Charlie at Semiaccurate says: Physics hardware makes Kepler/GK104 fast

[AnandThenMan]

So things could end up being: Nvidia hardware needs Nvidia games, otherwise performance is only okay. I see what is happening here, Nvidia doesn't have any of the upcoming console contracts, so they are going to try and force an Nvidia only gaming infrastructure on the PC. D:

 

[Olikan]

If NVidia gives the developers a lot of green pieces of paper with dead presidents on them, they'll bend backwards and use the middleware. And it's a bit more than half of the people, Steam survey says that NVidia is around 60% while ATI around 40%.

it's pointless the market share...all nvidia cards prior kepler won't have benefits of the hardware physX....

kepler will have to sell TONS of cards, to have it's own market share, enought to make developers pay atention to it ( ie. without nvidia paying)

 

[AnandThenMan]

The whole TWIMTBP deal was bad for consumers. Games should be coded to use features that are open and NOT proprietary.

Agreed, but that is completely the opposite of the direction Nvidia seems to be heading.

kepler will have to sell TONS of cards, to have it's own market share, enought to make developers pay atention to it ( ie. without nvidia paying)

Which is why Nvidia will pay for Nvidia only coded games.

 

[Crap Daddy]

it's pointless the market share...all nvidia cards prior kepler won't have benefits of the hardware physX....

kepler will have to sell TONS of cards, to have it's own market share, enought to make developers pay atention to it ( ie. without nvidia paying)

It's not pointless. You do realize that if what Charlie sayz is true we will have a 300$ card that will perform like a GTX580 in all games and smoke every card on the market in certain titles while being a small power efficient chip. Is this bad?

 

[lavaheadache]

A 7970 nvidia equivalent for $299 and you guys still complain? Would you rather pay $550 or what? This is good news. Less power draw than a 7970, smokes a 580, beats a 7970 in some games. Sounds like a 8800gt gtx 460 type card that changes the game with an awesome level of value/performance.

I wouldn't call it a 7970 equivalent by the way it was described.

It looks like what was trying to be said was that in games like HAWX 2 the card will scream, but in other games like Crysis the role will reverse.

I pulled the those titles from my pooper but my point is that there were games that nVidia had a silly lead in for tainted reasons while there were plenty of titles that seemed to be very neutral regarding whose flag you were swinging.

 

[AnandThenMan]

It's not pointless. You do realize that if what Charlie sayz is true we will have a 300$ card that will perform like a GTX580 in all games and smoke every card on the market in certain titles while being a small power efficient chip. Is this bad?

You need to actually read the article.

 

[tviceman]

All this information, while explained in a somewhat clear manner, is in itself confusing. Software physx will be hardware accelerated? Hmmmm, I doubt it. If this is a forward looking architecture, that is leveraged with new games that have heavy use of physics, like battlefield 3 or hardware-accelerated physx games, then great that is fine but if it's limited entirely to the physx API, then I question Nvidia's decisions to completely close off their hardware potential outside of their ecosystem.

It has to be more general than just the physx API for this to be a game changer. It sounds like there are going to be the exact same arguments over which games to benchmark with Kepler and GCN as there was with Cypress and Fermi.

But for $300, if it's at worst a gtx580 equivalent and at best 10% faster than an hd7970, then it is a hell of a lot more value over an hd7970 (so long as the peaks are occur just as often in newer games as do the valleys).

 

[Stuka87]

It's not pointless. You do realize that if what Charlie sayz is true we will have a 300$ card that will perform like a GTX580 in all games and smoke every card on the market in certain titles while being a small power efficient chip. Is this bad?

Nowhere did he say it would perform like a 580 in non-optimized games (unless I missed it). From what I gathered, it would perform like an AMD 78x0 in non-optimized games, and beat a 7970 in "highly" optimized games. And I am sure that level of optimization will most likely not be the case all the time. As it would most likely really depend on the game's content. Games with less physics will see less of a benefit.

 

[Larnz]

Yeah I dislike the idea of causing fragmentation in games by who pays the most into their dev kitty.

So if this is true any game that flashes an Nvidia logo will run like demon yet any AMD evolved sponsored game you end up playing with effectivly a mid range card...

1080p gamers might be ok with that as even with the slight hit to fps in non phys games they will probably still be ok. Anyone who wants a single GPU to game at 1600p howeevr would probably steer clear as it might make AMD sponsored games very laggy for a "high" end card.

IMO I call shenanigans.

 

[OILFIELDTRASH]

I wouldn't call it a 7970 equivalent by the way it was described.

It looks like what was trying to be said was that in games like HAWX 2 the card will scream, but in other games like Crysis the role will reverse.

I pulled the those titles from my pooper but my point is that there were games that nVidia had a silly lead in for tainted reasons while there were plenty of titles that seemed to be very neutral regarding whose flag you were swinging.

I could be a little over zealous. But it sounds like its around a 580 some games and in twimtbp games better than a 7970.

Really for $300 that is a heck of a deal IMO.

 

[Imouto]

This sounds too good to be true and only this guy is the one coming up with all of it, sounds like a bucketload of ****.

This is starting to get closer and closer to ala Bulldozer fiasco...

word

 

[OILFIELDTRASH]

I'm wondering if BF3 is a "heavily" optimized game? Is bf3 a twimtbp game?

If so that would be great. If this card overclocks like a 7970 then I'm really excited about it.

not really sure about Charlie though. Is his info reliable?

 

[blastingcap]

I could be a little over zealous. But it sounds like its around a 580 some games and in twimtbp games better than a 7970.

Really for $300 that is a heck of a deal IMO.

EDIT to clarify:

The way I (re)read it, he's saying that

- for certain games, GK104 will be 10-20% slower than 7970 and destroy a GTX 580. For a few games the GK104 might actually be faster than a 7970.

- for all other games, the GK104 will be somewhere around a 7870.

Charlie also mentions that GK104 will be a relatively small chip with power draws much lower than the 225W rumors that have been going around, meaning it will probably also draw less power and make less noise/heat than 79x0 chips.

If what Charlie says is true, and if GK104 is priced at $300, then 7870 should be priced accordingly. Like $250 or less.

Price war, anyone?

 

[OILFIELDTRASH]

This sounds too good to be true and only this guy is the one coming up with all of it, sounds like a bucketload of ****.

This is starting to get closer and closer to ala Bulldozer fiasco...

lol. Bulldozer will forever be linked with fail.

Now when something blows it's Bulldozer. lol

 

[Stuka87]

He said Tahiti which could refer to 7950; it's ambiguous.

The way I read it, he's saying that

- for certain games, GK104 will be significantly faster than a 7970 and destroy a GTX 580.

- for all other games, the GK104 will be about 10-20% slower than either a 7950 or 7970 (this is ambiguous)

He said performance in non-optimized games would be similar to Pitcairn, which is the 78x0 series chips. Not Tahiti.

 

[tviceman]

He said Tahiti which could refer to 7950; it's ambiguous.

The way I read it, he's saying that

- for certain games, GK104 will be significantly faster than a 7970 and destroy a GTX 580.

- for all other games, the GK104 will be about 10-20% slower than either a 7950 or 7970 (this is ambiguous)

I think if GK104 sells for $300 and is 10-20% slower than a 7950 in most games and much faster in others, then GK104 will sell out at launch. And if it's the 7970 that Charlie was comparing to, then the price/perf difference would be just that much greater. Charlie also mentions that GK104 will be a relatively small chip with power draws much lower than the 225W rumors that have been going around, meaning it will probably also draw less power and make less noise/heat than 79x0 chips.

Price war, anyone?

20% slower than an hd7950 would put it right in gtx570 territory. $300 for gtx570 speed with occasional big performance jumps (if it is only occasional) is, again, nothing at all to be excited about. So I hope he's talking about the hd7970, and since he's back to being his good old self, I hope he's also talking worst case scenarios here. At least he did say the chip is "small" and power draw looks very good.

I just don't think Nvidia would build a chip that only accelerates certain games and situations that can only exist when made within their ecosystem. And hardware accelerating code not meant to be hardware accelerated doesn't make sense either. As much as Nvidia wants physx to take over the world, it just doesn't make sense from a business perspective.... then again, I always thought it didn't make sense to lock physx out of AMD/Nvidia hybrid setups. *sigh* I guess we'll see.

 

[blastingcap]

He said performance in non-optimized games would be similar to Pitcairn, which is the 78x0 series chips. Not Tahiti.

Upon closer inspection I think you are right, but what threw me off was when he talked about how some titles may run on GK104 faster than even Tahiti, not 10-20% slower.

So I guess he's actually saying something more like:

- 10-20% slower than 7970 in some Nvidia-optimized games, with a few games where the GK104 is even faster
- otherwise, somewhere around 7870 levels in other games
- much lower power draw than 225W
- probably less heat/noise since it's a small chip with lower power draw
- $300 price point

Given that the 7950/70 are both priced so much more than $300, that's not necessarily a bad combination... like, hey buy this $300 card from us that is sometimes almost as fast as a $550 card from them. I don't know what 78x0 is supposed to be priced at, but it better be significantly less than $300, if what Charlie is saying turns out to be true.

 

[taltamir]

tahiti is MIMD, not SIMD

The whole GPU is... both me and charlia said the shaders (although we said different things about the shaders).
AMD used to have 5way and later 4way shaders while nvidia used 1way shader. Now AMD also uses 1way shader. aka a SIMD shader, an array of such SIMD shaders make up a MIMD GPU

 

[AnandThenMan]

I just don't think Nvidia would build a chip that only accelerates certain games and situations that can only exist when made within their ecosystem.

I believe Nvidia wants to (and probably did) build a chip that is designed to greatly accelerate games they sponsor. So they may indeed be happy to leave the rest of the games to show adequate performance, and use their influence to push reviewers to bench the NV happy titles. Nvidia has done this before, not such a stretch that they would expand on this if Kepler is more specifically designed to accelerate PhysX enabled effects.

The above fits right into Nvidia's MO.

 

[Crap Daddy]

He said performance in non-optimized games would be similar to Pitcairn, which is the 78x0 series chips. Not Tahiti.

He says here:

Benchmarks for GK104 shown to SemiAccurate have the card running about 10-20% slower than Tahiti.

That's the GTX580 for you. We don't know if he's talking about XT or Pro. But we also don't know how Pitcairn will perform, do we?

 

[MrTeal]

The problem for Nvidia is that once you venture outside of that narrow list of tailored programs, performance is likely to fall off a cliff, with peaky performance the likes of which haven’t been seen in a long time. On some games, GK104 will handily trounce a 7970, on others, it will probably lose to a Pitcairn. Does this mean it won’t actually do what is promised? No, it will. Is this a problem? Depends on how far review sites dare to step outside of the ‘recommenced’ list of games to benchmark in the reviewers guide.

If SA is correct, on the games where GK104 fails, it would be below a 7870. That likely would mean it's also in the same range or below a 6970. That would not be so impressive if it's true. Hopefully even if it is true it's only a couple corner cases, and SA is just making the problems sound worse than they are.

 

[tincart]

You do realize that if what Charlie sayz is true we will have a 300$ card that will perform like a GTX580 in all games and smoke every card on the market in certain titles while being a small power efficient chip. Is this bad?

He didn't say that at all.

If performance is going to be dependent on TWIMTBP branding, I'm not buying. I'll have to wait and see if this is true, but it looks like AMD may be getting my business this round.

 

[Concillian]

The way I read it, he's saying that

- for certain games, GK104 will be significantly faster than a 7970 and destroy a GTX 580.

- for all other games, the GK104 will be about 10-20% slower than either a 7950 or 7970 (this is ambiguous)

I don't think it's really all that ambiguous, he comes right out and says it:

On some games, GK104 will handily trounce a 7970, on others, it will probably lose to a Pitcairn.

Losing to Pitcairn means it's going to be a lot closer to 20% slower than a 7950 than it will be to 10%.

Still, if it's priced like a Pitcairn, and performs at least close to a Pitcairn in the "poor" performance games, then there isn't a big problem, is there?
The problems will arise if it's priced higher than Pitcairn or if it's priced similarly and the performance in the un-optimized games is significantly slower than Pitcairn.

 

[blastingcap]

20% slower than an hd7950 would put it right in gtx570 territory. $300 for gtx570 speed with occasional big performance jumps (if it is only occasional) is, again, nothing at all to be excited about. So I hope he's talking about the hd7970, and since he's back to being his good old self, I hope he's also talking worst case scenarios here. At least he did say the chip is "small" and power draw looks very good.

I just don't think Nvidia would build a chip that only accelerates certain games and situations that can only exist when made within their ecosystem. And hardware accelerating code not meant to be hardware accelerated doesn't make sense either. As much as Nvidia wants physx to take over the world, it just doesn't make sense from a business perspective.... then again, I always thought it didn't make sense to lock physx out of AMD/Nvidia hybrid setups. *sigh* I guess we'll see.

Sounds more like a $300 card that is a bit faster and a lot cooler and a lot less power hungry than a GTX 570 in most games, with some big jumps up to 79x0-caliber in Nvidia-optimized games.

I think he was saying that Nvidia was going to leverage their expertise with the PhysX API, in GK104. PhysX is in a lot of games (second to Havok?), though my recollection is fuzzy and out of date.

I also believe that the big brother of GK104 will be different. That Nvidia has bifurcated their line and the big brother of GK104 will be compute oriented with all the bells and whistles.

 

[AnandThenMan]

If SA is correct, on the games where GK104 fails, it would be below a 7870. That likely would mean it's also in the same range or below a 6970. That would not be so impressive if it's true. Hopefully even if it is true it's only a couple corner cases, and SA is just making the problems sound worse than they are.

Then it boils down to where GK104 actually performs. Remember, PhysX could most certainly run on AMD hardware, the limitation is software not hardware. Nvidia can't perform some kind of silicon magic here, they will have to use silicon resources to accelerate the eye candy. So what they will probably do is on a non-Nvidia card, shunt the workload to the CPU, or just disable it, like they do now.