Kepler Before Xmas

[SirPauly]

It's always much more ideal to have many players competing and trying to innovate because it may improve the technology landscape and gaming experience, too.

Still believe AMD's GPU division is formidable competition.

 

[tviceman]

You Americans place no emphasis on perf/watt at all.

Come back and consider when your electricity is charged at 30-40 US cents per kilowatt hour, instead of the current 8 cents.

http://www.anandtech.com/show/4137/...-6950-1gb-xfxs-radeon-hd-6870-black-edition/6

A gtx560ti uses 22 more watts per hour on average than a hd6950 while gaming. So, at 50 cents per kilowatt hour, a gtx560ti would cost (50 cents / 1 hour cost of electricity) * ( 22 watts / 1000 watt-hours which is 1 kilowatt-hour) = 1.1 cents more per hour to use while gaming.
A gtx570 uses 21 more watts per hour on average than an hd6970 while gaming. So that would cost 1.05 cents per hour more than an hd6970.
A gtx580 uses 49 more watts per hour on average than an hd6970. So that would cost 2.45 cents more power than an hd6970.

I'm sorry was there something meaningful in what you were trying to say?

 

[Silverforce11]

Yeah, over the lifetime of the card.

Ppl fuss over $20 difference in prices... some even less. Like rebates and specials, $10 discount!!

Point was if you're those type of people, then factor in power use because the difference in the long run is probably more.

 

[SirPauly]

Performance per watt is very important but the higher one goes with a price-point; does it really matter over performance, quality and gaming experience potential?

One can make the case that Fermi could of done a better job in this area but let's see what they do with Kepler.

 

[Silverforce11]

Another factor for HPC, perf/watt is probably the most important criteria after software support. Server farms are expensive to run and cool.

But NV can do whatever because they have no competition since STREAM development is crap compared to CUDA.

 

[nenforcer]

Another factor for HPC, perf/watt is probably the most important criteria after software support. Server farms are expensive to run and cool.

But NV can do whatever because they have no competition since STREAM development is crap compared to CUDA.

Expect this to change with the Radeon HD 7000 series and beyond. AMD is taking GPGPU seriously and the architecture will introduce many new features in software because I believe the hardware itself is just a 28nm die shrink of Radeon HD 6000 series.

 

[SirPauly]

Another factor for HPC, perf/watt is probably the most important criteria after software support. Server farms are expensive to run and cool.

But NV can do whatever because they have no competition since STREAM development is crap compared to CUDA.

nVidia placed themselves into this position by being pro-active with GPU processing, investments into hardware and software.

 

[ocre]

Expect this to change with the Radeon HD 7000 series and beyond. AMD is taking GPGPU seriously and the architecture will introduce many new features in software because I believe the hardware itself is just a 28nm die shrink of Radeon HD 6000 series.

Honestly, i will believe it when i see it. You still see ppl claiming CUDA is a loss but look what its done for Nvidia. I am still waiting on all the AmD promises to become, DX11 is soon to be replaced by DX12 and still CUDA remains.

ps,
Where is AMDs physx by the way. Not saying that they need it or anything, its just more hog wash, empty promises.

 

[Genx87]

Honestly, i will believe it when i see it. You still see ppl claiming CUDA is a loss but look what its done for Nvidia. I am still waiting on all the AmD promises to become, DX11 is soon to be replaced by DX12 and still CUDA remains.

ps,
Where is AMDs physx by the way. Not saying that they need it or anything, its just more hog wash, empty promises.

Bullet Physics like many things with AMD has languished. Havok at this point has a higher chance of running on a GPU than Bullet does of becoming a meaningful Physics API.

 

[RussianSensation]

I am often extremely impressed with your insightful post. They are from very well thought, complete understandings which manifest to extremely clear points of view.

Your post on this subject are a great example. There is a lot if one could gather from your post. I am glad you still take the time. Its sad that most ppl wont pay much mind, or grasp the real importance and ramifications intertwined within. It is a deterring factor on my standpoint but i do still sometimes find myself rambling. So I commend you for taking your time, as you do. Hopefully there are enough ppl that will value it as such.

Thanks for the compliment ocre!! :$

I think the next generation of 28nm GPUs will satisfy the cravings of those who want even better energy efficiency (esp. in the mid-range) and those of us who want even more performance on the high-end.

Huang provided more hints after the earnings conference call:

"We have more notebook design wins for the Ivy Bridge cyle than we ever had in notebooks. And this is likely the most successful notebook cycle we've ever experienced. And so we've got a lot of engineers dedicated to getting those notebooks into production. The reason for our success, I believe, is because our kepler generation of GPU was designed for intense energy efficiency. And with energy efficiency, we were able to translate that to simultaneously higher performance, as well as longer battery life. "

and

"At the GPU Technology Conference last year, Huang revealed that Kepler is designed to crank out almost three times as many double-precision FLOPS than the current Fermi architecture per watt."
Source

While double-precision performance doesn't directly translate into more FPS, I think we can take away from this that the Kepler architecture will be far more efficient than Fermi in general.

you reek if bias

and if you can't understand the meaning of relative then I can't help you

if it had not been for nVidia's relative blunder and AMD's relative success, I wouldn't have been surprised if we only had nVidia as a discrete GPU option today, although I'm sure you'd be thrilled if that were the case

I think you are being a little unfair to bryan1995. His point was more about AMD losing potential revenue by selling their cards for prices that many of us deem to be too low. For example, AMD could have priced HD4870 much closer to $399 since it was easily trading blows with the $399 GTX260. But even more so, when AMD was far in the lead, they released a $259-269 HD5850 and a $369 HD5870. Those cards would have sold extremely well at much much higher prices since they were faster than anything NV had at the time. Even today, it's shocking to believe how much performance a gamer was getting by buying an HD5870 for $370 on launch date. I am sure a lot of individuals who pay $430-500 for the GTX580 would have still paid $430-500 for the HD5870 if AMD chose to price it that way. That could have lowered price after Fermi arrived in late March/early April. Instead, AMD tried to raise prices after launch by $20-30. Strange strategy to me to say the least.

 

[bunnyfubbles]

I think you are being a little unfair to bryan1995.

I don't think so, just calling it as I see it.

His point was more about AMD losing potential revenue by selling their cards for prices that many of us deem to be too low. For example, AMD could have priced HD4870 much closer to $399 since it was easily trading blows with the $399 GTX260. But even more so, when AMD was far in the lead, they released a $259-269 HD5850 and a $369 HD5870. Those cards would have sold extremely well at much much higher prices since they were faster than anything NV had at the time. Even today, it's shocking to believe how much performance a gamer was getting by buying an HD5870 for $370 on launch date. I am sure a lot of individuals who pay $430-500 for the GTX580 would have still paid $430-500 for the HD5870 if AMD chose to price it that way. That could have lowered price after Fermi arrived in late March/early April. Instead, AMD tried to raise prices after launch by $20-30. Strange strategy to me to say the least.

if AMD hadn't priced the 4800s low, they wouldn't have put nVidia under as much strain as they did. Don't forget that on the other side of the 4800s there was the G92 still looming large in the lower-mid range budget, the later revisions of which could give the 4850 a go.

also, AMD mispricing the 5800s also reinforces my point about Fermi really being a very significant blunder by nVidia, so much so that it even hurt AMD as much as it did nVidia, if AMD had known they'd be scott-free with no competition for their 5800s there's no way they would have priced them so competitively right out of the gate, the pricing was specifically set forth to put pressure on nVidia yet again just as they had done with the 4800s, their mistake in assuming that Fermi wasn't as far off as it ended up being.

Also do not forget that the masses were also initially disappointed by the 5800 debut pricing as they had come to expect the newly established 4800 tiered pricing. Pricing the 5800s too much higher might have caused a sales slump and a dud launch, on the flip side they sold like hot cakes which only improved the reputation of the series, even when they were reflexively increasing the prices.

 

[Silverforce11]

AMD can't afford to wage a price war vs NV, they are not making a big profit while NV is and is sitting on big cash reserves.

If 7970 is near 6990 performance, they should price it >$500 and not their previous gen prices at around ~$370. They also need to make the 7950 not so close to the 7970, that it will cannibalize sales.

 

[3DVagabond]

AMD can't afford to wage a price war vs NV, they are not making a big profit while NV is and is sitting on big cash reserves.

If 7970 is near 6990 performance, they should price it >$500 and not their previous gen prices at around ~$370. They also need to make the 7950 not so close to the 7970, that it will cannibalize sales.

I understand what you are saying and the ##50's definitely take sales from the ##70's. Problem is they also have to watch where the competition's chips are performance wise. For example, leave a bigger gap between the 6970 and 6950 and the 560ti will beat it.

 

[Silverforce11]

If the competition intends to slot in their 2nd tier high end closer to its 1st tier and cannibalize its own sales, thats fine.

Doing what they did with the 5850/5870 and especially 6950 unlock fiasco really hurts their high premium top dog.

 

[3DVagabond]

If the competition intends to slot in their 2nd tier high end closer to its 1st tier and cannibalize its own sales, thats fine.

The competition wouldn't have to change a thing. If the 6950 was slower than what it is, it would be beat by the 560ti as they stand right now.

Doing what they did with the 5850/5870 and especially 6950 unlock fiasco really hurts their high premium top dog.

I don't think the 6950's unlocking is a fiasco at all. It's obviously is a calculated marketing move. That's why the board partners are releasing them with the unlocked bios already loaded. If you were buying a 6950 right now, which one would you buy? I'll bet one of the unlocked Sapphires.

They need to use the dual bios on the 6970 to create a better market position for them. For example the 6970's with the performance bios in the 2nd position.

http://www.hardwareheaven.com/revie...hd6970-graphics-card-review-introduction.html

This allows you to market a card that's higher performance but still has the same power, heat, and noise specs of a stock card. That's until you flip the switch, but then it's your own fault the card's loud, hot, and inefficient (relatively speaking). They get the improved benches though. Smart move. :thumbsup:

 

[Silverforce11]

I think its great for the consumer but if you look at the GPU charts, 6950 outsold 6970 by a huge margin (well, based on steam anyway). The same didn't really happen with 4850 vs 4870 and not as bad 5850/70.

Essentially the 7970 needs a big gap in performance vs the 7950 if its to be an attractive buy at a premium.

 

[3DVagabond]

I think its great for the consumer but if you look at the GPU charts, 6950 outsold 6970 by a huge margin (well, based on steam anyway). The same didn't really happen with 4850 vs 4870 and not as bad 5850/70.

Essentially the 7970 needs a big gap in performance vs the 7950 if its to be an attractive buy at a premium.

I'll bet the 570 outsells the 580, and the 6850 outsells the 6870. That's the nature of the beast.

They need to use the dual bios on the 6970 to create a better market position for them. For example the 6970's with the performance bios in the 2nd position.

Note this part of my previous post. The dual bios has been a marketing win for the 6950. They need to do something like that for the 6970. IMO, they need to further bin Cayman chips for the 6970 like they do for the 6950. The best 6950 chips are used for the toxic. These are chips that are fully functioning 6970's but maybe don't meet the specs exactly. They'll still do 1540shaders/880MHz, but might not have enough headroom above spec to be used in 6970's. Then there's the chips that'll run 1540, but might not do 880MHz and be fully stable. Etc...

Bin some Cayman chips that'll do 1GHz and RAM that can run the full 6GHz. Load that setting into the second bois. Put premium components and coolers on it and sell it for a premium. Say +$350.00. Have two more tiers below that, maybe. With the bottom tier being full spec'd 6970's with cheaper components, boards, etc... Sell those for $300.00.

Just an example off the top of my head. I'm sure it would require more thought than the 10min or so that I just gave it. This is just to explain the concept.