frame 1659
    Average Tri/Sec: -36789168, Tri/Frame: -789525
    Recorded/Played Tris ratio: -1.16
TimeDemo Play Ended, (4 Runs Performed)
==============================================================

5770 with 10.12

Spoiler

Code:

==============================================================
TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
    Play Time: 75.97s, Average FPS: 26.33
    Min FPS: 19.80 at frame 1945, Max FPS: 39.35 at frame 1017
    Average Tri/Sec: -21089428, Tri/Frame: -801033
    Recorded/Played Tris ratio: -1.14
!TimeDemo Run 1 Finished.
    Play Time: 64.96s, Average FPS: 30.79
    Min FPS: 19.80 at frame 1945, Max FPS: 43.95 at frame 76
    Average Tri/Sec: -24299354, Tri/Frame: -789302
    Recorded/Played Tris ratio: -1.16
!TimeDemo Run 2 Finished.
    Play Time: 66.97s, Average FPS: 29.86
    Min FPS: 19.80 at frame 1945, Max FPS: 43.95 at frame 76
    Average Tri/Sec: -23558378, Tri/Frame: -788842
    Recorded/Played Tris ratio: -1.16
!TimeDemo Run 3 Finished.
    Play Time: 65.84s, Average FPS: 30.37
    Min FPS: 19.25 at frame 1949, Max FPS: 44.46 at frame 79
    Average Tri/Sec: -23932526, Tri/Frame: -787903
    Recorded/Played Tris ratio: -1.16
TimeDemo Play Ended, (4 Runs Performed)
==============================================================

 

Depends, if you game at low resolutions and up to 4x AA, then GTX480/GTX580 may be 15% faster or more compared to 5870/HD6970. Common sense though, why would you need such cards then? If you game with them as intended, 1920/2560+ with 8x AA or so, then difference is negligible, up to 5%. If you go for even higher res, then 6970 CF or even 6950 CF spanks GTX 580 SLI.

Inconsistency of FPS increase has to be immature drivers, just watch perf increase of 5800 gen. over the last year, and thats pretty much same arch. as previous just with extra features, while Caymans is a new generation. To expect sizeble gains is common sense. Not going to claim that 6970 will be faster than GTX580 in the 2H 2011, but I doubt there will be much difference between them either, if their data in 2560 8x AA is already within few %.

 

Cayman's performance inconsistency seems to have much to do with nVidia's developers program.

Look at Lost Planet 2 and HAWX 2 for instance, AMD's performance are just abysmal here and Cayman didn't improve that much over Cypress, meaning Cypress' bottleneck in these games hasn't changed.

Most of the performance difference seen between Cypress and Cayman seems to come from the increased texturing throughput, perhaps the improved RBE's buffering too. There's nothing proving any driver immaturity, rather the opposite.

 

If you have calculated such an average, I'd like to see the exact numbers and reviews you've used.

 

Seems more than that to me. In some games 6970 is 40-50% behind. In few does it get closer than 5-10%.

 

It may be common sense to ask that question, but it raises questions over why frame-rate minima or "CPU bottlenecks" or "driver bottlenecks" or "geometry bottlenecks" are making NVidia better at lower pixel counts. Those bottlenecks, whatever they are, are real and are worth exploring.

Of course no-one actually explores those - except for Mintmaster's recent Dirt2 regression. Which, for some reason, Dave Baumann has ignored.

 

Unfortunately, it is not very rosy like that , only at 2500x1600 and 8X AA does the HD 6970 catch up to GTX 580 , anything lesser than that , and GTX 580 is about 15% faster .

Also people usually own 1920x1080 displays , not the other way around , so performance at this resolution is very important .

 

I dont know if 150 FPS or 200 FPS is such a victory. Maybe its realy just more driver overhead with ATI-s architecture. In 3Dmarks or other pure benchmarks nvidia doesnt have that advantage with lower pixel count.

Edit: Maybe if sites would also bench with a AMD Phenom setup and not just OC Intel core7 CPU-s.

 

It's framerate minima that spoil game play.

I don't know what you're alluding to in this case, though there are cases where Intel processors really screw with game performance if HT is on:

http://forum.beyond3d.com/showthread.php?t=57620

 

If u check out for example tomshardware review http://www.tomshardware.com/reviews/radeon-hd-6970-radeon-hd-6950-cayman,2818.html than u realize than the minimum framerates of cayman(and other AMD) cards are same than nvidia has to offer.

 

The issue is not "min framerate" but high resolution-independant frame time, although I'm not sure it's real.

In other words, significantly reduced pixel time as you increase resolution.

 

HAWX2 is sort of academic, though, as 2560x1600@4xAA is 60fps on the 6950. It makes a huge difference in averages, though, because the 580 is twice as fast. Some insight can be found at Tomshardware:
http://www.tomshardware.com/reviews/radeon-hd-6970-radeon-hd-6950-cayman,2818-3.html

You can see that Crossfire scaling works fine, as 2x6850 is 102fps, while 1x6870 gets 61 fps with tessellation enabled. With tessellation disabled, however, 2x6850 hits a wall at 126 fps, while 1x6870 jumps to 102 fps. Meanwhile, GTX 580 gets 184 fps with tessellation disabled. It seems like ATI has some substantial overhead in this game unrelated to pixels or vertices.

There's not a lot of data out there on LP2, unfortunately.

 

To be fair, Dave did pop in to say that it was front-end limited. I just wish we knew whether he meant command buffer, driver, or geometry, as they're all pretty much at the front of the pipeline.

 

Could it be all 3? Thus why he said front-end limited.

 

I was thinking about it the other day, and my guess is, Nvidia has undiscriminating brute force approach (need it or not, here comes 200FPS on your 1024x.. LCD!" ), while AMD has more focused both HW and drivers approach, i.e. high-end GPUs like Caymans are specifically tweaked with high resolutions in mind. The higher res you got, the better it is compared to competition. Like an example of 6850 CF beating more than twice more expensive GTX580 SLi in ultra-high res. Or 6970 being ~= GTX580 in 2560x with 8x AA. If you have a small screen, then it doesnt really matter if FPS is 100 or 150, thats probably the line of thinking in AMDs camp (I'm not claiming to know whats in their heads, just it seems that way ).