First GTX 670 review(s) up (tt & oc.net) * TT OC review added*

[aaksheytalwar]

I agree. But there is no point in a gaming card having good compute, we can have a gaming card and a balanced card. That would be better

 

[Don Karnage]

The 285 was slower then a 5770? Really?

I thought 260 = 5770

 

[aaksheytalwar]

In old games 5770 was slower. But in newer games without dx11, it catches up. I have proved it. Sorry, rest is just your ego.

 

[aaksheytalwar]

The 285 was slower then a 5770? Really?

I thought 260 = 5770

At the time of release that was true, but with newer games 2xx scales badly while 5xxx scales much better

 

[omek]

Missed it... pictures are broken.
Did anyone catch them?

 

[tviceman]

1. It's okay bro, AMD did it to NV with the 58xx. No big deal, just price cuts on a mid-range gpu being sold at enthusiast prices.. not as bad as an enthusiast gpu being forced to sell at mid-range prices, is it??

AMD made a more power and size efficient die than Nvidia did at 40nm and 55nm. They didn't make a faster die, though. So again, AMD's die strategy is now failing them. Nvidia has the smaller, more efficient, and FASTER die.

2. If its any indication, it means gk104's dynamic turbo + custom cooled pcb = no point in a gtx680. As to their "third tier" chip, you mean gk106? That's less than HALF a gk104 on specs, you think its gonna beat up on 78xx? Nah.

No, I meant exactly what I said - a third tier GK104. gtx680 is 1st tier, gtx670 is second tier, and whatever they name the third tier chip, will be priced lower than the hd7950, probably at or cheaper than the hd7870, and will likely have a performance lead over hd7870 equal to what the gtx670 will have over the hd7950. As is evident by the sheer number of different GK107 mobile parts AND Fermi OEM parts, Nvidia is really good at (for better or worse) harvesting as many usable dies as they can and making some kind of product out of it.

edit: See all the doom and gloom for AMD, you guys are smoking something I'm not? 79xx is a small chip, it's not meant to be competing/priced in the top-bracket, the fact that it has been doing so is a bonus for AMD, a price-cut on it is no biggie considering they have been getting away with ripping off customers for many months until they are challenged by NV.

Yeah, it's doing great for them if by great you mean less GPU revenue vs. last quarter and/or last year despite higher selling prices.
Less revenue, same prices = less chips sold.
less revenue, higher prices = even less than less chips sold. <--- This is where AMD was at with their quarterly report. This is, in your opinion, great.

 

[LOL_Wut_Axel]

In old games 5770 was slower. But in newer games without dx11, it catches up. I have proved it. Sorry, rest is just your ego.

At the time of release that was true, but with newer games 2xx scales badly while 5xxx scales much better

That's incorrect. The GTX 260 is slightly slower than the HD 4870, which is the same speed as the HD 5770. The GTX 260 Core 216 is a tiny bit faster than the HD 4870 and therefore HD 5770. GTX 285 vs HD 5770 there's no chance: the GTX 285 will slaughter it.

 

[Silverforce11]

snip

I believe the situation is reversed here, NV has a more efficient die but no compute, thus no Tesla variants. I don't see your doom and gloom, its a balanced chip thats a tiny bit bigger than gk104, with great compute and gaming capabilities. It's not as if 79xx is a giant 500mm2 die here which would restrict its ability to maintain margins at a mid-range pricing.

Revenue is down for both companies if you didn't notice. Maybe you can figure out why? Tip: TSMC.

As to the 2 cluster disabled gk104 variant vs 7870, obviously NV can set the target performance higher, they already know the bar set by the competition since they are late to the party. It's not surprising at all, but then we're talking about a gk104 vs an even smaller and more efficient chip. If NV can price it low, nothing is stopping price cuts on 78xx if AMD needs to, right?

 

[LOL_Wut_Axel]

I believe the situation is reversed here, NV has a more efficient die but no compute, thus no Tesla variants. I don't see your doom and gloom, its a balanced chip thats a tiny bit bigger than gk104, with great compute and gaming capabilities. It's not as if 79xx is a giant 500mm2 die here which would restrict its ability to maintain margins at a mid-range pricing.

Revenue is down for both companies if you didn't notice. Maybe you can figure out why? Tip: TSMC.

Pretty much. The only problem is, the vast majority of enthusiasts and gamers don't care for compute performance. Regardless, from a technical standpoint, the blend AMD did in compute and gaming is nearly flawless.

 

[Mad_dawg]

Pretty much. The only problem is, the vast majority of enthusiasts and gamers don't care for compute performance. Regardless, from a technical standpoint, the blend AMD did in compute and gaming is nearly flawless.

Agreed, people don't buy the gtx 580 for it's compute powers but they bought it because its the fastest chip of it's time... The hd 7970 don't have that luxury vs the the gtx 680....

 

[Silverforce11]

Pretty much. The only problem is, the vast majority of enthusiasts and gamers don't care for compute performance. Regardless, from a technical standpoint, the blend AMD did in compute and gaming is nearly flawless.

If you're going down that approach, I can add this statement: The vast majority of enthusiasts who spend big bucks don't care about power use, so efficiency is pointless. Once efficiency is not a factor, OC 79xx vs OC gtx680 is either a tie or a win for 79xx. Gtx680 is only a win if efficiency & no OC were major factors. I personally think the gtx680 is superior simply because its about 100w less OC vs OC.

 

[Don Karnage]

I don't get the whole "Compute" excuse. What programs utilize compute?

 

[Silverforce11]

I don't get the whole "Compute" excuse. What programs utilize compute?

I dunno but it seems important. Maybe all the explosion in next-gen sequencing machines have to do with it, generating massive datasets that need analyzing??

 

[RussianSensation]

No, it's pretty clear to me that Tahiti is very good when it comes to compute.

Very true but unfortunately for AMD, 99% of gamers don't care. I bet they'd take 48 ROPs vs. 32 in 7970 and trade compute transistor space for them. After all HD7970 has 63% more transistors than HD6970 but only ~45% more performance. Power efficiency has not improved since 6970.

If your preference for NVIDIA can't let you see that, too bad.

Last 2 of my 3 card purchases were AMD : 4890, GTX470, 6950. :thumbsup:

AMD was able to blend gaming and compute performance almost perfectly, making a card that had comparable to higher compute performance than the GTX 580 while having high gaming performance, a reasonable die size, reasonable performance/mm^2, and good performance/watt.

Many would say Pitcairn is a better balanced chip than Tahiti is. Considering the PC gaming market didn't care much for 8800GTX/9800GTX/280/285/480/580's compute performance all this time since 2006, it's not very relevant that HD7970 is good for compute for 99% of gamers. Like I said, most people here would take an HD7970 with compute no better than GTX680 if it had 1200mhz clocks, 48 ROPs, etc.

NVIDIA was only able to get high compute performance at the cost of the three last things that were mentioned. So, you are very much wrong because AMD was able to blend gaming and compute much better than NVIDIA did--and on the first try, which makes it even more impressive.

Next generation games will use higher resolution textures and even more tessellation. See adaptive tessellation. Kepler as an architecture, not as an SKU, is 1 generation ahead of GCN architecture in both of those areas. Therefore, AMD will need GCN 2.0 / Enhanced at minimum just to match Kepler architecture going forward. Their best bet for HD8000 is to chop off GPGPU functions, go with 1200mhz, 48 ROPs, and 2560 SPs. Forget compute altogether since it's hardly relevant for gamers. If they can afford to make a power efficient and fast gaming chip and retain compute, sure. But with HD7900 series they have shown that they cannot. It consumes 100W more power with overclocking against an OCed 680. I personally don't care but given how 680 is selling out, it seems gamers want a fast power efficient chip.

The problem for AMD is that a very few amount of enthusiasts care about compute performance. NVIDIA obviously realized this, and took advantage of it: they made a card with next to no compute/FP64 performance and went all out for a gaming card, which is what AMD did to an extent with Pitcairn.

The product is made for gamers. How does GPGPU compute relate to gaming performance? I don't recall arguments being made in favour of G80, GT200b, GF100/110 for gamers because they had good compute. Again if a poll was made and gamers were asked if they'd take a 48 ROP, 1200mhz 7970 if AMD dropped GPGPU compute to GTX680 levels, I would wager most gamers would take that route.

Your whole argument of "GCN is overhyped" falls to its knees when you consider AMD has cards with both great gaming performance and mediocre compute performance and cards with great gaming performance and great compute performance.

I disagree. Kepler is almost 2x faster in tessellation and texture performance. That means NV can use this architecture safely for at least 2 years. It also puts less pressure on Maxwell knowing that for gaming Kepler is a full generation ahead for next generation gaming performance features. AMD will need to catch up eventually because as we've seen with HD5870 and 6870 get trounced in tessellation in modern games, you cannot afford to be behind for too long in these features. AMD will have to improve texture and tessellation performance with GCN Enhanced or it will have to use brute force to match Kepler's higher-end offerings after GTX680.

 

[LOL_Wut_Axel]

If you're going down that approach, I can add this statement: The vast majority of enthusiasts who spend big bucks don't care about power use, so efficiency is pointless. Once efficiency is not a factor, OC 79xx vs OC gtx680 is either a tie or a win for 79xx. Gtx680 is only a win if efficiency & no OC were major factors. I personally think the gtx680 is superior simply because its about 100w less OC vs OC.

People that don't care about efficiency typically don't care about bang-for-buck and will go straight for the highest performance even if the cards in the tier just are below are 2-5% slower clock-for-clock. That means straight for the GTX 680 and HD 7970. The total cost of a card is 1) what it costs upfront and 2) what it costs to run over a given amount of time, which is why I laugh at people buying GTX 480s for $200-250 now.

I agree that caring for efficiency of a $500+ card is kinda stupid, but these cards will go down in price as the year progresses. If you're gonna get a card for $300-400 and run it for two+ years, which is what most people do, efficiency will matter because the costs start adding up. It's people that have $500 or more to spend that won't give two rats ass about efficiency because they love to waste money and get the highest performance.

 

[Jaydip]

Rage utilized the cuda cores to decompress textures,but considering rage lets just move on.I think Civ5 is also benefited from compute power not sure.

 

[Silverforce11]

In the next round, I think AMD has to split their architectures up, one purely focused on compute and another on gaming. Lumping them together is silly as it wrecks efficiency.

Big K is going to be the compute beast, rumored (and pretty accurate source so far) ~600mm2 but only +25% performance over gk104. That's the sacrifice when going for compute. NV if they can actually produce enough Keplers will have a winning design for HPC and gaming. That's how it should be done. (Think back a few years ago, this was the biggest criticism towards NV's designs, they needed to split up HPC and gaming.. they've done exactly that, they do listen/learn!).

BUT... let's take efficiency out of the equation, suddenly the only fault of 79xx is that its conservatively clocked at stock. That's hardly a fault at all given enthusiasts are avid over-clockers.

Edit: @LOL_Wut_Axel: I agree with your point of view, efficiency matters because electricity in most places around the world is not cheap. But considering we're on an US forum where they enjoy really low kwh rates and they don't seem to care about power use...

 

[Jaydip]

People that don't care about efficiency typically don't care about bang-for-buck and will go straight for the highest performance even if the cards in the tier just are below are 2-5% slower clock-for-clock. That means straight for the GTX 680 and HD 7970. The total cost of a card is 1) what it costs upfront and 2) what it costs to run over a given amount of time, which is why I laugh at people buying GTX 480s for $200-250 now.

I agree that caring for efficiency of a $500+ card is kinda stupid, but these cards will go down in price as the year progresses. If you're gonna get a card for $300-400 and run it for two+ years, which is what most people do, efficiency will matter because the costs start adding up. It's people that have $500 or more to spend that won't give two rats ass about efficiency because they love to waste money.

U said it yourself,Gtx 480 ~ 200$ is still a good buy.

 

[LOL_Wut_Axel]

U said it yourself,Gtx 480 ~ 200$ is still a good buy.

Except no, because of how much it costs to run. That's the whole point: it's inexpensive upfront, and expensive as hell on the long run. People typically go for inexpensive upfront and on the long run, somewhat expensive upfront and inexpensive on the long run, or expensive all-around. People going for inexpensive upfront and very expensive on the long run are doing it wrong: they probably don't understand finances very well given that they were buying the card in the beginning for its low price yet are paying up the bottom for it in the long run.

But that's a topic for another day.

 

[Crap Daddy]

Efficiency was extremely useful to make GTX690.

 

[SirPauly]

People that don't care about efficiency typically don't care about bang-for-buck and will go straight for the highest performance even if the cards in the tier just are below are 2-5% slower clock-for-clock. That means straight for the GTX 680 and HD 7970. The total cost of a card is 1) what it costs upfront and 2) what it costs to run over a given amount of time, which is why I laugh at people buying GTX 480s for $200-250 now.

I agree that caring for efficiency of a $500+ card is kinda stupid, but these cards will go down in price as the year progresses. If you're gonna get a card for $300-400 and run it for two+ years, which is what most people do, efficiency will matter because the costs start adding up. It's people that have $500 or more to spend that won't give two rats ass about efficiency because they love to waste money and get the highest performance.

I agree with some of this but not with a broad brush or sweeping view. I do think power efficiency, acoustics and thermals do help with some owners that may go multi-GPU or may desire some balance in their systems.

 

[Jaydip]

Except no, because of how much it costs to run. That's the whole point.

Exactly the enthusiasts don't give a damn.Can we adopt a rule here ,if we are going to edit a post please add "Editing" at last.

 

[boxleitnerb]

What is "texture performance"?
At higher resolutions, it is a tie. And we have yet so see if the tess performance makes a difference. Up to now, it has only in 2-3 games.

 

[blastingcap]

Until consoles use a feature, don't count on 99% of PC games using them. Tessellation, GPU-accelerated physics, etc.

 

[LOL_Wut_Axel]

In the next round, I think AMD has to split their architectures up, one purely focused on compute and another on gaming. Lumping them together is silly as it wrecks efficiency.

Big K is going to be the compute beast, rumored (and pretty accurate source so far) ~600mm2 but only +25% performance over gk104. That's the sacrifice when going for compute. NV if they can actually produce enough Keplers will have a winning design for HPC and gaming. That's how it should be done. (Think back a few years ago, this was the biggest criticism towards NV's designs, they needed to split up HPC and gaming.. they've done exactly that, they do listen/learn!).

BUT... let's take efficiency out of the equation, suddenly the only fault of 79xx is that its conservatively clocked at stock. That's hardly a fault at all given enthusiasts are avid over-clockers.

Edit: @LOL_Wut_Axel: I agree with your point of view, efficiency matters because electricity in most places around the world is not cheap. But considering we're on an US forum where they enjoy really low kwh rates and they don't seem to care about power use...

Efficiency and power usage for Tahiti are still good. The HD 7970 is more efficient than the last very efficient Performance card AMD had, the HD 6850. It's worse than GK104 in this metric, sure, but at least AMD didn't have the trainwreck NVIDIA did with GF100 and 110 when they needed to give up performance/watt and absolute power usage entirely to get very high compute performance. In terms of gaming performance, if you consider overclocking, Tahiti is definitely up there with GK104. The problem is that high overclocks on Tahiti, specifically the HD 7970=very high power consumption. The only way you're gonna get the same efficiency as GK104 with Tahiti is by overclocking slightly and undervolting slightly, but if you do that in terms of performance it'll be somewhat slower than GK104.

Low power usage matters because it's not just the card itself that will be contributing to it. If you run a very hot card, you will need much better cooling for the room and that costs additional. Even if you're in the US and have very low electricity rate, it'll add up not just because of the card but because of the increased air conditioning you'll need to run the card and PC at reasonable temperatures.