NVIDIA Kepler speculation thread

[Ailuros]

You ask the question and then answer it alone. Dump or simply postpone?

They didn't dump GF100 in favor of GF110 did they? Either way the performance part for Kepler was always planned to arrive first whether you want to believe it or not. GK104 taped out in fall 2011 and there was no high complexity chip tape out before that or anything close to that.

I might be expensive but this can in all cases be considered as investment and the technology will almost surely be used one day.
I wonder why you do bother about if it is expensive or not. NV has lots of money to dump.

Then you are obviously confusing NV with Intel.

 

[xDxD]

Considering it's Nvidia's only decently profitable segment, it doesn't make a whole lot of sense to be so late.

well, firts quarter doesn't appears so bad http://pressroom.nvidia.com/easyir/...releasejsp=release_157&xhtml=true&prid=755674 (and there wasn't new Geforce, new Tesla and new Quadro...) we'll see second quarter with the launch of kepler

 

[Man from Atlantis]

well, firts quarter doesn't appears so bad http://pressroom.nvidia.com/easyir/...releasejsp=release_157&xhtml=true&prid=755674 (and there wasn't new Geforce, new Tesla and new Quadro...) we'll see second quarter with the launch of kepler

If im not mistaken NV Fiscal year starts on February, so Q1 results include Kepler..

 

[A1xLLcqAgt0qc2RyMz0y]

If i'm not mistaken NV Fiscal year starts on February, so Q1 results include Kepler..

Yes the 1st quarter was from Feb 1 through Apr 30th.

However the Kepler GTX680 launched late in March (22nd). The GTX670 launched in May so it was not included in the 1st quarter.

Also remember the GTX was in extremely short supply so it really didn't add much to the 1st quarter.

The current quarter ending July 31st will have a full three months of GTX680 and 7 weeks of GTX670 availability and that should reflect positively in the earnings.

 

[trinibwoy]

Considering it's Nvidia's only decently profitable segment, it doesn't make a whole lot of sense to be so late.

What makes you think it's late? I love how people assume they know when products "should" be launched. The upgrade cycle in the pro segment is looooong.

 

[silent_guy]

What makes you think it's late? I love how people assume they know when products "should" be launched. The upgrade cycle in the pro segment is looooong.

Also, has there ever been a case where the Quadro/Tesla cards were launched together with the GeForce? The pro segment is not nearly as schedule and hit sensitive as the consumer segment. If a business needs as pro card it will buy what's available. It's not going to wait until the next gen comes out. It's not as if Nvidia has a Quadro competitor.

 

[Zaphod]

Yes the 1st quarter was from Feb 1 through Apr 30th.

However the Kepler GTX680 launched late in March (22nd). The GTX670 launched in May so it was not included in the 1st quarter.

The chips will surely have started shipping for revenue to partners in good time before product availability, which would probably place at least some GTX670 money in the Q1 accounts as well.

 

[A1xLLcqAgt0qc2RyMz0y]

The chips will surely have started shipping for revenue to partners in good time before product availability, which would probably place at least some GTX670 money in the Q1 accounts as well.

Very little if any GTX670 money was in the Q1. Shipping for revenue occurred mostly in early May.

Also remember that Nvidia's customers pay 30-45 days after receipt of product which pushes revenues further into Q2.

As for the GTX680, in Q1 they were made by Nvidia and just re-badged by the Vendors with "time to market" being just days from production to vendors to sales outlets to customer purchase.

 

[Zaphod]

Also remember that Nvidia's customers pay 30-45 days after receipt of product which pushes revenues further into Q2.

Whether some number goes into the current assets or outstanding claims row on the balance sheet doesn't matter one iota for the quarterly result. At the time product leaves your hands, it enters your books.

 

[Homeles]

They didn't dump GF100 in favor of GF110 did they? Either way the performance part for Kepler was always planned to arrive first whether you want to believe it or not. GK104 taped out in fall 2011 and there was no high complexity chip tape out before that or anything close to that.

You have nothing to back up that claim. What can be asserted without evidence can be dismissed without evidence.

 

[ninelven]

What can be asserted without evidence can be dismissed without evidence.

/jackie chan

Using that line of reasoning the entire topic does not merit discussion.

 

[Homeles]

/jackie chan

Using that line of reasoning the entire topic does not merit discussion.

So we're speculating randomly, completely disregarding how GPUs work, what Nvidia has done in the past, etc.?

Right.

 

[ninelven]

So you have evidence of a GK100? Or is that being asserted without evidence?

 

[Homeles]

So you have evidence of a GK100? Or is that being asserted without evidence?

Since Nvidia's default position since GT200 is to release 500mm²+ dies, the default position is to assume that at some point Nvidia was planning to release one for the 600 series. It is quite likely that at some point, GK100 was existed in some form -- be it a simple concept dreamed up by the engineering team, or a computer simulation, or a taped out chip.

It's far less likely that Nvidia had absolutely no intention of releasing a larger chip for the 600 series. It is likely that something happened -- whether it was the fault of TSMC, Nvidia, or both -- that made Nvidia decide not to go through with releasing the product.

The issue is that (s)he is making statements as if they were fact, but there is no evidence supporting that claim out there and also no evidence provided here. Speculating is fine -- claiming your baseless beliefs to be the truth is not.

The point of this thread is "what if?" It is not "I think Kepler is going to look like this, and all of you are wrong to think otherwise."

 

[ninelven]

I'll take that as a no?

 

[Homeles]

I'll take that as a no?

I see that you are illiterate, and I wonder why you are on the internet.

 

[imaxx]

I see that you are illiterate, and I wonder why you are on the internet.

Let me offer to the readers a quick analysis of your sentence: you are using an AD HOMINEM (latin term!) logic fallacy here.

Translating for the "not-literate" guys, it means you are attacking the arguer in place of the argument itself, in order to validate your reasoning.

In "the art of being right", the philosopher A. Schopenhauer classified it as the 38th trick to use -the last resort when you have no more bullets

 

[Homeles]

Let me offer to the readers a quick analysis of your sentence: you are using an AD HOMINEM (latin term!) logic fallacy here.

Translating for the "not-literate" guys, it means you are attacking the arguer in place of the argument itself, in order to validate your reasoning.

In "the art of being right", the philosopher A. Schopenhauer classified it as the 38th trick to use -the last resort when you have no more bullets

If he's not going to respect me by reading what I said, then I absolutely am not going to respect him. You can thank him for lowering the level of conversation.

Ad hominem is only an issue when I'm not attacking an argument... there's no argument to attack.

 

[Silent_Buddha]

I'll take that as a no?

He's basically saying that while most people here go with "I think..." hence speculation...

One or more people are going "I know..." as in it is pure fact and not speculation...

It's a complete fail of logic to assume that there were no plans for a GF100. How far GF100 got before it was cancelled is the only thing in question.

Nvidia has always lead with a base number.

NV10-40 and G60-80 (x0 number designations). With refreshes adding a 1 or a 5. G92 is the only real outlier here.

Since moving to 3 digit designations they've all launched with x00 designations for the first large chip of each new generation.

GT200, GF100. All first gen products have x0x while all refresh products have x1x.

Keplar follows suit. All first gen products carry x0x (GK104, GK107, etc.). Refresh products carry x1x. GK110 for example and probably eventually a GK114.

So, there's no doubt in my mind that GF100 existed at some point. Whether just computer simulations or taped out. They certainly weren't about to just skip a product cycle in the HPC/professional markets when they knew ATI was specifically targetting that with their next chip. Sure they are currently dominating that market segment, but to skip a product cycle gives your competition a big chance to get their foot in the door.

The only question is, at what point did Nvidia decide that GF100 was an unworkable product, even for the HPC/professional market, and scrap it in favor of dedicating more resources to GF110 or even Maxwell.

Regards,
SB

 

[KimB]

So you have evidence of a GK100? Or is that being asserted without evidence?

This is pretty darned good evidence that such a part is in the works:
http://www.atomicmpc.com.au/News/301204,nvidia-big-kepler-gk-110-announced-for-supercomputers.aspx

Whether or not there was ever a GK100 is pretty much an academic question at this point. But it is highly likely that we will see a "big Kepler" consumer part within the next 18 months.