I'm trying to do something a bit different and add a little value here, by instead of simply saying "look at this website or that website for video card comparisons", looking at two different websites and comparing their performance data in a high feature and performance demand scenario.
I know my study is WAY overkill, but I was bored...but I was also interested in this.
It was triggered by a post from Meats_of_Evil started in this thread at" http://www.hardwareanalysis.com/content/topic/77569/?o=20#592302
I started this thread so as to not hijack angryhippy's thread.
IMO, my study tends to confirm that test protocols are absolutely critical in assessing a card's relative performance and can result in massive differences in "which card is better" conclusions. You can't just take an arbitrary comparison benchmark, even from a "quality" website, and assume you will get information that applies to your planned usage.
With the cost of some of these cards, that can be really important to know.
I decided to look at the tomshardware rankings MOE brought up but specifically
for a gaming scenario with high quality settings
since that would probably be of more interest to HWAers where we're looking at using the more expensive video cards.
I was surprised that when limited to this specific gaming scenario data, the tomshardware RELATIVE RANKING results compared very nicely with the Passmark overall card rankings
no meaningful anomolies.
For a more fair comparison, I TRIED to exclude over-clocked cards data from the list. Not sure I got them all. I also tried to exclude unusually low-ranked and isolated cards as being likely "budget" versions of the typical implementation of a given card. For example, if there are 3 GTX 550Ti cards ranked considerably higher than a lone GTX 550Ti, I used a value from the higher-ranked cards to establish the comparison ranking point for that card (as long as they weren't overclocked/turbocharged/etc). If one had access to the details of the specific card, one might find the card had less or slower memory and/or was clocked lower than the "standard" card model to reduce costs.
With just a few anomolies, almost all the cards appear to pretty much hold consistently cross-ranked positions between Passmark and tomshardware, or any difference in the raw performance was so close it could be easily considered noise introduced by the test protocols.
Note that on tomshardware, the HD6990 performance (82) is almost TWICE the performance (47) of the GTX 580 AND
TWICE the performance (45) of the HD6970. This relative performance is consistent with a couple other sites I examined in which high game quality settings were involved (e.g. 2560x1600, 4X-AA, 16X-AF). However, Passmark rates the HD6990 "overall" performance as lower than the GTX 570. I conclude from this that the HD6990 and its drivers are extremely well optimized for high-quality game playing.
Note: raw performance values and rankings can change as data are updated on the respective websites.
To help show the consistency (or any lack thereof) between Passmark and tomshardware ratings (for their specific test scenarios) I created a baseline performance comparison by calculating the percentage performance of each card self-relative to the GTX 580. The closer the percentages match between tomshardware and Passmark for a given card, the more consistent the results. If the percentages differ significantly, it suggests there is an unknown difference in card specification or the test protocol introduced an anomoloy.
But keep in mind that what I
am MOST interested in is whether the relative RANKING
of the cards is consistent between Passmark and tomshardware.
Passmark raw performance ratings are in xxxx format, while tomshardware are xx.xx format.
(Data from 20110611 or 20110612)
tomshardware metrics were for THIS SPECIFIC scenario. Note: tomshardware didn't provide a rank
Aliens vs. Predator - Enthusiast
Maximum Settings (1920x1080, 4xMSAA, 16xAF, SSAO)
3867 01 GeForce GTX 580 (100% of Passmark GTX 580).
47.00 * Geforce GTX 580 1,53 GB GDDR5 (100% of tomshardware GTX 580)
3549 02 GeForce GTX 570 (92% of GTX 580)
41.00 * Geforce GTX 570 (87% of GTX 580)
3236 06 Radeon HD 6990 (83% of GTX 580)
* Radeon HD 6990 4GB (174%
of GTX 580)
3132 07 Radeon HD 6970 (81% of GTX 580)
42.50 * Radeon HD 6970 2GB (90% of GTX 570)
(Note: the tomshardware being 90% v. 81% is probably
due to tomshardware card having the 2GB RAM config)
3108 08 Radeon HD 6950 (80% of GTX 580)
37.90 * Radeon HD 6950 1GB (81% of GTX 580)
2955 10 GeForce GTX 560Ti (76% of GTX 580)
32.70 * Geforce GTX 560 Ti (70% of GTX 580)
2843 12 Radeon HD 6850 (74% of GTX 580)
27.60 * Radeon HD 6850 1GB (59% of GTX 580)
2664 17 GeForce GTX 560 (69% of GTX 580)
31.70 * Geforce GTX 560 1GB GDDR5 (67% of GTX 580)
2501 22 GeForce GTX 465 (65% of GTX 580)
no entry, tomshardware (n/a)
2497 23 Radeon HD 5850 (65% of GTX 580)
27.60 * Radeon HD 5850 1 GB (59% of GTX 580)
2353 25 GeForce GTX 460 (61% of GTX 580)
24.90 * Geforce GTX 460 1GB (53% of GTX 580)
21.90 * Geforce GTX 460 768 MB (47% of GTX 580) (included to show difference likely due to RAM config)
no entry, Passmark
23.50 * Radeon HD 6790 1GB GDDR5 (50% of GTX 580)
1864 40 GeForce GTX 550Ti (48% of GTX 580)
22.30 * Geforce GTX 550 Ti (47% of GTX 580)
1658 48 Radeon HD 5770 (43% of GTX 580)
19.80 * Radeon HD 5770 1GB (42% of GTX 580)
1442 56 GeForce GTS 450 (37% of GTX 580)
17.10 * Geforce 450 GTS
1 GB GDDR5 (36% of GTX 580)