New 28 nm Graphics Cards To Be 45 Percent Faster And Overclock Like Never Before?

[AsRock]

Sure their be a boost but 45% ?.. nVidia spread their shit AMD spread their shit and some one who makes the chips for them are spreading the shit for them ?.

Although i hope it turns out better thsn they think it will do so i have a excuse to get a new GPU because i'll need it for da wife lol.

 

[n-ster]

They aren't very good marketers if they would underestimate the speed of their new GPUs

Of course they will "spread their shit for them" as more sells = more profit for them too

 

[Benetanegia]

I'm very inclined to believe that a huge increase in clocks is posible (this time). They are jumping from bulk process to HKMG. It's been a long time too. This is not like when Intel made the jump almost 5 years ago, it's not a tech that is completely new and unproven, expertise on high-k materials is out there already, so all the cumulative improvements that we saw over the past 5 years may very well apply all at once on TSMC's 28nm and maybe they even got it "right" this time. 40nm was kinda awful, far worse than expected, 28nm just needs to be slightly better than average and it would already look much better.

The 45% figure seems exagerated, and it probably is, but 25% or some more seems reasonable. Now where's my 1200 Mhz (stock) GPU?

 

[AsRock]

They aren't very good marketers if they would underestimate the speed of their new GPUs

Of course they will "spread their shit for them" as more sells = more profit for them too

Over estimating just pisses people off. And in fact only facts sell me not this possible crap.. But hey hopefully they are being truthful we will see.


So maybe i should of just said I will believe it when i see it.

 

[Imhoteps]

45% smaller process= 45% power saving, does NOT equal 45% higher clocks. Idiots.

Making claims liek that need to be verified with an offical statement, not by an anonymous source. If there is no officail source, then it is jsut rumour, and shoud be ignored. I dunno why I still gotta say that...

Spelling, spelling...

 

[mediasorcerer]

I guess the biggest thing is with 28nm, taking advantage of the extra or increased bandwidth of pcie gen 3, that may account for the claims?
.

 

[Neuromancer]

I guess the biggest thing is with 28nm, taking advantage of the extra or increased bandwidth of pcie gen 3, that may account for the claims?
.

PCIE3 bandwidth is not bandwidth gain its latency
it has extra BW too, but the main thing is latency/ like imagine etherent with 2% overhead.. WOW that would be an almost 100% gain in speed.

Not as much on pcie but still impressive, and it leads to WAY more than gfx. It spells well for all kinds of devices.

 

[xenocide]

I guess the biggest thing is with 28nm, taking advantage of the extra or increased bandwidth of pcie gen 3, that may account for the claims?
.

I don't think so. GPU's aren't exactly Bandwidth starved. Going from 16x to 8x currently only results in a 1-3% performance drop, so I have severe doubts doubling the bandwidth will really accomplish a lot. We're not even sure the next generation of GPU's will natively be PCIe 3.0...

 

[Neuromancer]

I don't think so. GPU's aren't exactly Bandwidth starved. Going from 16x to 8x currently only results in a 1-3% performance drop, so I have severe doubts doubling the bandwidth will really accomplish a lot. We're not even sure the next generation of GPU's will natively be PCIe 3.0...

Its not about BW its about latency and overhead. (which implies BW.. but effectively)

 

[mediasorcerer]

Thats interesting to learn, i was just having a guess , but thanx for the info.

 

[Casecutter]

After reading what DailyTech’s said and all the comments here I say take a step back, and see this is what TSMC wants you to hear... and not from either Nvidia or ATI... WHY?

The main point of the write-up centers around "This speed improvement is based on the same leakage per gate". TSMCs' High-k Metal Gate (HKMG) and that it supports up to a 45 percent clock speed improvement over the firm's own 40G process used to make the last two generations of video cards.

Why is the supplier of this; their brand new 28 nm process talking about this? It’s like an OEM contract manufacture for Toyota or Honda, saying our new window regulator could do this and this. Does it matter..? Not really not until the customer spec’s their actual requirements. I see this as a case of we feel we got are crap right this time, and we don’t intend to be the fall guy (a second time) for blunder in the customer’s architecture.

The rest of the DailyTech’s write-up was fluff, while the cut-down synopsis here just watered it down more.

The Tahiti purportedly supports both Graphics Core Next (GCN) and XDR2 memory, and was said to be out end of 2011. I deem its’ been pushed back as AMD seeing Kepler being Q2 would rather hold out and tweak it more as it learns more on the competition. Although, it will release the first 28Nm graphics card in 2011; and it appears to be on the VLIW4 architecture released with Cayman. The question... is more transistors and TSMC new HKMG gates really going to boost performance and improve efficiency?

I'm thinking Tahiti is probably more end of January-February 2012.

 

[mediasorcerer]

Interesting analysis casecutter.thanx.

 

[N3M3515]

oooo sounds interesting, I'll definitely be expecting a big jump in performance similar to 4xxx series to 5xxx series AMD wise as the 5850 completely destroyed the 4850 in all aspects.

In all aspects but PRICE, where the 4850 completely destroyed it........

 

[Covert_Death]

i think we are all missing one important thing, don't expect 45% improvement on the FIRST itteration of 28nm tech... think about it, did the move to 40nm tech give max performance of 40nm? NO we are just now getting there after a few itterations. maybe in a few years when 28nm has reached its limit and we start to see the last few cards made on 28nm, maybe THOSE 28nm cards will offer 45% improvement over todays 40nm cards...

nvidia and AMD would kill future profits to make the best card for the next 5 years right now, what would they improve on the next few years, you have to offer something up gradable every year or your run out of buisness and sit idle.

 

[qubit]

i think we are all missing one important thing, don't expect 45% improvement on the FIRST itteration of 28nm tech... think about it, did the move to 40nm tech give max performance of 40nm? NO we are just now getting there after a few itterations. maybe in a few years when 28nm has reached its limit and we start to see the last few cards made on 28nm, maybe THOSE 28nm cards will offer 45% improvement over todays 40nm cards...

nvidia and AMD would kill future profits to make the best card for the next 5 years right now, what would they improve on the next few years, you have to offer something up gradable every year or your run out of buisness and sit idle.

You may well be right. It's certainly plausible and history has shown this pattern of improvement.