Upcoming Nvidia GPGPU brand

[INKster]

So they say in the real report:

Yay ! :D
http://www.beyond3d.com/content/news/165

Boo ! :P
http://www.thestreet.com/_dm/smallbusinesstech/smallbusinesstech/10347867.html

[Rys]

I'm wondering whether the product will be a rebranded Quadro board with a software and services value-add, or whether it's a completely new (set of) board design(s) with software and services layered on top.

I tend toward the latter, since back before G80 launched around September time last year (without checking), I got wind of a high number of PCBs being designed for G8x GPUs, more than I'd have expected for the desktop SKUs.

[Geo]

So they say in this three page report:

http://www.thestreet.com/_dm/smallbusinesstech/smallbusinesstech/10347867.html

Also on our front page, about three hours before this thread was posted. :cry:

[INKster]

Also on our front page, about three hours before this thread was posted. :cry:

Shouldn't be an obstacle for discussing the subject right here, i hope. :wink:

[Geo]

Of course not! But a tip of the cap wouldn't kill you, would it?. Rys has s/c/o/t/c/h/ cheese sandwiches to buy, y'know, and I still have hopes of an R600 to play with this summer.


*Alms! Alms for the poor! Alms!*

[Tim Murray]

I want an R600 to throw into PeakStream, but that's neither here nor there.

[Geo]

I want an R600 to throw into PeakStream, but that's neither here nor there.

Well, you get to be in line ahead of me for that very reason. :cry:

Geo needs a laptop this summer tho (current is 3 yrs old), so I don't see me buying a new high-end GPU (and, really, my 8800 GTX is liable to be a very adequate performer for the rest of 2007 from all indications).

[Jawed]

Of course not! But a tip of the cap wouldn't kill you, would it?.
Are they the same tips of the hat where front page news items result from members of the public making forum posts?

Jawed

[Geo]

Are they the same tips of the hat where front page news items result from members of the public making forum posts?

Jawed

I'm not sure what the policy is there, but certainly not the case in this particular situation.

[INKster]

Of course not! But a tip of the cap wouldn't kill you, would it?. Rys has s/c/o/t/c/h/ cheese sandwiches to buy, y'know, and I still have hopes of an R600 to play with this summer.


*Alms! Alms for the poor! Alms!*

Edited :D
I will change it again when the new brand and/or other details show up.

[geekcomputing]

im very curious to how Acceleware ties into all of this. Nvidia owns 15% of acceleware btw.

i wonder if acceleware is just simply a company that nvidia owns a little bit of and makes turn key solutions to gpgpu problems , while nvidia will do something similiar to a Gpgpu-Quadro and marketing directly to the programmers (and thus taking on a peakstream approach)

OR is acceleware deep in bed w/ nvidia helping write software (its possible since nvidia has been helping the company covertly for 2.5 years)

opinions anyone?

[Geo]

Creating a new brand is an interesting strategy. Ghu bless NVIDIA, they out-market ATI all day everyday, twice on Sunday. I would guess they don't see a whole lot of overlap with Quadro, but DO see a "value add" proposition to makeing this a separate brand, not dissimaliar from the Quadro brand in concept. . . .

[geekcomputing]

i just had an interesting thought.

we know that a future nvidia gpu will be double percision.

so i wonder if nvidia will make some its "gpgpu-quadro" card double percision and just have its other gpus and quadros normal.

hummmmm
uttar?

[Tim Murray]

i just had an interesting thought.

we know that a future nvidia gpu will be double percision.

so i wonder if nvidia will make some its "gpgpu-quadro" card double percision and just have its other gpus and quadros normal.

hummmmm
uttar?
Well, it's not like they're going to use GPGPU-specific chips. At least, not at this point--they would never make the money back. They might artificially limit FP64 performance on consumer-level chipsets, though, since it's not like any consumer-level apps will use it.

[Rys]

Are they the same tips of the hat where front page news items result from members of the public making forum posts?

Jawed
We've (hopefully) always tipped ours in the news post. See here (http://forum.beyond3d.com/showthread.php?t=38314) for example.

[jb]

im very curious to how Acceleware ties into all of this. Nvidia owns 15% of acceleware btw.

i wonder if acceleware is just simply a company that nvidia owns a little bit of and makes turn key solutions to gpgpu problems , while nvidia will do something similiar to a Gpgpu-Quadro and marketing directly to the programmers (and thus taking on a peakstream approach)

OR is acceleware deep in bed w/ nvidia helping write software (its possible since nvidia has been helping the company covertly for 2.5 years)

opinions anyone?


We are using their cards in a 3D EM Simulator for all of our products. They are simulating this data faster than our CPU clusters! However the price is very very high. They are taking NV cards and doing something to them... Not sure if its a BIOS and a Driver tick. But that in conjunction with EM Simulation tool (as they needed a special patch of their software in order to use these cards) seems to make our desingers very happy. Not making me very happy as I just dont see what they have to modify and why does it cose so much..we are talking about x15 times increase in price per card in a SLI config :(

[silent_guy]

But that in conjunction with EM Simulation tool (as they needed a special patch of their software in order to use these cards) seems to make our desingers very happy. Not making me very happy as I just dont see what they have to modify and why does it cose so much..we are talking about x15 times increase in price per card in a SLI config :(

I don't find that strange at all. The sales volume of their EM simulation tool is probably 2 orders of magnitude lower than that of an average game. The money to pay the programmers has to come from somewhere... And, hey, it makes your designers happy! :wink:

[jb]

I don't find that strange at all. The sales volume of their EM simulation tool is probably 2 orders of magnitude lower than that of an average game. The money to pay the programmers has to come from somewhere... And, hey, it makes your designers happy! :wink:


My frustration is that there is no reason why it should cost this much. We are talking about inflating the price a normal workstation which is putting this out of reach of most of our engineers. With mimimum effort and lower prices you could extand this solution to may other forms of ECAD work. But not at its current price point :(

[mhouston]

You are paying for their (original) development time and support. You *could* roll your own app atop Peakstream/RapidMind/Brook/CUDA/CTM, but it's going to take a lot of development effort. You basically need to decide if the hardware and software cost they are charging is less than your development cost and using cheaper consumer boards. If the speed up saves engineering time and lets you get your product out faster, then what they charge is likely worth it. Usually these companies have eval periods where you can decide if their product is worth it.

It's a similar model to the Quadro/FireGL boards in that it's the same hardware as the consumer stuff, but the boards go through more rigorous testing and extra driver development time is spent on various pro apps (CAD packages, modeling, off line rendering, etc). A single pixel error isn't an issue when gaming, but is a *huge* issue in the pro space. You are also paying for support, so you should make use of it and bitch and complain if things aren't exactly working for your app.

[silent_guy]

My frustration is that there is no reason why it should cost this much. We are talking about inflating the price a normal workstation which is putting this out of reach of most of our engineers. With mimimum effort and lower prices you could extand this solution to may other forms of ECAD work. But not at its current price point :(

Say you need 4 engineers full time to program your application. In the Bay Area, with benefits, that will cost you easily $120K per year per person, if not more (Google has played a major factor in raising the bar of entry.) Add a couple of marketing and sales guys, an admin, infrastructure, plane tickets and hotel rooms to go the customers and trade shows etc. Let's say $800K for the first year and $1.2M for the second.

Obviously, the first year you don't make a dime because you're still working on your product. And after that, sales pick up slowly. After the second year, you've sold 3000 licenses. 2M/3000 = $660 per license and just to break even.

In reality, you're going to need more than 4 programmers to create a kick-ass product, so real cost will be higher. And at the end of the day, the whole idea is to actually make money. It's really not surprising they are asking as much as they do. Just looking at the real cost of the hardware is wrong way to look at it.

[Geo]

Say you need 4 engineers full time to program your application. In the Bay Area, with benefits, that will cost you easily $120K per year per person, if not more (Google has played a major factor in raising the bar of entry.)

Yeah, fully loaded I'm thinking that's on the low side. Assuming they are experienced engineers, and for what we're talking about I don't think this is going to the be entry level jobs, then I'd say not less than $150k and maybe as high as $200k, with all the stuff (including office space).

[geekcomputing]

acceleware = 45 employees btw.

not sure on the breakdown between management, sales, and R+D but 45 mouths is still a lot.

no idea on what a canadian programmer in calgary makes either.