NVIDIA shows signs ... [2008 - 2017]

[dkanter]

Thanks david, tho ive already read your article before

I wasnt sure on this but are IBM and GF's process pretty much the same? And i have the same query as 3dilettante, will that make any difference?

Gate last should mean higher pfet performance.

IBM and GF aren't exactly the same, but they will be very close. GF isn't presenting separate data at IEDM yet, but I'm hoping they will some day.

DK

 

[BRiT]

Nvidia cuts Tesla's performance (By Lawrence Latif)

DESIGNER OF WARM GPUS Nvidia has cut the performance of its Fermi based Tesla GPGPU aimed at the high performance computing (HPC) market.

The latest specifications detail a 18 per cent clock decrease, coupled with a previously announced 12.5 per cent cut in the number of "stream processors" from 512 to 448, and the board tipped up drawing 10 per cent more power than previous vague estimates that had been bandied about by the firm late last year.

This is perhaps not surprising given the months of delays that Nvidia has had trying to get its Fermi GPUs out the door. Regardless of who you believe, even the most fortuitous yield reports are abysmal and Nvidia's repeated attempts to rein in the power consumption of a chip that has the same thirst for power as a tinpot dictator seem to have failed, with the firm even having admitted that its Fermi GTX480 GPU chip runs hot.

 

[pcchen]

I thought C2050 was 1.15GHz from the start. Since M2050 has the same frequency as C2050, I don't see any "failed promises" here.

 

[hoom]

And Charlies' take
http://www.semiaccurate.com/2010/05/05/nvidia-downgrades-tesla-again/
Ouch.
Thats some serious performance slippage there

 

[Silent_Buddha]

I thought C2050 was 1.15GHz from the start. Since M2050 has the same frequency as C2050, I don't see any "failed promises" here.

Perhaps the cut is in reference to the information Nvidia was giving out to potential Fermi based Tesla customers at the end of last year? Which would make them far less attractive to those same potential customers.

Regards,
SB

 

[pcchen]

This is hardly news worthy though. C2050 was already known to be 1.15GHz almost a month ago. M2050 is the same (M2050 is basically a fanless version of C2050). If they think this is a big deal, they should do that when C2050 was released.

 

[hoom]

Its a big deal because at NVs Evergreen spoiler 'launch' they promised 512 SP, 1.5ghz clock & 2.5ghz memory within 175W & during 2009.
They have fallen a very long way short of all of them.

 

[Sontin]

Its a big deal because at NVs Evergreen spoiler 'launch' they promised 512 SP, 1.5ghz clock & 2.5ghz memory within 175W & during 2009.
They have fallen a very long way short of all of them.

They only promised 8x DP performance over GT200.

Yeah, i hope you don't believe this "may numbers". They came from the source who promised that cypress will faster in DX10 and DX11...

 

[Malo]

They only promised 8x DP performance over GT200.

Yeah, i hope you don't believe this "may numbers". They came from the source who promised that cypress will faster in DX10 and DX11...

"may numbers" ?

 

[CarstenS]

Its a big deal because at NVs Evergreen spoiler 'launch' they promised 512 SP, 1.5ghz clock & 2.5ghz memory within 175W & during 2009.
They have fallen a very long way short of all of them.

I'd really love to see the source of this. Could you please, hoom?

 

[no-X]

"There will be 512 shader parts on both sides" [Tesla/GeForce] --- nVidia, supercomputing conference

nV's PR promised power consumption slightly under GTX28x levels, he also promised 2009 launch

for "8x the peak double precision floating point performance over GT200" (Fermi_Compute_Arch_Whitepaper.pdf) it is needed to achieve at least 1400 MHz on 512 SPs

no idea where the "2.5ghz memory" came from...

 

[Jawed]

So the key numbers to work out are: performance/watt and performance/$. What was promised and what's been delivered.

Jawed

 

[Kaotik]

no idea where the "2.5ghz memory" came from...

Probably means 1250MHz GDDR5, some places just double it to make it "more understandable for some people" (though imo it's more confusing)

 

[hoom]

some places just double it to make it "more understandable for some people"

places like the NV product guide...
It says 1.5ghz in the new specs & 1.8-2.0ghz in the screenshotted Nov 09 one.
I don't know where Charlie got 2.5ghz but given that the Nov specs were in other ways scaled back its likely that mem clock was too.

 

[Sontin]

nVidia is a very boring company. Even in Q1 they increased their revenue and profit.

-- Revenue grew 2 percent quarter-on-quarter to $1.0 billion
-- GAAP net income increased to $137.6 million, or $0.23 per diluted
share
-- GAAP gross margin improved to 45.6 percent

http://phx.corporate-ir.net/phoenix.zhtml?c=116466&p=irol-newsArticle&ID=1426701&highlight=

 

[Raqia]

nVidia is a very boring company. Even in Q1 they increased their revenue and profit.


http://phx.corporate-ir.net/phoenix.zhtml?c=116466&p=irol-newsArticle&ID=1426701&highlight=

Looks like they're doing just fine w/ their low end stuff, which probably accounts for most of their revenues; this is where AMD can make a major dent w/ Fusion. In the high-end, doesn't seem like GPU based computer has taken off yet, though I'm sure they still make a good deal of money off of workstation boards. The sliver of performance enthusiast products probably doesn't contribute quite so much to the bottom line, so all the drama of the GTX480 vs the HD5870 isn't going to hit profits so hard.

 

[Silent_Buddha]

Interesting, Nvidia did slightly better Quarter to Quarter but worse Year over Year than ATI in Revenue.

Quarter to Quarter
Nvidia: +1.9%
ATI: -2.8%

Year over Year
Nvidia: +50.8%
ATI: +87.6%

Although this isn't exactly comparable, quarters differ by over a month, and I'm not sure if MB chipsets would be included with the CPU side of AMD's balance sheets or with the Graphics side.

The almost 2% growth compared to loss for ATI could be attributable to slight sales dampening of ATI's high end chips due to impending Fermi launch, coupled with lack of high end Nvidia chips for Q4 with a few weeks of high end sales in Q1.

In other words, any potential customers holding off on sales for ATI chips pending launch of Fermi would only show as a loss of sales. Potential sales gains after Fermi performance reveal and impact of low retail quantities wouldn't show up until next quarter. Quarter ending Mar. 27th and Fermi reveal Mar. 26th. So any positive impact will show up the quarter after.

Nvidia on the other hand would benefit from moving from lack of high end sales for Q4 and 2/3rds of Q1 to having a slight but positive revenue influx from whatever quantities of Fermi managed to ship.

Year over Year shows the larger overall trend of ATI regaining some ground on Nvidia. However Nvidia still dominates the discrete graphics segment it would appear with over twice the revenue and almost three times the profit of AMD's graphics division (~137 million USD versus ~47 million USD).

One final note, it's good to see both companies posting positive YoY growth as well as positive cash flow. With only 2 major graphics chip companies, it would be a very sad day if one were to go down.

Regards,
SB

 

[thop]

NVIDIA Q1 2011 Earnings Call Transcript

Reception for the new GeForce 400 series amongst gamers and press has been phenomenal.

In the first quarter, we shipped a few hundred thousand units as yields have exceeded our expectations. The 480 and 470 are widely available worldwide and continue to sell out on a weekly basis.

And so I think that with respect to manufacturability of 40 nanometers, I think we're surely out of the woods. I think we can now officially say that. We are clearly out of the woods, and I think that's all David was trying to say. The 40 nanometers at TSMC is now a world-class node and the manufacturability of it is very, very good.

Because we were late with Fermi, we were out of this marketplace for some six months to maybe eight months. And so there's a very large pent-up demand for high-end products out there.

 

[dizietsma]

NVIDIA Q1 2011 Earnings Call Transcript

God, they spouted a lot of poppycock there....

 

[Arun]

Unsurprisingly, they are down in after hours trading despite slightly beating analyst expectations: their guidance for Q2 is pretty disappointing. Previously, they claimed they'd be capacity constrained for quite some time which would mean seasonality was not very relevant and they'd grow throughout the year. But suddenly when capacity improves ahead of expectations (good bit thanks to yields), that upside headroom has gone away. How convenient! I don't think they gave a wrong guidance on purpose though, and their explanations don't satisfy me. I heavily suspect there was some double ordering and much of that demand was never really there. Oops.

BTW, a few *very* interesting tidbits I haven't many people mention:
- Confirmed that the main Tegra2 smartphone design wins are for Android 3.0, starting in Q4 2010. There have barely been any rumors on 3.0 yet, so that's something.
- GF100 was back-end limited, not wafer/yield limited. Due to complexity, testing/packaging/assembling can take "many weeks".
- Currently sampling Fermi down to entry-level notebooks, i.e. GF108. Nicely dodged that completely when an analyst asked for details later.
- 70 Optimus design wins, only 11 in production so far. That's in addition to the MCP89 IGP (GeForce 320M) obviously, as well as any AMD-based notebook design wins.