NVIDIA preparing four Maxwell GM204 SKUs (VideocardZ via S/A)

[ShintaiDK]

Q2 2010

We talk to AMD about their new Market Share Lead

http://www.decryptedtech.com/editorials/we-talk-to-amd-about-their-new-market-share-lead

I honestly doubt the numbers. I will try see if I can find numbers with Q2 2010. But This one is from Q4 2010. And Last quarter would be Q3 2010.

That would have been as fast a loss as it was gained.

But I assume the numbers you linked are simply manipulated or wrong.

 

[SirPauly]

Q2 2010 was 51.8 total discrete market share for AMD and 48.8 for nVidia -- nVidia still did hold on to desktop discrete leadership according to Mercury Research!

 

[ShintaiDK]

Q2 2010 was 51.8 total discrete market share for AMD and 48.8 for nVidia -- nVidia still did hold on to desktop discrete leadership according to Mercury Research!

Are you talking discrete, or are you talking different discrete segments? AMD have always been bad in the mobile segment for example. So its not there their share was. The only way to reach 51.8% would be desktop discrete cards only. If not isolated to the DIY channel as its seen before with inflated CPU share numbers.

 

[SirPauly]

Talking about total discrete which includes discrete mobile and discrete desktop! AMD did very well with mobile discrete based on their efficient designs in that time-line!

 

[Cloudfire777]

So you think GM204 will only be 40-45% more efficient than GK110, even though GM107 is 70% more efficient than all of Kepler?

I`m saying GM204 cores will have 35% better performance than GK110 cores

GTX 750 Ti runs at 1080MHz so Maxwell seems to do fine on high clocks. GTX 780 Ti runs at around 1000MHz and I`m thinking GTX 880 Ti might hit 1080MHz too, so that is where the extra performance comes from.
If GTX 880 runs at the same clock as GTX 780 Ti, we are looking at a +20% over 780 Ti assuming GM204 will have a +35% performance/core over Kepler.

 

[RussianSensation]

You are essentially complaining that Intel doesn't sell you their 16 Core CPU for 300€. It's the exact same situation, but I see no one complaining.

What? Our CPU forum is full of threads of people complaining about the lack of progress being made in the CPU space. Outside of video encoding/rendering and the like, Intel had released nothing worth buying since Jan 3, 2011 when 2600k came out. You can cherry pick all you want but you will be hard pressed to find anything faster than 2600k @ 4.8 Ghz almost 4 years after it launched. The CPU space is pathetic. Since i7 920@4.0Ghz came out in Nov 2008, Intel hasn't produced any viable CPU upgrade for 95% of consumers. If you take a system with a 2008 I7 920 @ 4.0ghz and pair it with 7970Ghz CF, you can't find anything more than 60% faster for games right now. That's almost unbelievable in computer timeline terms. Moore's Law is out the window in CPU choices and it is near EOL for CPUs.

 

[witeken]

What? Our CPU forum is full of threads of people complaining about the lack of progress being made in the CPU space. Outside of video encoding/rendering and the like, Intel had released nothing worth buying since Jan 3, 2011 when 2600k came out. You can cherry pick all you want but you will be hard pressed to find anything faster than 2600k @ 4.8 Ghz almost 4 years after it launched. The CPU space is pathetic. Since i7 920@4.0Ghz came out in Nov 2008, Intel hasn't produced any viable CPU upgrade for 95% of consumers. If you take a system with a 2008 I7 920 @ 4.0ghz and pair it with 7970Ghz CF, you can't find anything more than 60% faster for games right now. That's almost unbelievable in computer timeline terms. Moore's Law is out the window in CPU choices and it is near EOL for CPUs.

Moore's Law is still going on.

You must blame quantum mechanics, not Intel. The exact same story could be told if you replace Intel with AMD.

Haswell was a superb update of the Core architecture.

 

[KaRLiToS]

Moore's Law is still going on.

You must blame quantum mechanics, not Intel. The exact same story could be told if you replace Intel with AMD.

Haswell was a superb update of the Core architecture.

Superb update?

We don't have the same definition of "Superb"

 

[notty22]

Superb..

Christine

 

[mla]

I`m saying GM204 cores will have 35% better performance than GK110 cores

GTX 750 Ti runs at 1080MHz so Maxwell seems to do fine on high clocks. GTX 780 Ti runs at around 1000MHz and I`m thinking GTX 880 Ti might hit 1080MHz too, so that is where the extra performance comes from.
If GTX 880 runs at the same clock as GTX 780 Ti, we are looking at a +20% over 780 Ti assuming GM204 will have a +35% performance/core over Kepler.

what Source?
tank

 

[witeken]

Superb update?

We don't have the same definition of "Superb"

Because you don't have a Haswell laptop? You don't use AVX2.0? You don't use TSX? And don't forget GT3 and Crystalwell. For most consumers it isn't a big upgrade, but that doesn't mean Haswell isn't a great update.

 

[KaRLiToS]

what Source?
tank

This is obviously from a nVidia presentation slide show.

Because you don't have a Haswell laptop? You don't use AVX2.0? You don't use TSX? And don't forget GT3 and Crystalwell. For most consumers it isn't a big upgrade, but that doesn't mean Haswell isn't a great update.

I didn't say there are no updates. I just meant that I don't call that a "Superb" update. A "tiny" update would have been more appropriate.
__________________
No flame intended.

 

[NTMBK]

Because you don't have a Haswell laptop? You don't use AVX2.0? You don't use TSX? And don't forget GT3 and Crystalwell. For most consumers it isn't a big upgrade, but that doesn't mean Haswell isn't a great update.

AVX2 is a fairly nice instruction set, but frankly until Intel sorts out its implementation of gather it's not going to be much use. I hope that Gather performance is fixed in Skylake, or it's going to have similar problems with AVX-512.

 

[Galatian]

What? Our CPU forum is full of threads of people complaining about the lack of progress being made in the CPU space. Outside of video encoding/rendering and the like, Intel had released nothing worth buying since Jan 3, 2011 when 2600k came out. You can cherry pick all you want but you will be hard pressed to find anything faster than 2600k @ 4.8 Ghz almost 4 years after it launched. The CPU space is pathetic. Since i7 920@4.0Ghz came out in Nov 2008, Intel hasn't produced any viable CPU upgrade for 95% of consumers. If you take a system with a 2008 I7 920 @ 4.0ghz and pair it with 7970Ghz CF, you can't find anything more than 60% faster for games right now. That's almost unbelievable in computer timeline terms. Moore's Law is out the window in CPU choices and it is near EOL for CPUs.

What? While this is completely off-topic, you are only looking at raw performance. Performance per Watt has increased tremendously and so has iGPU performance. I can get a tablet PC (Surface 3) with which I can play CiV on the go for example. That was simply not possible with Nehalem/Sandy/Ivy.

Also there is more raw performance to be had, because you can get socket 2011 CPUs but we see the core scaling on current games.

The point is: the progress that has been made on CPUs in the last year caters 99% of all users...power efficiency is what matters now. If you need raw performance like in the server space Intel even makes CPUs for you.
We just have to face it: we gamers just are not a big enough group to get dedicated CPUs for. Blame it on part to the developers so refuse to get their code enhanced for more then 4 cores.

 

[mla]

This is obviously from a nVidia presentation slide show.



I didn't say there are no updates. I just meant that I don't call that a "Superb" update. A "tiny" update would have been more appropriate.
__________________
No flame intended.

This is not accurate calculation

 

[stahlhart]

Get this thread back on topic now.
--stahlhart

 

[tviceman]

I`m saying GM204 cores will have 35% better performance than GK110 cores

GTX 750 Ti runs at 1080MHz so Maxwell seems to do fine on high clocks. GTX 780 Ti runs at around 1000MHz and I`m thinking GTX 880 Ti might hit 1080MHz too, so that is where the extra performance comes from.
If GTX 880 runs at the same clock as GTX 780 Ti, we are looking at a +20% over 780 Ti assuming GM204 will have a +35% performance/core over Kepler.

But you said it would have a 225watt TDP, which is only slightly less than GK110. If Maxwell is 70% more efficient per watt than Kepler, then a 225 watt TDP should translate into ~50% better performance.

 

[blackened23]

Arguing semantics over high end and mid range is absolutely stupid because of two reasons. One, if the estimates for the CUDA cores being implemented on GM200 are accurate (i've read 5000-8000) then there's absolutely no way that is feasible on 28nm. Therefore they have to wait on TSMC's 16nm to push forward with GM200.

Aside from this, whatever is the highest performing chip on the market at the time of release is high end.

Period. End of story. You can argue all day long about something being "mid range" but if the next tier of chip is having yield issues, or more to the point, limitations due to the node being used (as is probably the case with GM200), then the fact is, GM204 will be the high end when it launches - because it will likely be the single GPU king in terms of performance.

I mean if someone wants to be silly, do they want to argue that Tahiti was "mid range" despite being massively overpriced and relatively underperforming at launch? Especially given the software deficiencies that plagued Tahiti for 1.5 years post launch especially. No, when Tahiti launched it was high end despite the uarch being displaced by Hawaii some time later. Hawaii wasn't ready, so what WAS ready was high end. When GK104 was launched, GK110 had yield issues which allowed limited quantities. GK110 wasn't ready in March 2012. GK104 was. So because GK104 existed, and GK110 did not, GK104 was the new high end due to being the single GPU performance leader.

In this case, I really don't think it's an issue of NV holding GM200 back despite how desperately some people would push that view. If it needs as much transistors as I think it will, for more than 5000 CUDA cores, make no mistake, that just is not possible on 28nm. There are gains possible on 28nm, but there are very real limitations on how far you can stretch it, and that is a fact.

 

[Mand]

Aside from this, whatever is the highest performing chip on the market at the time of release is high end.

And there will always be something better next year.

 

[ShintaiDK]

And there will always be something better next year.

Sometimes it takes 2 years

 

[n0x1ous]

do they want to argue that Tahiti was "mid range" despite being massively overpriced and relatively underperforming at launch? Especially given the software deficiencies that plagued Tahiti for 1.5 years post launch especially. No, when Tahiti launched it was high end despite the uarch being displaced by Hawaii some time later. Hawaii wasn't ready, so what WAS ready was high end. When GK104 was launched, GK110 had yield issues which allowed limited quantities

The difference is, history taught us that Nvidia would have a larger chip coming along. Tahiti at its die size was about what we would expect for AMD's top chip for a node considering that AMD had not crossed 400mm2 since R600 in 2007. Hawaii was a surprise. GK110 was not - hence the midrange talk for GK104 but not for Tahiti

 

[tviceman]

The difference is, history taught us that Nvidia would have a larger chip coming along.

But, as has already been said here several times, there is always something better coming.

Tahiti at its die size was about what we would expect for AMD's top chip for a node considering that AMD had not crossed 400mm2 since R600 in 2007. Hawaii was a surprise. GK110 was not - hence the midrange talk for GK104 but not for Tahiti

Yet, at 65nm AMD released a bigger, faster chip than the hd4870 (hd5890). On 40nm, they released a bigger, faster chip than the hd5870 (hd6970). And now history has repeated itself at 28nm. AMD released a bigger, faster chip than the hd7970 (r9 290x). Help me out here, I can't see a pattern. Can you???? Tahiti wasn't considered mid-range, at all, under any circumstances. Double standards are double standards.

Blackend is right - whatever the fastest chip is on the market is the high end. If you don't want to buy it because it doesn't meet your die size requirements, then sit back on the bench.

 

[n0x1ous]

But, as has already been said here several times, there is always something better coming.



Yet, at 65nm AMD released a bigger, faster chip than the hd4870 (hd5890). On 40nm, they released a bigger, faster chip than the hd5870 (hd6970). And now history has repeated itself at 28nm. AMD released a bigger, faster chip than the hd7970 (r9 290x). Help me out here, I can't see a pattern. Can you???? Tahiti wasn't considered mid-range, at all, under any circumstances. Double standards are double standards.

Blackend is right - whatever the fastest chip is on the market is the high end. If you don't want to buy it because it doesn't meet your die size requirements, then sit back on the bench.

Better yes - bigger? not necessarily. 6970 to 7970...

Pretty sure 4870 was 55nm not 65 and there was no hd 5890. if you meant 4890 that was 55nm as well and the same chip as 4870. I dont disagree with your overall point necessarily, just that the market has/had different expectations for what the top chip in a given node would be - size wise for the two vendors..... Hawaii was a surpise GK110 was not

as you say, I will indeed be sitting on the sidelines till the big die's come out......(nvidia - a certaintiy) (AMD - maybe)

 

[SirPauly]

Where the price/performance was more-so evolutionary and incremental was with the introduction of 28nm with AMD's 7970 and 7950! The days of sweet spot pricing was over. Still glad to see strong competition and choice for many price-points!

 

[f1sherman]

But you said it would have a 225watt TDP, which is only slightly less than GK110. If Maxwell is 70% more efficient per watt than Kepler, then a 225 watt TDP should translate into ~50% better performance.

70% more efficient AT WHAT CLOCKS / TDP?

How about at same/similar clocks @150W and giving same perf. as 780 Ti?
(My uneducated extrapolation @2560 Maxwell cc from this )

Arguing semantics over high end and mid range is absolutely stupid

PERIOD!

I mean if someone wants to be silly, do they want to argue that Tahiti was "mid range" despite being massively overpriced and relatively underperforming at launch?

How about AMD milking us through-out the history with their small sub 400, sub 300mm2 dies?

I am pretty sure that behaviour used to be called - engineering marvel, efficiency, AMD die-size advantage over brute-force NVidia with their huge monolithic dies (what ever that means LOL).

Now that Nvidia does it - all of sudden that's MILKING IT, and ripping off their customers.

So which one is it?

• Is AMD able to produce 500mm2 chip or not? If YES, by the same logic are they not milking us and holding back with performance all this time?. If NOT, wth are they really that incompetent?
• Is it OK if Nvidia produces small sized chip with great performance/TDP? Or is it OK only if they don't release even bigger chip down the road