Nvidia’s Drive PX 2 prototype allegedly powered by Maxwell, not Pascal

[csbin]

DRIVE PX 2

gtx980m

 

[RaulF]

So I guess no HBM?

 

[ShintaiDK]

HBM may be flagship only, just like AMD.

GDDR5(X) will rule for a long time in GTX980/970/390/390X and down.

 

[3DVagabond]

Can it play Crysis?

 

[parvadomus]

That thing is 250watts TDP, and its 8TFlops SP (probably running highly binned gpus). So pascal will get around 4TFlops @ 125 watts? (less than 980TI performance @ 250watts), however it wont match 980TI peformance with that 128/256bit bus. I think its safe to say that Pascal will offer no more than 2X performance per watt over Maxwell.

 

[Krteq]

Total computing performance of PX 2 stated in slides is a sum of performance of all ARM cores and GPUs, not GPUs only

 

[RaulF]

HBM may be flagship only, just like AMD.

GDDR5(X) will rule for a long time in GTX980/970/390/390X and down.

That makes sense. :thumbsup:

 

[ShintaiDK]

That thing is 250watts TDP.

Where did you see TDP?

 

[Krteq]

 

[ShintaiDK]

Thanks

Its a unit wattage tho.

 

[Krteq]

Yeah, It's a total TDP per board as well as computing performance (All Cortex/Denver cores + GPUs).

 

[caswow]

Nice one nvidia Not even pascal gpus on the board. They fooled a Lot of you guys into thinking these are pascal gpus

 

[Piroko]

Disregarding the (potential) stage mockup, that is genuinely a lot of power to crunch video inputs and other sensory data, I can see this being very useful for its intended application. Only the watercooling and general complexity seems a bit cost prohibitive for mainstream application.

 

[Krteq]

Yes, they showed us only some numbers (What the hell are DL TOPS???) and board prototype fitted with GTX 980Ms and unidentified Tegra chips, but it's still better than sawed off PCB with wooden screws (aka "This puppy here is...")

No signs of Pascal based chips :/

 

[AnandThenMan]

Very odd looking board.

 

[Pariah]

If I was Nvidia, I wouldn't be using Pascal modules for public mockups either. The product hasn't been officially announced. Makes more sense to have internet trolls mock that it's a fake than try and draw conclusions of the actual product before it is revealed.

 

[pooptastic]

Looking at a water cooled solution for a car makes me think of countless cars i've seen (or had) with rusted out and non-functional heaters, and crappy broken radiators.

Wouldn't you want passive cooling on this, even if the heat sync weighed 30 lbs or something?

 

[MrTeal]

Looking at a water cooled solution for a car makes me think of countless cars i've seen (or had) with rusted out and non-functional heaters, and crappy broken radiators.

Wouldn't you want passive cooling on this, even if the heat sync weighed 30 lbs or something?

Do you know the lengths an engineer working on a small car would go to eliminate 30 lbs of curb weight from a vehicle? A low temperature closed loop cooling system isn't infallible, but it's a lot less harsh of an environment than a high temperature, high pressure mixed metal loop like you see in the primary cooling systems of many cars.

 

[pooptastic]

Do you know the lengths an engineer working on a small car would go to eliminate 30 lbs of curb weight from a vehicle?

Fair point i guess. Really with current safety standards 'little' cars suffer from tons of safety bloat, so i'd think 30-50lbs wouldn't even matter that much since it's less than 1% more weight.

Just strange to see an active cooling system is all. i'm used to seeing those ECM units which are passively cooled. Obviously this requires a ton more processing power than those did.

What i'm wondering now, is which AT forum member will re-locate theirs and put it in a plexiglass box in their SUV to show it off.

 

[RussianSensation]

If I was Nvidia, I wouldn't be using Pascal modules for public mockups either. The product hasn't been officially announced. Makes more sense to have internet trolls mock that it's a fake than try and draw conclusions of the actual product before it is revealed.

So now it's trolling to point out how a company announces a new product while using old product chips in the public announcement? Man, I didn't get the memo on how to do new product launches/business introductions. Guess for iPhone 7, Broadwell-E, New E-Class Benz, companies should show off last gen's products and claim them to be next generation so they can see reaction of "Internet Trolls".

If Intel launched Skylake in August 2015 with paper slides and then flaunted i7 4790K Skylake as the next gen's product, it would have been SO much more professional, way better than having real demos running in real time, powered by real next gen chips. Fake mock ups and power slides is next gen way now. /sarcasm

Also, amusing to see NV using water-cooling on a 250W TDP product but the competitor was mocked for months for "requiring" LC to function.

Good to see NV continues to push into other industry sectors though since the glory days of record volume into graphics card sales are well behind the graphics card industry.

 

[3DVagabond]

If I was Nvidia, I wouldn't be using Pascal modules for public mockups either. The product hasn't been officially announced. Makes more sense to have internet trolls mock that it's a fake than try and draw conclusions of the actual product before it is revealed.

So you think they actually have them with Pascal GPU's but they are just being stealthy? Really? I think you'd be the only one. The rest of us would think they didn't show what they don't have.

 

[3DVagabond]

Looking at a water cooled solution for a car makes me think of countless cars i've seen (or had) with rusted out and non-functional heaters, and crappy broken radiators.

Wouldn't you want passive cooling on this, even if the heat sync weighed 30 lbs or something?

It's really too bad that it needs water cooling.

 

[monstercameron]

Oh the sweet, sweet irony.
Paper launch...check
No products ready...check
Needs liquid cooling... Check.

 

[AtenRa]

I didnt know we needed that much GPU power for AI self driving cars.

 

[Pariah]

So you think they actually have them with Pascal GPU's but they are just being stealthy? Really? I think you'd be the only one. The rest of us would think they didn't show what they don't have.

Absolutely. The very first post here, someone already analyzing the design and noted no HBM. Don't reveal anything about the design until it is done and you can reveal benchmarks.