I'm fortunate enough both to own a GTX580 and have funds available to purchase a 7970, but I won't be. It's a good effort from AMD and I may well end up getting one eventually, but not at launch prices. And certainly not before I've seen what kepler can do. For those here wetting their pants with joy over AMD's latest offering, you go right ahead and put your money where your mouth is, I'll just sit back and watch for a while.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Just seen 7970 Price @ £452.15 inc vat.
- Thread starter Scooby-DoobyDoo
- Start date
As Jokester pointed out, 3GB version should be around £349 and 1.5GB one around £309.
Surely Kepler`s 28nm process will allow similar percentage clock increases over last gen cards. Overclocking easily takes the 7970 behond GTX580 performance in the same way that overclocking a GTX480 made it sail past AMD`s previous gen 4890 cards (the latter difference being much much greater). Just imagine what a 1024 or 768 shader Fermi clocked at 1100MHz plus would do and you can get a rough idea of what Kepler may offer.
The 79xx series cannot be fully judged until the Kepler steps into the ring. 28nm certainly seems to clock well, and that should apply to both camps.
At launch, the 7970 is approximately 20% faster than the GTX580, plus offers good gains through overclocking. When the GTX480 was launched, it outperformed the same generion 58xx cards by a similar margin. The 7970 should really be spanking the GTX580 but it does not, especially in key titles such as BF3.
The 79xx series cannot be fully judged until the Kepler steps into the ring. 28nm certainly seems to clock well, and that should apply to both camps.
At launch, the 7970 is approximately 20% faster than the GTX580, plus offers good gains through overclocking. When the GTX480 was launched, it outperformed the same generion 58xx cards by a similar margin. The 7970 should really be spanking the GTX580 but it does not, especially in key titles such as BF3.
Last edited:
Soldato
This sums up the situation in a nutshell.
The 7970 is a nice step forward from the GTX580 in terms of every useful metric (performance per $, per Watt, per mm^2 die space). But this is absolutely to be expected from a reduction in manufacturing process - particularly one so sharp as the drop from 40nm->28nm. The fact remains that, in a number of ways, the 7970 has not advanced as much as could be expected from previous die-shrinks. The overclockability of the chip is impressive, but again, can only be judged against the equivalent 28nm Nvidia chips.
Still, AMD have managed to bring their high-end 28nm cards to market a few months ahead of Nvidia. That in itself is an achievement, and I have no problem with AMD pricing the cards where they are. They are the fastest cards around, and they should be priced as such. I expect sharp price drops once Kepler arrives however, as the Nvidia cards look set to offer significantly better performance.
Soldato
Performance might improve when the 12.1 or later drivers appear, maybe they are waiting to see what hand Nvidia plays with Kepler first.
Remember GCN Architecture is new, & will offer more flexibility, also the 7xxx are DirectX 11.1 ready, so AMD might still have a few surprises for us.
http://www.anandtech.com/show/4455/amds-graphics-core-next-preview-amd-architects-for-compute/2
http://developer.amd.com/afds/assets/presentations/2620_final.pdf
Remember GCN Architecture is new, & will offer more flexibility, also the 7xxx are DirectX 11.1 ready, so AMD might still have a few surprises for us.
http://www.anandtech.com/show/4455/amds-graphics-core-next-preview-amd-architects-for-compute/2
http://developer.amd.com/afds/assets/presentations/2620_final.pdf
Last edited:
Looking into a crystal ball Kepler will be fast and expensive. The top GTX680 part will be a significant improvement over GTX580, providing 50-80% real world gains (at high res and settings). The lesser GTX670 will no doubt be shader and memory bandwidth hobbled, but shall offer ~80% of GTX680 performance for two-thirds of the price. Like with the GTX4xx vs 58xx battle, the GTX680 will reign supreme, and the GTX670 should match or slightly exceed the 7970. Priced at £450 & £300 (680/670) AMD will be forced to cut prices and may once again become the budget king, although NVidia will hold the crown.
NVidia`s advantage is that they already know the oppositions cards, and that it carries a performance advantage over from the last gen battle. AMD`s advantage is being first to market and creaming the cash until competition arrives.
The question is not whether Kepler outperforms 79xx, but rather by how much and whether it becomes embarrassing.
Quote me again in 3 months (or whenever Kepler arrives) and flame as required.
NVidia`s advantage is that they already know the oppositions cards, and that it carries a performance advantage over from the last gen battle. AMD`s advantage is being first to market and creaming the cash until competition arrives.
The question is not whether Kepler outperforms 79xx, but rather by how much and whether it becomes embarrassing.
Quote me again in 3 months (or whenever Kepler arrives) and flame as required.
Last edited:
Associate
I think I will wait for Maxwell!
Kepler is already in production, it's stupid to think that they're using current data on AMD's cards to improve their next gen.
Also looking at my crystal ball, Kepler might turn up as bad as Fermi at launch.
Guess work doesn't work here unless you have any data available.
Soldato
Looking at my crystal ball, I forsee Intel coming along with a line of GPU's and wiping the floor with both the red and green teams.
I can hope
I can hope
people saying wait for kepler, as only a few months away, You could be saying the same when thats out as amd might have a updated version out of gcn a few months away aswell.
Its a cat and mouse game, you are better just to get what currently best out there for you and your budget when you have money to upgrade rather this constant waiting for next release.
Its a cat and mouse game, you are better just to get what currently best out there for you and your budget when you have money to upgrade rather this constant waiting for next release.
It is stupid to assume that NVidia will not tailer clocks and shader count (locked/unlocked) to outperform the oponent. Fermi was not bad. Sure initial yeilds were poor and run a tad hot, but the performance was there and TSMC could be blamed for the process problems.
I make a living out of statistical analysis and guesswork, and I am often incorrect.
Last edited:
Associate
Me too tbh.
Soldato
what he said ^
Here are the 7970 vs 580 1.5GB vs 6970 results for the games I play at 1920x1200 (highest settings). Results have been taken from a variety of reviews (not cherry picked).
Overall, the 7970 is 20% faster than the 580 and 40% faster than the 6970. Given that the 7970 has 60% more transistors, 50% more memory bandwidth, plus a small clock increase over the 6970, I find it slightly disappointing. Against the 580 it has 40% more transitors, 35% more memory bandwidth and 18% higer clock speed.
Overall, the 7970 is 20% faster than the 580 and 40% faster than the 6970. Given that the 7970 has 60% more transistors, 50% more memory bandwidth, plus a small clock increase over the 6970, I find it slightly disappointing. Against the 580 it has 40% more transitors, 35% more memory bandwidth and 18% higer clock speed.
Soldato
clock all GPUs to their max though and you'll see how much better the 7970 really is
Associate
True, but given the overclocking headroom of the 7970 we will inevitably see a revised model to compete with nVidia's next offering. And nVidia typically charges more, so AMD will still have plenty of room to manoeuvre with regards to pricing - worst comes to the worst they'll simply drop their prices.
I'm more interested to see how Metro: Last Light turns out. If it launches around the time of nVidia's new cards then it will be the definitive benchmark title and given nVidia's edge with the original it will be very interesting to see.
But then AMD will be reliant solely upon clockspeed to pull clear of the opposition rather than clever design. NVidia's new 28nm cards will also share this additional clockspeed benefit.
What I am trying to get at is that if you slow down the 7970 to GTX580 speeds (782core/1000mem), performance will be almost identical. AMD's current advantage over the 580 comes mostly from clock speeds rather than core design. If NVidia scales the 580's core transistor count by 50% (possibly 100% according to some rumors), increase memory bandwidth to 512bits (as expected), and hits similar clocks to the 7970, there will be no contest.
Clock for clock there is very little performance difference between the 580 and 7970. NVIdia could possibly get very close to 7970 performance by simply shrinking the existing 580 die and taking advantage of 28nm clockspeads, but they won't be doing that.
Last edited:
You're making assumptions.
"What if" is a game of hit and miss, I don't expect GTX 680 to be slower than HD 7970 but at max core speed it may just trade blows.
We don't know if Kepler solves all heat output issues that are the main limitation in developing the fastest GPU and even though you claim tailoring GPU clock speed will give Nvidia advantage (because, as you say, they're similar clock for clock with the current GTX 500 series), you can just assume
Also going with higher bus speed will increase the cost of production inevitably. AMD can play a price war game at any point.
That's a very odd argument. Nvidia have two different clock speeds- one for the core and one for the SP which usually run at double the core. The architectures are too different for a clock to clock comparison to be meaningful.