nVidia wins this round - Charlie D.

[AtenRa]

@ Grooveriding

What makes you think NV will release the GK104 first ??

 

[SirPauly]

And that's the key how will the GK-104 compare to the GTX-580? Time will tell.

 

[SirPauly]

I do agree with buyers that demand the best performance right now may buy the HD7970 and in that context, the choice is still very welcomed for early adopters or enthusiasts that desire the latest and greatest.

 

[Arkadrel]

A 580 on 28nm would be like 310mm^2.

That would still be a nice card though.
That would make it around ~18% smaller than the 7970.

With some higher clocks on that card, it wouldnt be 24% slower on avg than the 7970. Im guessing just clocking it higher could catch up most of that differnce (at the cost of higher TPD).

It would still be a awesum card, that could be sold cheaper say....
440$ or so? and be near equal to the 7970.

 

[SirPauly]

@ Grooveriding

What makes you think NV will release the GK104 first ??

I still have no clue what they will do because nVidia themselves are so quiet.

 

[Olikan]

Wow, that brings everything into focus

With statements like that, it's looking like CD is only looking for web hits.

well, yes and no...

it seems nvidia didn't screw up this time, and they managed to bring the performance, nm², watts, yields, heat... at radeon levels....something that charlie was and still is very vocal at.

but the "win all around" was indeed web hits

 

[blastingcap]

Wow, that brings everything into focus

With statements like that, it's looking like CD is only looking for web hits.

Are you guys sarcastic or something? Because Charlie said that it's mostly interesting on the financial side. If it's just on raw performance, he never said GK104 is faster than 7970, and in fact it's implied by his choice of words (Kepler vs GCN arch) that he's comparing price/perf or perf/watt or perf/die size.

In other words, the GK104 will probably be more efficient in most metrics than its AMD counterpart, which is probably either the 7950 or 7870. So Nvidia will make more money than AMD, even if the price/perf to the consumer is comparable on both sides.

Hence Charlie telling that guy to stick to the 7970 for fastest performance.

People ought to actually read what he said, before implying that he said something he didn't.

 

[Grooveriding]

@ Grooveriding

What makes you think NV will release the GK104 first ??

I won't round them all up, but every rumour for months; since we've heard inklings of nvidia on 28nm - has been saying mid-range first. Now these more recent rumours are saying, again, mid-range. I'd also say when is the last time you have seen high-end and mid-range come together ? It will be one, or the other. Rumours are saying it's mid-range.

Frankly, I'm still skeptical they have anything close to ready. Until I see leaked board shots on chiphell - there is nothing close to ready.

I am ruminating on what little there is in way of rumour, and that all is saying a mid-range card. For me it's very dissapointing if accurate that again nvidia can't execute well on delivering their high-end cards in a timely manner to compete with AMD. A mid-range card is nice as there are no mid-range 28nm options right now for people wanting to spend in that bracket, but not what I'm looking for.

 

[exar333]

A 580 on 28nm would be like 310mm^2.

That would still be a nice card though.
That would make it around ~18% smaller than the 7970.

With some higher clocks on that card, it wouldnt be 24% slower on avg than the 7970. Im guessing just clocking it higher could catch up most of that differnce (at the cost of higher TPD).

It would still be a awesum card, that could be sold cheaper say....
440$ or so? and be near equal to the 7970.

A straight-up die-shrink of the 580 (with 3GB RAM) with a price between $300 and $350 would be killer. This is what I want to see! This would be a great intro 28nm product on the mid-range, with the high-end to follow.

 

[SirPauly]

Are you guys sarcastic or something? Because Charlie said that it's mostly interesting on the financial side. If it's just on raw performance, he never said GK104 is faster than 7970, and in fact it's implied by his choice of words (Kepler vs GCN arch) that he's comparing price/perf or perf/watt or perf/die size.

In other words, the GK104 will probably be more efficient in most metrics than its AMD counterpart, which is probably either the 7950 or 7870. So Nvidia will make more money than AMD, even if the price/perf to the consumer is comparable on both sides.

Hence Charlie telling that guy to stick to the 7970 for fastest performance.

People ought to actually read what he said, before implying that he said something he didn't.

https://twitter.com/#!/CDemerjian/status/160082027255828480

 

[MrK6]

https://twitter.com/#!/CDemerjian/status/160082027255828480

Lol, more or less. And when someone says "just about every metric," well, "performance" is a metric that's pretty high on everyone's list.

 

[blastingcap]

https://twitter.com/#!/CDemerjian/status/160082027255828480

Talk about selective reading.

http://twitter.com/#!/CDemerjian/status/160089353861861378

The fact that Charlie compared arch and not particular chips against each other, coupled with his fascination with die sizes and financials, tells you want you need to know. But some people don't read between the lines or even the lines themselves apparently and interpret his article to mean GK104 > 7970 which is definitely not what he said.

 

[SirPauly]

Forums are fun and entertaining!

He may be talking about winning metrics like Performance/watt; Performance/nm; Performance/dollar with the HD 7970 vs GK104

 

[blastingcap]

Forums are fun and entertaining!

He may be talking about winning metrics like Performance/watt; Performance/nm; Performance/dollar with the HD 7970 vs GK104

Yes, more efficient, like I translated it. Doing more with less, on price/perf, perf/die size, thermals, energy efficiency, etc. He did say that it wins handily in most metrics, implication being, most metrics that he obsesses about such as perf/die size.

You then posted a twitter link that added no new info.

Since you like twitter so much, I posted a more telling tweet by Charlie D.

And now you are seeing the light. Better late than never I guess, but that was what anyone closely reading what he wrote saw all this time: he's saying it's a more efficient architecture than flawed Fermi, at least for gaming, and noted its relatively small die size and implications for cost and thus profit potential for Nvidia.

Charlie is not a typical gamer, in that he cares more about the financial implication of chips, than in the actual price or performance. He apparently only cares about price/perf and thermals and such to the extent that it would affect revenue and profit. But then he is a "journalist" and is trying to drum up a consulting business, so naturally he's playing his incentives. So it's perfectly understandable why he doesn't view the world the way a gamer does.

Nvidia will likely not heavily discount anything, but price them at just low enough a point (or with extra goodies like free games) to claw back lots of market share. Do not expect for a microsecond for GK104 to be faster than 7970 yet priced at $300.

 

[AtenRa]

I won't round them all up, but every rumour for months; since we've heard inklings of nvidia on 28nm - has been saying mid-range first. Now these more recent rumours are saying, again, mid-range. I'd also say when is the last time you have seen high-end and mid-range come together ? It will be one, or the other. Rumours are saying it's mid-range.

Frankly, I'm still sceptical they have anything close to ready. Until I see leaked board shots on chiphell - there is nothing close to ready.

I am ruminating on what little there is in way of rumour, and that all is saying a mid-range card. For me it's very dissapointing if accurate that again nvidia can't execute well on delivering their high-end cards in a timely manner to compete with AMD. A mid-range card is nice as there are no mid-range 28nm options right now for people wanting to spend in that bracket, but not what I'm looking for.

There are a few things that make me believe TSMCs 28nm HP is in much better position than 40nm was at the same time (first GPU launches from AMD & NV) and NV is closer to release a high end card than most people think.

1: No small die chips launched yet. AMD went straight for a big die chip at 28nm.
Usually when yields are low you manufacture the smaller die chips first (entry and low level) in order to make a profit and raise yields for the bigger die chips.
That happened with both AMD and NV at 40nm, AMD launched the HD4770 and NV launched the GT210 before they begun the production of their biggest die chips (AMDs Evergreen and NVs GF100) .
This time at 28nm there are no small die chips yet.

2: TSMCs 28nm is the second try with 193nm immersion, so i believe they didnt have a lot of problems like their first try at 40nm. Their first try with HKMG caused them a lot of difficulties but i believe they overcome them by now.
They have also introduced RDR (Restricted Design Rules) at 28nm and that could make things better.

3:One month after AMD's HD7970 GCN introduction and NV haven't even bothered to release any info about Kepler yet. Last time with Fermi they rushed to release the architecture before the end of September 2009, a few weeks after AMDs Evergreen (HD58xx) launch.

To sum up, 28nm must be easier both for TSMC and AMD/NV and i believe NV will launch a high end card first and then the middle range GK104. unless the GK104 is not a middle range chip.

 

[Grooveriding]

I would hope a mid-range card would have better performance/dollar than a top end part like the 7970. Isn't that the point of mid-range cards ?

I think some are beaming in their information in a manner similar to this

 

[blastingcap]

3:One month after AMD's HD7970 GCN introduction and NV haven't even bothered to release any info about Kepler yet. Last time with Fermi they rushed to release the architecture before the end of September 2009, a few weeks after AMDs Evergreen (HD58xx) launch.

To sum up, 28nm must be easier both for TSMC and AMD/NV and i believe NV will launch a high end card first and then the middle range GK104. unless the GK104 is not a middle range chip.

How noisy or silent was Nvidia right before dropping the G80, better known as the legendary 8800 GTX?

AMD needs to watch its back. It's fighting a losing war with Intel, draining resources at a time when a well-financed Nvidia is seeking to regain its lost market share.

http://www.nasdaq.com/article/amd-slips-to-loss-in-q4---quick-facts-20120124-01430

 

[Grooveriding]

There are a few things that make me believe TSMCs 28nm HP is in much better position than 40nm was at the same time...

TSMC 28nm process is in trouble Opinions are that some of the issues of the unmentioned chips are related to designs not being optimized for the process.

Sounds just like 40nm to me. Possibly marginally worse/better if you were to know the specifics.

And about this comment you made

3:One month after AMD's HD7970 GCN introduction and NV haven't even bothered to release any info about Kepler yet. Last time with Fermi they rushed to release the architecture before the end of September 2009, a few weeks after AMDs Evergreen (HD58xx) launch.

I'm confused by this. AMD released 58XX in September 09 and nvidia released nothing until 6 months later in April 2010, not 'a few weeks after', in that time there was nothing but rumours and nothing concrete about the card beyond a fake launch at GTC 09. Exactly what we are seeing so far again on 28nm, a delay with nothing in sight or confirmed by nvidia and just rumours.

 

[SirPauly]

In other words, the GK104 will probably be more efficient in most metrics than its AMD counterpart, which is probably either the 7950 or 7870. So Nvidia will make more money than AMD, even if the price/perf to the consumer is comparable on both sides.

People ought to actually read what he said, before implying that he said something he didn't.

That's not what he said:

https://twitter.com/#!/CDemerjian/status/160082027255828480

People ought to actually read what he said, before implying that he said something he didn't -- the irony of your own advice!

 

[blastingcap]

That's not what he said:

https://twitter.com/#!/CDemerjian/status/160082027255828480

People ought to actually read what he said, before implying that he said something he didn't -- the irony of your own advice!

Some people read the titles of books, other people actually read the inside text of the book without caring much about the title. Guess which category of people is more informed?

P.S. Of course, even better informed are those who read the addendums and erratas that come out after the first edition of the book was published.

 

[SirPauly]

The context is usually pretty important, one may imagine!

 

[AtenRa]

I'm confused by this. AMD released 58XX in September 09 and nvidia released nothing until 6 months later in April 2010, not 'a few weeks after', in that time there was nothing but rumours and nothing concrete about the card beyond a fake launch at GTC 09. Exactly what we are seeing so far again on 28nm, a delay with nothing in sight or confirmed by nvidia and just rumours.

I mean NV rushed to release the information about Fermi architecture, they rushed to release something about their next architecture because they knew they were behind schedule.
This time no information about Kepler architecture,nothing at all.

 

[SirPauly]

I would hope a mid-range card would have better performance/dollar than a top end part like the 7970. Isn't that the point of mid-range cards ?

I think some are beaming in their information in a manner similar to this

Talk to Charlie; he was the one with beaming nVidia accolades. Personally have no idea what nVidia may do.

 

[BallaTheFeared]

I would hope a mid-range card would have better performance/dollar than a top end part like the 7970. Isn't that the point of mid-range cards ?

I think some are beaming in their information in a manner similar to this

I would expect large improvements from a die shrink and core redsign, currently AMD is positioning themselves to give us mid-high end people nothing to that effect.

 

[Grooveriding]

I mean NV rushed to release the information about Fermi architecture, they rushed to release something about their next architecture because they knew they were behind schedule.
This time no information about Kepler architecture,nothing at all.

There was GTC 2009 prior, but that was just a case of timing. Last year's GTC was October. In terms of information released after 5870/5850 prior to 480/470 there was exactly what we see now; nothing but rumours.

What to believe. If you believe this one GK104 should be 384 bit not 256

There is nothing real or factual anywhere, just like we saw with fermi. For all we know nvidia won't even release a mid range card before the Summer at this rate. Until there are board shots, there is nothing.

And with the aforementioned TSMC issues I linked, it does not look good for anyone on 28nm TSMC. Nvidia releasing these big die chips for their halo parts is starting to hamper their ability to release on new nodes competitively against AMD - first with fermi now with kepler. With the 28nm tsmc issues I can see them having the same problems trying to put out their halo card. So I find the possibility of a mid-range rumour to be the most plausible.

But again, there are no board shots or rumours that are not contradicted by other rumours. We may very well continue to see nothing at all until May, just like we saw on 40nm with nvidia. Lots of rumours for six months followed by a release.