nVidia wins this round - Charlie D.

[Mezzanine]

Awesome, I might go buy one...oh wait..

 

[GaiaHunter]

Good if the prices go down - I wouldn't mind buying a card with 7950-7970 performance levels for €250.

But somehow I doubt it I'll see that performance level for that price in the near future... :'(

 

[Olikan]

4. Site traffic needed a boost today
5. Site got hacked
6. He was fed misinformation

Btw people are saying it is GDDR5 and realistically that is all it can possibly be for a gaming card. Note that NVDA had major memory controller problems for Fermi. If they cleared that problem up, and then some, that may account for the discrepancy.

7. it's a trap! (like the cpu-after-haswell story)

well, actually, this is not really THAT surprising. Tahiti actually have lot's of it's bandwidth left without use.

here, tahiti can get 14% more performance without overclocking the memory.

 

[Skurge]

7. it's a trap! (like the cpu-after-haswell story)

well, actually, this is not really THAT surprising. Tahiti actually have lot's of it's bandwidth left without use.

here, tahiti can get 14% more performance without overclocking the memory.

Actually, there are situations where tahiti is bandwidth starved.
http://www.xbitlabs.com/articles/graphics/display/radeon-hd-7970_9.html#sect0

 

[LOL_Wut_Axel]

LOL, no. Maybe they'll have the fastest single-GPU card, but that in no way, shape, or form means they "won this round". NVIDIA needs to learn to make more efficient consumer GPUs. They also need to release cards sooner so AMD doesn't get such a big head start. Also, GCN looks like a winning architecture, and AMD can very easily bring a card to market that's 30% faster than the HD 7970. Plus AMD has been better when it comes to pricing for around three years (with the exception of the GTX 460 and a few others) Very much doubt NVIDIA wins this round, just like they didn't win against the HD 5000-6000 series overall.

 

[SlowSpyder]

Who knows...

Maybe AMD got wind of Keplar and that is why the 7970 was moved up? I guess we'll see in a few months.

Either way, I am happy with my 7970... for now.

 

[GaiaHunter]

LOL, no. Maybe they'll have the fastest single-GPU card, but that in no way, shape, or form means they "won this round". NVIDIA needs to learn to make more efficient consumer GPUs. They also need to release cards sooner so AMD doesn't get such a big head start. Also, GCN looks like a winning architecture, and AMD can very easily bring a card to market that's 30% faster than the HD 7970. Plus AMD has been better when it comes to pricing for around three years (with the exception of the GTX 460 and a few others) Very much doubt NVIDIA wins this round, just like they didn't win against the HD 5000-6000 series overall.

We don't know what was the price toll for the increased GPGPU capabilities on GCN.

It was easy to see that GF100 was paying a big price toll due to the 5xxx series being out and the GTX480 not being more power efficient than the GTX285 and GTX295.

It might be possible that without the increased GPGPU AMD could have obtained much better performance, performance/watt and smaller die size.

And NVIDIA is exploring that due to the fact they already faced harsh trade-offs.

Or Charlie can be joking/wrong.

Who knows?

 

[boxleitnerb]

LOL, no. Maybe they'll have the fastest single-GPU card, but that in no way, shape, or form means they "won this round". NVIDIA needs to learn to make more efficient consumer GPUs. They also need to release cards sooner so AMD doesn't get such a big head start. Also, GCN looks like a winning architecture, and AMD can very easily bring a card to market that's 30% faster than the HD 7970. Plus AMD has been better when it comes to pricing for around three years (with the exception of the GTX 460 and a few others) Very much doubt NVIDIA wins this round, just like they didn't win against the HD 5000-6000 series overall.

Nvidia made more or equally efficient GPUs before Fermi. So what? It changes back and forth. Also I wouldn't call 2-3 months a big head start, especially at those ridicilous prices.

 

[Olikan]

Although Charlie has been spot on on nVidia stuff lately.

actually no, he isn't.

"the CPU is under-performing badly, missed internal clock targets, and costs too much to be financially viable" .....about tegra 3

"The big problem for Nvidia is that the first batch of 28nm parts isn’t a new design, it is a shrink of the GF11x parts currently on sale"

 

[BallaTheFeared]

lol sorry, couldn't resist.


Did I miss something, cause 130GB/s to 172GB/s of memory bandwidth nets me all of about 7% more performance.

 

[Silverforce11]

All NV has to do is a direct die shrink of the 580, up clocks and they get a GPU with a relatively small die size that can match the 7970.

AMD didn't set the target bar very high when they went with a 365mm2 die and ~210W TDP. These kind of design targets fall right in line with NV's mid-range part. Unless they screw up majorly, we can expect gtx580 +10-20% from gk104. Great for consumers as there will be major pricedrops. AMD is milking it now because they can.

 

[IonusX]

What about the rest of the card? Memory bandwidth alone doesn't make a card.

if they rush job it kepler wont be doing that anytime soon. i also dont trust anything that ever comes out of s/a's mouth. even charlie's they all a band of lairs n cheats!

 

[96Firebird]

See that arrow on the "A" in the logo? That's where this one landed...

 

[blackened23]

Did I miss something, cause 130GB/s to 172GB/s of memory bandwidth nets me all of about 7% more performance.

I'll be exited for Kepler as well Balla, but what gives with the leaks? There's one saying that it will be equal to 7970 (OBN). And then Charlie trolls and says it will be a monster gpu, but have a 256 bit memory bus? What?

You have to admit there is so much conflicting information here, I just do not know what to believe. I actually think NV is probably behind some of this, informational wars is what they do best (just look at what they did with Fermi a year before GTX 480's release)

 

[LOL_Wut_Axel]

We don't know what was the price toll for the increased GPGPU capabilities on GCN.

It was easy to see that GF100 was paying a big price toll due to the 5xxx series being out and the GTX480 not being more power efficient than the GTX285 and GTX295.

It might be possible that without the increased GPGPU AMD could have obtained much better performance, performance/watt and smaller die size.

And NVIDIA is exploring that due to the fact they already faced harsh trade-offs.

Or Charlie can be joking/wrong.


Who knows?

Well, NVIDIA already has huge die sizes coupled with an all-purpose GPGPU design. AMD also made a GPGPU design, but not at the cost of power efficiency or die sizes. Tahiti has a somewhat smaller die size than Cayman and a tiny bit higher power consumption. NVIDIA won't change its huge die size strategy, so AMD looks set to win in power efficiency yet again. Yields should also prove better for AMD, again. And in terms of performance, since the HD 7970 probably won't be enough to counter the fastest single GPU NVIDIA has, they can easily increase performance by 30% via higher clocks, and more texture units and SPs.

I'm really not a fan of NVIDIA's huge die/power consumption strategy. The card I have now is by no means inefficient, but one of the reasons GF104/114 was made was to address power efficiency somewhat. The HD 6870 is still comparable to my card in performance and consumes less power. And why do I care so much about power consumptions? Because PR is one of the worst 1st world countries when it comes to energy costs, not to mention I'm a sucker for efficiency in general.

 

[BallaTheFeared]

I'll be exited for Kepler as well Balla, but what gives with the leaks? There's one saying that it will be equal to 7970 (OBN). And then Charlie trolls and says it will be a monster gpu, but have a 256 bit memory bus? What?

You have to admit there is so much conflicting information here, I just do not know what to believe. I actually think NV is probably behind some of this, informational wars is what they do best (just look at what they did with Fermi a year before GTX 480's release)

Who really knows... Right now I have no plans to upgrade though even if Kepler is a smash hit.

In my personal experience though, even with stupid high core/shader clocks on my 470s memory bandwidth doesn't mean jack.

 

[LOL_Wut_Axel]

Nvidia made more or equally efficient GPUs before Fermi. So what? It changes back and forth. Also I wouldn't call 2-3 months a big head start, especially at those ridicilous prices.

Except the GTX 680/780/785 probably won't be released until July or August, of course. Maybe June if NVIDIA is really lucky...

So no, it's nowhere near 2-3 months. Like with the HD 5000 series, AMD will have a headstart of about 6 months. That's the price NVIDIA has to pay for making such huge monolithic GPUs.

 

[MrK6]

Actually, there are situations where tahiti is bandwidth starved.
http://www.xbitlabs.com/articles/graphics/display/radeon-hd-7970_9.html#sect0

Yep. Even in my own testing I've found Tahiti to be extremely bandwidth limited. Clocking the memory can give the same or greater performance than clocking the core, which is very rare.

I'm curious about the whole article though. What did Charlie "see" 3-4 months out from release? Why did he switch his stance all of a sudden? It'd be nice to have some hard data on Kepler, but with it being this far out, who knows. With an extra 4+ months under its belt, I would hope it would be competitive or better than an AMD part.

 

[blastingcap]

http://semiaccurate.com/2012/01/19/nvidia-kepler-vs-amd-gcn-has-a-clear-winner/ vs http://semiaccurate.com/2011/12/01/can-nvidia-supply-apple/ So which is it, 256 or 384 bit? Because one article says one, the other says the other, about GK104.

 

[bunnyfubbles]

LOL, no. Maybe they'll have the fastest single-GPU card, but that in no way, shape, or form means they "won this round". NVIDIA needs to learn to make more efficient consumer GPUs. They also need to release cards sooner so AMD doesn't get such a big head start. Also, GCN looks like a winning architecture, and AMD can very easily bring a card to market that's 30% faster than the HD 7970. Plus AMD has been better when it comes to pricing for around three years (with the exception of the GTX 460 and a few others) Very much doubt NVIDIA wins this round, just like they didn't win against the HD 5000-6000 series overall.

AMD has been better about pricing and efficiency for consumer concerns, however GCN is currently a step in the opposite direction in those areas. The 7970 is not only the most expensive single GPU card AMD has released since acquiring ATI, its currently drawing more power under load than the 6970 falling just short of the GTX580. Clearly AMD had to sacrifice gaming-oriented performance and power efficiency with this architecture.

Since nVidia already made that step towards HPC several generations ago, it really wouldn't be too surprising if AMD ends up disadvantaged overall this round. Luckily for AMD (and all of us really) they've got a head start in this race, as I think they'll need it.

AMD caught a pretty big break when nVidia screwed up Fermi so badly initially. Had nVidia gotten Fermi right the first time and had released GTX500 level parts from the get-go, Evergreen wouldn't have stood a chance and it would have been AMD scrambling for Northern Islands.

Not so sure we can count on nVidia fumbling with Kepler as badly as they did with Fermi

 

[Panopticon]

I'm not sure what to make of this as the information regarding the names of their chips seems confusing to say the least. I have seen some conflicting answers regarding the internal naming of their chips so is he saying that a 670 or 660 part beats a 7970? I fully expect nvidia to have a more powerful flagship card and I bet this generation ends up being like 6970 vs 580 but if a 660 part beats the 7970 AMD is in trouble. The problem with that is that I don't see that even being a possibility. Could be SA just screwing around honestly because most of the time even their "leaks" have more concrete info than this.

 

[LOL_Wut_Axel]

AMD has been better about pricing and efficiency for consumer concerns, however GCN is currently a step in the opposite direction in those areas. The 7970 is not only the most expensive single GPU card AMD has released since acquiring ATI, its currently drawing more power under load than the 6970 falling just short of the GTX580. Clearly AMD had to sacrifice gaming-oriented performance and power efficiency with this architecture.

Since nVidia already made that step towards HPC several generations ago, it really wouldn't be too surprising if AMD ends up disadvantaged overall this round. Luckily for AMD (and all of us really) they've got a head start in this race, as I think they'll need it.

AMD caught a pretty big break when nVidia screwed up Fermi so badly initially. Had nVidia gotten Fermi right the first time and had released GTX500 level parts from the get-go, Evergreen wouldn't have stood a chance and it would have been AMD scrambling for Northern Islands.

Not so sure we can count on nVidia fumbling with Kepler as badly as they did with Fermi

I'm not really sure if you're serious with this. Every new GPU release has meant higher power consumption, but that doesn't mean lower power efficiency. The HD 5870 consumed more power than the HD 4890 but was also more efficient due to the huge increase in performance, while the HD 6970 consumes around 5% more power than the 5870 and is 15% faster, meaning it was also a step up in efficiency. The HD 7970 is on par with the GTX 570 when it comes to power consumption, meaning it consumes 6% more power than the HD 6970. However, it's also 35% faster than the HD 6970, so efficiency has gone up. Also, remember that AMD purposefully lowered the clocks of the HD 7970, similar to what NVIDIA did with the GTX 460. Like I said, it's a GPU that with minor tweaks can get a 30% performance boost.

I don't think Kepler will be a failure, but I just don't expect it to be better than Southern Islands overall.

 

[LOL_Wut_Axel]

I'm not sure what to make of this as the information regarding the names of their chips seems confusing to say the least. I have seen some conflicting answers regarding the internal naming of their chips so is he saying that a 670 or 660 part beats a 7970? I fully expect nvidia to have a more powerful flagship card and I bet this generation ends up being like 6970 vs 580 but if a 660 part beats the 7970 AMD is in trouble. The problem with that is that I don't see that even being a possibility. Could be SA just screwing around honestly because most of the time even their "leaks" have more concrete info than this.

By the time the Enthusiast Kepler cards are here AMD will be able to release a new HD 8000 series...

 

[BallaTheFeared]

Fermi fumbled right into me getting two great cards for $300 total... I hope kepler "fumbles' in the eyes of the world, so I can get another great deal on some amazing cards.

I'd have to flash my bios higher than what I currently have (3500MHz default) for memory, but even with my slider maxed out (4550MHz) I only see a 7% gain in fps, from underclocked ram @ 900 core.

 

[IonusX]

See that arrow on the "A" in the logo? That's where this one landed...

id like to see your time machine?