When will we start seeing more reviews showing time to render though? Couldn't they just use frapscalc and cap the game at 30 fps (to see what happens when the frame rates are to low), 60 FPS and 90 FPS (even if you have to reduce the resolution to get it). Just to see what the results would show. It wouldn't be a test to see which CF/SLI setup is the fastest. It would be to only show how long it took to render.
AMD: Southern Islands (7*** series) Speculation/ Rumour Thread
- Thread starter UniversalTruth
- Start date
The marking is "ENG" I.e. engineering sample. This should not have been sent for review.
HardWare.fr has a review of the GTX 690 here and, interestingly, the HD 7970 is now faster than the GTX 680.
Damien says that's because:
So going by HFR's results, the GTX 680 used to be faster and cheaper, now the HD 7970 is faster and cheaper. Gotta love competition.
Damien says that's because:
- He's now using a retail GTX 680 instead of a press sample,
- AMD has finally fixed their MSAA problem in Batman Arkham City,
- NVIDIA has a problem with Total War Shogun 2,
- Because it's a review for a dual-GPU card, he's used very high settings, with which the GTX 680 has trouble (memory bandwidth & capacity constraints).
So going by HFR's results, the GTX 680 used to be faster and cheaper, now the HD 7970 is faster and cheaper. Gotta love competition.
Silent_Buddha
Legend
Glad to see he's using a retail card instead of an Nvidia provided card. Still going to be a matter of luck whether he gets a high clocking representation or a lower clocking representation of the card, however.
Basically makes most reviews with GTX 6xx relatively worthless, IMO.
Regards,
SB
Basically makes most reviews with GTX 6xx relatively worthless, IMO.
Regards,
SB
In 10 years i have never get so much headache reading the " different reviews" for compare them... ( i like read different ones, like that i get a better figure ), But theres so much difference from one to another.
The result on Hardware.fr are really different of what you can find on Hardwarecanuck ( for the example, and it was the first name who was come to my mind. ).
I was going to post this one in xtremesystem, and finally decided to not... i will be call for trouble it seems. lets wait someone else do it.
The result on Hardware.fr are really different of what you can find on Hardwarecanuck ( for the example, and it was the first name who was come to my mind. ).
I was going to post this one in xtremesystem, and finally decided to not... i will be call for trouble it seems. lets wait someone else do it.
Last edited by a moderator:
Random number between 7970 and 690 pricing.
I think they'll just go for twice the price of the 7970, i.e. $899.99, maybe even $949.99 if it really has "good" power-efficiency. They will probably clock it just high enough to beat the 690—otherwise, what would be the point?—and call it a day.
Beating the 690 would require at least same clocks as HD7970 (of course depending on the actual games used in benchmarking but that's about the lowest I could see it happen as an average).
Given that 2x7970 would have somewhat too high power draw but the 7970 clearly has some headroom left for lower voltage that seems doable but probably not much more.
I'd tend to agree only if the lower clocks were true for the majority of cards sold in retail.
FWIW - reviewers get send selected samples every time. Our 680 hits 1124 until 70°C where it drops to 1093. Our 7970 otoh is way more silent than any retail card with reference cooler we've had in the lab.
UniversalTruth
Veteran
That's exactly the price range where both 6990 and 590 were sold at. Nothing random.
And also, fixing the awful micro-stuttering crossfire is characterised with, and the applications' support out of the gate, not fixes every now and then with new CAPs, etc
Last edited by a moderator:
check reviews of GTX 690, micro-stutering is still present; it isn't an AMD's speciality
Given AMD's upcoming HSA plans and the fact that the Tahiti die - even after taking all the GPGPU stuff into account - still seems somewhat big, I actually hope for some kind of chip-to-chip interconnect and/or unified memory surprise in HD7990.
A dual-GPU card that actually works like a single GPU-card, reaching nearly perfect scaling across all games (dropping the need of game specific crossfire profiles for good), featuring 6GB of shared address space (not 2x3GB) and solving all those bothersome dual-GPU micro stuttering alternate frame rendering problems and inconsistencies - now that would be card actually worth $1000. :smile:
Computerbase: HD 7970 underclocked to GTX 680 bandwidth looses 7% performance in avg.
So 256-Bit versions of HD 7950/7970 could be also good performers, especially if GPU clock is raised a bit.
They would be also better positioned against NVs cheaper GK104 SKUs.
So 256-Bit versions of HD 7950/7970 could be also good performers, especially if GPU clock is raised a bit.
They would be also better positioned against NVs cheaper GK104 SKUs.
HardWare.fr has a review of the GTX 690 here and, interestingly, the HD 7970 is now faster than the GTX 680.
Damien says that's because:
- He's now using a retail GTX 680 instead of a press sample,
- AMD has finally fixed their MSAA problem in Batman Arkham City,
- NVIDIA has a problem with Total War Shogun 2,
- Because it's a review for a dual-GPU card, he's used very high settings, with which the GTX 680 has trouble (memory bandwidth & capacity constraints).
So going by HFR's results, the GTX 680 used to be faster and cheaper, now the HD 7970 is faster and cheaper. Gotta love competition.
Same here: http://ht4u.net/reviews/2012/nvidia_geforce_gtx_680_kepler_sli_evga_zotac_test/index56.php
Seems like the HD7970 is a tad faster than the GTX680 when not using press samples. How's that one might ask? Did they give out the cards with the highest turbo clk? Seems kinda shady
this make zero difference in 2 games listed ( Skyrim, Dirt3 ), but in some case it goes to 10fps ( 64 > 54fps ). ( so the memory speed look even more important of the pure 3gb 384bits and high bandwith ) hard to say if it is only the memory speed, or in the case of Sykrim and BF3 ( high memory usage ), the 3Gb available negate this too.
When you bench with the 7970 ( CFX or single ), you clearly see if you only increase the clock speed, passed beyond a certain point, the scaling decrease.. you need up the memory according...
I believe 1520mhz is good till 1200-1250mhz, passed this point and specially over 1300mhz you want to increase to 1600-1700mhz the memory. Its really a parrallel scaling.
I can be wrong i just base myself on the constation made when overclocking them..
The thing is when AMD have release the 7970, they have choose really conservative core and memory speed.. they could have set it at 1005mhz and 6000mhz without worrry.. Now they had to make a choice, and the different Asic, TDP, production etc have push them in this direction. Without saying, i think they have really think the lineup, from the lower 7770 to 7970 from start... and so decided the 1ghz will be for the 7870 ..
When you bench with the 7970 ( CFX or single ), you clearly see if you only increase the clock speed, passed beyond a certain point, the scaling decrease.. you need up the memory according...
I believe 1520mhz is good till 1200-1250mhz, passed this point and specially over 1300mhz you want to increase to 1600-1700mhz the memory. Its really a parrallel scaling.
I can be wrong i just base myself on the constation made when overclocking them..
The thing is when AMD have release the 7970, they have choose really conservative core and memory speed.. they could have set it at 1005mhz and 6000mhz without worrry.. Now they had to make a choice, and the different Asic, TDP, production etc have push them in this direction. Without saying, i think they have really think the lineup, from the lower 7770 to 7970 from start... and so decided the 1ghz will be for the 7870 ..
Last edited by a moderator:
Similar threads
