[gkar1]

From the review comments thread: http://www.techpowerup.com/forums/sh...4&postcount=12

Originally Posted by W1zzard

yeah i think driver optimizations. just waiting for a final whql driver that supports all new amd cards so i can rebench

[CarstenS]

Originally Posted by mczak

So it's a bit unfair saying it doesn't scale as well with CUs than clock[...]

I've run the HD 7950 through our parcours at 7970 clocks, those results are what led me to believe the CU scaling is still lacking compared to Nvidia for example. But then, Nvidia has to scale at max to 16 units, whereas AMD has to distribute to as much as 32. But then again, AMD has had 20 SIMDs/CUs and more for two and a half years. But then again, Nvidias scaling was less impressive in GT200, which also had a higher unit count. But then...

[mczak]

Originally Posted by CarstenS

I've run the HD 7950 through our parcours at 7970 clocks, those results are what led me to believe the CU scaling is still lacking compared to Nvidia for example. But then, Nvidia has to scale at max to 16 units, whereas AMD has to distribute to as much as 32. But then again, AMD has had 20 SIMDs/CUs and more for two and a half years. But then again, Nvidias scaling was less impressive in GT200, which also had a higher unit count. But then...

I think 7850 at 7870 clocks would be interesting. In theory it should show more of a difference than 7950 at 7970 clocks. Haven't seen that anywhere, but judging by overall scores the 7870 is just barely 20% faster than the 7850, so considering the clock difference it looks like those 25% percent more CUs indeed don't do much again (probably less than 10%, ideally it should be around 12.5% considering the the area these CUs use).
But I do expect kepler scaling to be less impressive than fermi, simply because they will also have "more than enough" SMs (or argued the other way round, fermi SM scaling was impressive because they had a lack of shader power compared to the capabilities of the rest of the chip).

[Ryan Smith]

Originally Posted by Psycho

Anyone who retested the 7900 with the same drivers as used for 7800? (and especially for BF3, which itself received 7000 series improvements (path changes) in the meantime).
7870 really seems too close to the 7950.

Intrestingly enough, the 7000 series did not benefit in any way from the BF3 patch, at least not at the settings we use. Performance was +/- 1% in all benchmarks on all 7K cards.

[Silent_Buddha]

I have to say I'm extremely impressed with the 7870. Offering roughly GTX 580 performance (just slightly lower on average) at 5850 levels of power conumption.

Or, ~100-150 USD cheaper (depending on rebate) and over 100/20/30 watts less at the wall under load/idle/long idle than GTX 580. Nvidia really needs Kepler soon, as GTX 580 is completely irrelevant in the market now.

And Performance is close enough to 7950, that I'm not sure 7950 justifies the 100 USD price premium. Perhaps we'll see the 7950 pull further ahead in future Dx11 games utilizing more compute and shader resources.

Overclocking while good is disappointing compared to 79xx. It appears that 78xx has the same max clockspeed, but has a higher base clockspead so the differential is less.

It does make me agree with mczak that AMD might have lost out on an opportunity to have 7970 clocked at 1 ghz base clockspeed so that the entire 7xxx range could be marketed as Gigahertz cards.

And with the small die size, there's plenty of room for price manuevering if they need to respond to Nvidia. 1 GB cards also being an option if they need to reduce price depending on what Nvidia comes out with.

OC editions will be interesting as the OC'd 7870 appears to spank the GTX 580 in many games.

Regards,
SB

[Squilliam]

SB Maybe the reason why the 7950 and 7870 are so close is that the former was benchmarked on older drivers?

[Silent_Buddha]

Games that aren't as reliant on compute or shader power will do relatively better on the 7870 due in large part to the higher clock speed. Driver's aren't going to do much about that.

After all 78xx and 79xx have the same number of ROPs and same tesselation hardware. So anything heavily reliant on those will take advantage of the higher base clockspeed of the 78xx to close the performance gap. So for example in pure tesselation power, the 7870 is faster than the 7950, but considering the rest of the architecture on it's own won't allow it to pull ahead. But things like that will help to close the performance gap.

You can see this at Anandtech. I believe they always rebench all cards with the newest drivers for example, and there's a few games where that is the case. And one where the 7870 is actually faster than the 7950 at 1920x1200. In that one case clockspeed appears to be what's putting it ahead at the lower resolution but once you get to 2560x1600, the advantages of the 7950 starts to come through, albeit not enough for it to pull ahead by much.

I'd suspect that if you clocked both cards at the same clockspeed the 7950 would manage a consistent lead across the board. And considering the 79xx and 78xx cards have roughly the same max overclock, when overclocking the 7950 should always be consistently ahead.

Regards,
SB

[Psycho]

Originally Posted by Ryan Smith

Intrestingly enough, the 7000 series did not benefit in any way from the BF3 patch, at least not at the settings we use. Performance was +/- 1% in all benchmarks on all 7K cards.

Strange.. would expect it to simply switch to the "nvidia" path for 7xx0 in some places:

Originally Posted by Dave Baumann

Additionally we are finding things out right now such as the VLIW opimised FXAA code that DICE graciously implemented under an AMD path is now not as optimal as the NVIDIA FXAA path with GCN!

Or maybe it's ONLY the fxaa shader, but looks like you tested both with and without FXAA-high.
Repi? (who is probably too busy at GDC)

[Lightman]

Originally Posted by Ryan Smith

Intrestingly enough, the 7000 series did not benefit in any way from the BF3 patch, at least not at the settings we use. Performance was +/- 1% in all benchmarks on all 7K cards.

Isn't that because more recent launch driver for HD7970 had an 'optimization' for BF3 forcing quicker nVidia path anyway? If you still have old launch driver it would make sense to test it in patched BF3 now and compare results to unpatched game.

I remember Dave Baumann specifically mentioning this optimization as a reason for updated review driver just few days before official launch.

[Ryan Smith]

Originally Posted by Lightman

Isn't that because more recent launch driver for HD7970 had an 'optimization' for BF3 forcing quicker nVidia path anyway?

I suspect that's the case. AMD's BF3 performance jumped a bit more than 3% with their first driver update.

[Dave Baumann]

The driver can't force a different path in an app.

[Davros]

perhaps you can use 3danalyse to con bf3 into thinking you have an nv card

[silent_guy]

Originally Posted by Davros

perhaps you can use 3danalyse to con bf3 into thinking you have an nv card

Much easier to just replace the shader?

[trinibwoy]

Today is retail day right? What's the word on availability?

[Tchock]

2 OCed versions sold out on NE, 1 stock and 1 Gigabyte's ugly triple breasted OC version going for 359.

Nothing here at base MSRP though, but we'll see how it goes in a few days.

[AlphaWolf]

Originally Posted by trinibwoy

Today is retail day right? What's the word on availability?

Limited (less than a full spectrum of partners seem to have product atm). Although I see 1.1Ghz gigabyte models available. Many places just have pre-order listed.

[Albuquerque]

Eww, that Gigabyte model IS ugly.

I was almost entirely convinced that I was gonna buy a 7950, but then saw the benches for the 7870. So then I'm almost entirely convinced that I was gonna buy a 7870, and now I see the benches for Kepler. And then we start seeing rumors about 670Ti, and FFFFFUUUUUUUUUUUUU.....

I love that we are finally starting to see competition on the horizon, and despite my lack of decision making skills, I'm very happy that my decision is getting harder.

[Davros]

kepler ?
http://www.ggmania.com/?smsid=32621

[homerdog]

I would like to see more reviewers putting older cards in their reviews. For a card like this, I would like to see my current GTX260 thrown in there to see how it compares. Surely there are a lot of folks like me who have aging mid-high end GPUs and are looking to upgrade to something more modern, but so many reviews just compare to cards that are hardly more than a year or two old.

[Kaotik]

Originally Posted by homerdog

I would like to see more reviewers putting older cards in their reviews. For a card like this, I would like to see my current GTX260 thrown in there to see how it compares. Surely there are a lot of folks like me who have aging mid-high end GPUs and are looking to upgrade to something more modern, but so many reviews just compare to cards that are hardly more than a year or two old.

The problem is that reviewers want to use DX11 games / settings, which the GTX260 for example can't do

[UniversalTruth]

Well, you can indirectly compare your card to those new ones. I'll put your card in the 7750-7770 performance range. Which is not very bad... but not good either.

[AlphaWolf]

No, it puts it in the 'can't execute that code' range for any DX11 titles.

[UniversalTruth]

Originally Posted by AlphaWolf

No, it puts it in the 'can't execute that code' range for any DX11 titles.

Honestly speaking, it's not a big loss, if any at all. First, a tiny fraction of all applications out there in the wild do strictly require a DX11 hardware only. And then- the image quality enhancements in DX11 mode are negligible if seen by naked eyes.

But that's another discussion. Until the new generation of consoles arrives in late 2013 and a good number of new-generation games appear, all this nonsense about DX10, DX10.1, DX11, DX11.1 is essentially worthless.

[Rangers]

Originally Posted by homerdog

I would like to see more reviewers putting older cards in their reviews. For a card like this, I would like to see my current GTX260 thrown in there to see how it compares. Surely there are a lot of folks like me who have aging mid-high end GPUs and are looking to upgrade to something more modern, but so many reviews just compare to cards that are hardly more than a year or two old.

They always do that. I'm interested in seeing just how much my aged Q6600 is holding me back these days, but nobody would dream of benchmarking it anymore, even in a special article.

Tech sites only care about the absolute latest and greatest, to a sometimes annoying degree.

[Davros]

Thas because they fear the mighty Q6600

ps: with a really good cooler (unlike my stock cooler with 1 pin replaced by a nut and bolt) you may be able to get 3.6ghz (G0 stepping)