The GT300/Fermi Thread - Part 2!

[zalbard]

Current ATI cards get castrated when tessellation is enabled.

Quite a strong statement considering the fact that the very top offer from Nvidia isn't even 2 times faster than a cheaper 5870 (not to mention 5970 which is on par with (or better than) GTX480 in Haven and a much slower GTX470) in the over-the-top tessellation environment of Haven, not to mention some people speculating about GTX480 scoring much closer to 5870 with high IQ settings...
You are not being totally rational here.

[Teemax]

Quite a strong statement considering the fact that the very top offer from Nvidia isn't even 2 times faster than a cheaper 5870 (not to mention a much slower GTX470) in the over-the-top tessellation environment of Haven, not to mention some peope speculating about GTX480 dcoring much closer to 5870 with high IQ settings...
You are not being totally rational here.

Agreed that it's such an aggressive statement, coming from a respected reviewer.

Unless it's back with facts, forgive me if I take the statement with a LOT of salt. Somehow, SKYMTL has turned into an NVIDIA's ambassador in this forum. Weird.

[BenchZowner]

Agreed that it's such an aggressive statement, coming from a respected reviewer.

Unless it's back with facts, forgive me if I take the statement with a LOT of salt. Somehow, SKYMTL has turned into an NVIDIA's ambassador in this forum. Weird.

I heard nVIDIA gave him 20k$.

And you've been here for how long to know SKYMTL well ?

[RaV[666]]

From my simple understanding of the tesselation, it seems it could be "tweakable" just like AA or aniso settings are, you could have 2x more polys/4x more polys and etc.It would be a smart move for ati to implement such override in drivers.I wonder if thats possible.

[Teemax]

And you've been here for how long to know SKYMTL well ?

I haven't registered in this forum for a long time, yet, I'm a regular reader of HardwareCanucks. I like the depth of their investigations and the often unbiased, logical arguments for their conclusions.

However, SKYMTL's tone sounds a bit odd for an supposedly neutral reviewer. Claiming that Evergreen has "castrated tessellation performance" is a HUGE claim, he should at least back it up with some evidence.

[Sam_oslo]

Tessellation is one the few and most important new futures introduced by DX11. There is so much new here and it needs an open mind to investigate and understand it without getting offensive toward others ideas ans meanings.

Apparently, nVidia is using a different implementation of tessellation and it can affects both the picture details/quality and the performance. It is important to know the real differences based on a contractive discussion .

[Motiv]

Tessellation is one the few and most important new futures introduced by DX11. There is so much new here and it needs an open mind to investigate and understand it without getting offensive toward others ideas ans meanings.

Apparently, nVidia is using a different implementation of tessellation and it can affects both the picture details/quality and the performance. It is important to know the real differences based on a contractive discussion .

Very true, I expect the G4xx products based on fermie to be monsters at tessellation, physics and any other math intensive part of the DX world. The GPU has been built from the ground up to be incredible at these things.

What it look like it may fall down on, is the standard DX9/10/11 feature set and thus need 'gimmicks' to get people to choose it, in the absence of many DX11 games.

Will be interesting to see ATIs response to the tessellation question in their refresh this year. With more games exploiting DX11, they may need to improve it before next year.

[RaZz!]

Nvidia real demo run vs Nvidia PR slide (Hint. Check 60-120 sec):
*snip*

there's nothing wrong with the PR slide tbh. they clearly state on the slide that it's the "dragon, cobblestone road sequence", which is the sequence from 60-120secs on the demo graph. they picked a tessellation heavy sequence to show fermi's advantage in such scenarios. fair enough.

Very true, I expect the G4xx products based on fermie to be monsters at tessellation, physics and any other math intensive part of the DX world. The GPU has been built from the ground up to be incredible at these things.

What it look like it may fall down on, is the standard DX9/10/11 feature set and thus need 'gimmicks' to get people to choose it, in the absence of many DX11 games.

Will be interesting to see ATIs response to the tessellation question in their refresh this year. With more games exploiting DX11, they may need to improve it before next year.

true, however, even if fermi lacks some raw performance in standard dx9/10/11, future games will all make use of tessellation, where fermi can compensate the lack of raw performance with more tessellation performance (atleast that's what one would expect if the performance rumors are true).

i really hope the gtx480 provides more performance than just a 5% increase over a 5870 though. a 5% increase could easily be surpassed by a high clocked 5890 or smth. all i want is the prices to go down

[Piotrsama]

Nvidia real demo run vs Nvidia PR slide (Hint. Check 60-120 sec):

Good stuff there.

As for the dual GF100, just like G200, it will need a shrink to happen unless Nvidia and TSMC have some magic pixie dust to use on the Bx silicon.

Or maybe they can release some limited Mars (or something), where they jump the PCI-E power limit and use 500W+ on stock for the dual gpu part.

[SKYMTL]

However, SKYMTL's tone sounds a bit odd for an supposedly neutral reviewer. Claiming that Evergreen has "castrated tessellation performance" is a HUGE claim, he should at least back it up with some evidence.

It is a huge claim but see my statement above. Once again, everyone seems to be jumping down on the statement without looking at numbers that actually back it up. Look at DiRT, AvP, Heaven and BF: BC.

What ATI did is a good thing: they concentrated on making a DX11 compatible GPU by adapting an older architecture instead of taking a ton of time designing a newer one. They knew there was no point in concentrating too hard on DX11 so early into its life cycle and they were able to actually beat NVIDIA to market while offering a perfectly suitable solution.

So, yes the HD 5000 series takes a massive hit in DX11 but it was never meant to be their end-all DX11 product anyways. That name goes to their upcoming architecture that will be released when there is a good amount of games that support DX11.

[eric66]

It is a huge claim but see my statement above. Once again, everyone seems to be jumping down on the statement without looking at numbers that actually back it up. Look at DiRT, AvP, Heaven and BF: BC.

What ATI did is a good thing: they concentrated on making a DX11 compatible GPU by adapting an older architecture instead of taking a ton of time designing a newer one. They knew there was no point in concentrating too hard on DX11 so early into its life cycle and they were able to actually beat NVIDIA to market while offering a perfectly suitable solution.

So, yes the HD 5000 series takes a massive hit in DX11 but it was never meant to be their end-all DX11 product anyways. That name goes to their upcoming architecture that will be released when there is a good amount of games that support DX11.

have you looked at dirt 2 benchs ? i don't see awesome dx11 power of fermi series

[SKYMTL]

have you looked at dirt 2 benchs ? i don't see awesome dx11 power of fermi series

As I said: I wasn't comparing ATI's cards to NVIDIA's with my statement. Rather, it was a general observation regarding the HD 5000 series' performance in a DX11 environment. C'est toute.

HOWEVER, supposedly the GF100 series was built with DX11 in mind which could translate into great performance in supporting applications. But that is just a rumor at this point and only time will tell if it translates into better performance.

[BenchZowner]

have you looked at dirt 2 benchs ? i don't see awesome dx11 power of fermi series

have you verified and figured out that those benchmark numbers are real ?

[ since you like playing 20q ]

[Arxagelos]

have you looked at dirt 2 benchs ? i don't see awesome dx11 power of fermi series

no max fps given and don't expect in dirt2 with only water,flags and crowd tesselation to see deference, in heavy tesselated games will see difference if it exist.

[Iconyu]

It is a huge claim but see my statement above. Once again, everyone seems to be jumping down on the statement without looking at numbers that actually back it up. Look at DiRT, AvP, Heaven and BF: BC.

What ATI did is a good thing: they concentrated on making a DX11 compatible GPU by adapting an older architecture instead of taking a ton of time designing a newer one. They knew there was no point in concentrating too hard on DX11 so early into its life cycle and they were able to actually beat NVIDIA to market while offering a perfectly suitable solution.

So, yes the HD 5000 series takes a massive hit in DX11 but it was never meant to be their end-all DX11 product anyways. That name goes to their upcoming architecture that will be released when there is a good amount of games that support DX11.

When you look at Fermi it doesn't shout 'gaming' it shouts 'GPGPU', I'm still not convinced that nvidia's tesselation isn't shader bound. When you have a GPGPU that powerful you can get away with a lot of stuff on syth' benchmarks, as they tend to focus on one thing at a time.

I was excited too when I first saw the numbers, but I need to see where the power comes from, because when you turn off the tesselation in that section of the benchmark there is next to nothing there.

[Chickenfeed]

What ATI did is a good thing: they concentrated on making a DX11 compatible GPU by adapting an older architecture instead of taking a ton of time designing a newer one. They knew there was no point in concentrating too hard on DX11 so early into its life cycle and they were able to actually beat NVIDIA to market while offering a perfectly suitable solution.

So, yes the HD 5000 series takes a massive hit in DX11 but it was never meant to be their end-all DX11 product anyways. That name goes to their upcoming architecture that will be released when there is a good amount of games that support DX11.

Couldn't agree more. Never in history has a launch card using a new API been able to master it first generation. I doubt we will see strong DX11 cards until at least 3rd generation, perhaps 2nd gen refresh best case. That said, what they did with Cypress was a smart buisness move none the less. I'm sure Nvidia will push DX11 now they can call it relavent but if the past repeats itself, I am not buying it (their marketing). Hell I had a 8800gtx launch week for a good 14 months and never felt DX10 was relavent until much later...

[zerazax]

Couldn't agree more. Never in history has a launch card using a new API been able to master it first generation. I doubt we will see strong DX11 cards until at least 3rd generation, perhaps 2nd gen refresh best case. That said, what they did with Cypress was a smart buisness move none the less. I'm sure Nvidia will push DX11 now they can call it relavent but if the past repeats itself, I am not buying it (their marketing). Hell I had a 8800gtx launch week for a good 14 months and never felt DX10 was relavent until much later...

Yep, I had a G80 earlier, and while it crushed DX9 games, it could not run DX10 as well as the 2nd generation of cards (G92s were better, then RV770 & GT200 finally saw DX10 be relevant).

The best bet is always to buy the card that you can use now, and not what you will worry about a year down the line because by then, it will already be outdated/slow

[***Deimos***]

Couldn't agree more. Never in history has a launch card using a new API been able to master it first generation. I doubt we will see strong DX11 cards until at least 3rd generation, perhaps 2nd gen refresh best case. That said, what they did with Cypress was a smart buisness move none the less. I'm sure Nvidia will push DX11 now they can call it relavent but if the past repeats itself, I am not buying it (their marketing). Hell I had a 8800gtx launch week for a good 14 months and never felt DX10 was relavent until much later...

Mr obvious. Ofcourse future cards are faster with better features - why else would folks upgrade and buy them?

And I think you underestimate the ENORMOUS effort in architecture, design, floorplanning, validation, etc to get even chips as similar as 9800GT and 9600GT made. FYI 5800fx and 7900GTX were both DX9, but they are very very different.

its getting ridiculous how many people think gf100 does not have fixed function tessellation. if they are that incompetent maybe they should hire people off of tech forums to architect their gpu's. most people probably dont know the difference from fixed function logic or programmable logic anyways.

what they did was fairly simple. gf100 basically is setting up the scene in parallel compared to serial setup of other gpus. it works well for all of the small triangle tessellation creates.

Does it really matter? The means to an end? R600 doesn't have either 2D core or correctly working AA hardware. Yet you can surf web and play games with AA.

Obviously fake

lol, my memory bandwith is better

Ghz -> GB = 1000000 / 1024^2 = 0.9313 GiB
512bit DDR = 512 * 2 / 8 = 128 (384=96, 320=80)
For GDDR5 double the clock rate shown, because GDDR5 fetches twice #bits at a time.

(if you dont use the Ghz->GB conversion factor, you will get same number as GPU-Z)
So, for your GTX295: (512/4) x 1.512Ghz = 128 x 1.512 x 0.9313 = 180.2 GigaBytes/s (GiB/s)
GTX480 from screenshot would be: (384/4) x 1.8 x 0.9313 = 160.9 GigaBytes/s (GiB/s)
For reference 5870: (256/4) x 2.4 x 0.9313 = 143.0 GigaBytes/s (GiB/s)

Almighty GTX480 only a smidgen ahead ... pff.. marketing PR wont stand for that.
Using same clocks as 5870, 480 will be 50% more, or 214.6GiB/s (or 230GB/s in marketing speak - very close to 5970's 256GB/s!!)

[Motiv]

I heard nVIDIA gave him 20k$.

And you've been here for how long to know SKYMTL well ?

He went right down in a lot of peoples estimations, when he did that very biased review of Nvidias 'big bang' drivers.

Everyone could see the reviewers were told the apps they had to run, as 90% of the website rand the 4 same apps, yet he was on here arguing it was all fair and above board.

That's why a lot of people lost respect for him and the canucks website that day.

[Tim]

Agreed that it's such an aggressive statement, coming from a respected reviewer.

Unless it's back with facts, forgive me if I take the statement with a LOT of salt. Somehow, SKYMTL has turned into an NVIDIA's ambassador in this forum. Weird.

He doesn't do nVidia bashing like half the users here, and uses the round thing that is on his shoulders. That's the difference.

[SKYMTL]

Quite a strong statement considering the fact that the very top offer from Nvidia isn't even 2 times faster than a cheaper 5870 (not to mention 5970 which is on par with (or better than) GTX480 in Haven and a much slower GTX470) in the over-the-top tessellation environment of Haven, not to mention some people speculating about GTX480 scoring much closer to 5870 with high IQ settings...
You are not being totally rational here.

We will see when benchmarks are released with tessellation enabled in AvP, Heaven and some other games. I am not even talking about the performance in relation to NVIDIA cards here since the GF100's performance in DX11 is nothing but rumors and vague benchmarks at this point.

With tessellation enabled we have seen massive decreases in performance in the two apps mentioned above. Actually, PCGamesHardware did a good AvP comparison on a HD 5870. Their results are VERY easily repeatable so I suggest you check it out.

It is just a matter of fact that ATI's current hardware isn't designed around DX11's needs. Rather it is a DX9 / DX10 architecture with DX11 functionality added on. This isn't a bad thing per se since the number of DX11 games out there is minimal at best. I am sure when DX11 becomes mainstream, ATI will have an architecture that properly support all of its advanced rendering paths but until then, I will maintain that the HD 5000 series' DX11 performance isn't that great at all.