Are NVIDIA Tesla 20-series GPUs designed for gaming?

[maverick786us]

I just came across an aritcal on NVIDIAs Tesla 20-Series GPUs. I am quite impressed with it, as it looks like a powerhouse. Are these graphic cards designed for gaming or just ray tracing to 3D cloud computing to data analytics?

 

[vjcsmoke]

No. No they are not.

 

[evilsofa]

From the linked article: "look for these to run between $2,499 and $18,995 when they roll out sometime in the second quarter of 2010"

From the wikipedia article on Nvidia Tesla: "The lack of ability to output images to a display is the main difference between Tesla products and ordinary video cards."

Both of those issues effectively kill the idea of using them for gaming.

 

[Lorien]

I just came across an aritcal on NVIDIAs Tesla 20-Series GPUs. I am quite impressed with it, as it looks like a powerhouse.

There's nothing really impressive about these cards. If you read the press release they state:
- Double precision performance in the range of 520GFlops - 630 GFlops

The 3 month old Radeon HD5870 does 544GFlops.

 

[sirmonkey1985]

tesla cards are for industrial rendering farms.. they usually only come with 1 video port but have double or triple the video ram of a consumer gaming card.. most dont need high double precision performance, lorien.. there are very few programs that actually use it right now and even doing 3D renderings still use single percision which will be much higher then ATI's cards..

basicly telsa = nurfed consumer card for reliability and double the video ram..

 

[obobski]

Tesla for gaming....I'm so glad these threads took years to start, I remember back in the day "LOL QUADRO FX COSTS $3000 IT MUST BE KILLER FOR GAMING!!!!111oneoneoneone"

and I'll say here what I would say then:
this hardware isn't designed for gaming or consumer use, its for professional applications
the Tesla is great at what its intended to do, and really does make a difference, but for straight up gaming, GeForce is the nVidia product you want, if buying something today doesn't work, wait for the new Fermi based nVidia cards instead of wasting money on a pro card (which is going to use the same processor as GeForce has available today, so basically the highest end Quadro FX and Tesla, today, meaning GeForce GTX 280/285, not worth $2500+ imho, especially considering that most of what you're buying is a higher precision driver, commercially recognized branding, and so on)

 

[rincewind]

There are very few programs that actually use it right now and even doing 3D renderings still use single percision which will be much higher then ATI's cards..

HD5870 has 2.72 TFlops single-precision performance, assuming you can fully load each of the 320 shader clusters.

EDIT: Turns out that GT200 cards are incredibly weak at DP, coming in at 1/12 of SP performance.

 

[Lorien]

tesla cards are for industrial rendering farms.. they usually only come with 1 video port but have double or triple the video ram of a consumer gaming card..

It is obvious you don't fully know what you are talking about. First, Tesla cards come with NO video ports, the C20 products are the first to have one.

most dont need high double precision performance, lorien.. there are very few programs that actually use it right now

Second, double precision is what is used for the kind of GPGPU work these cards are designed and sold to do. The fact that a 3 month old card that wasn't designed specifically for GPGPU work has just as much computing power is embarrasing.

single percision which will be much higher then ATI's cards..

Third, please go read the white papers of Cypress and Fermi so you don't spew uninformed nonsense like that. Single precision numbers of Fermi based on the shader clocks of the tesla card(1.23ghz) are at 1.26 TeraFlops while ATI's are 2.72 TeraFlops. Tesla products are usually 10% slower than consumer Geforce products, 10% faster shader clocks will not magically give it 1.5 TeraFlops of performance.

So, basically there is nothing impressive about these cards other than the extra transistor budget for ECC and cache coherency for GPGPU programming work, but that does absolutely nothing for gaming.

wait for the new Fermi based nVidia cards instead of wasting money on a pro card

The current C10xx Tesla products are based on the GTX285. These Tesla C20xx cards are based on Fermi, which gives us a good indication of what to expect from the consumer Geforce product...besides if you bothered to read the press release they said that these cards will be available in Q2 of next year.

As for the Fermi based Geforce consumer products:
http://www.nvidia.com/object/io_1258360868914.html

Editors’ note: As previously announced, the first Fermi-based consumer (GeForce®) products are expected to be available first quarter 2010..

 

[rincewind]

Second, double precision is what is used for the kind of GPGPU work these cards are designed and sold to do

Exactly - I couldn't be bothered to point it out, but anyone who has done calculations with 32-bit floating-point numbers will be able to confirm that results obtained are often inaccurate. Double-precision isn't perfect either, but it's close enough for 95% of situations. The binary number system cannot represent decimal values accurately unless the integer and decimal portions are stored separately.

 

[obobski]

The current C10xx Tesla products are based on the GTX285. These Tesla C20xx cards are based on Fermi, which gives us a good indication of what to expect from the consumer Geforce product...besides if you bothered to read the press release they said that these cards will be available in Q2 of next year.

As for the Fermi based Geforce consumer products:
http://www.nvidia.com/object/io_1258360868914.html

someone apparently woke up on the wrong side of the bed this morning....

my point was, don't buy the pro cards, buy the GeForce cards (if you do nVidia), as the performance is in-line with gaming, and costs far less, wasn't trying to imply anything else in that statement

 

[trinibwoy]

Second, double precision is what is used for the kind of GPGPU work these cards are designed and sold to do. The fact that a 3 month old card that wasn't designed specifically for GPGPU work has just as much computing power is embarrasing.

You know you're really close to having a point. Except for the little part about theoretical performance numbers being meaningless. Even more so in HPC.

 

[maverick786us]

I apologize if this post created an argument among you guys. I was just curious when they associated it with super-computing, I was just curious, whether it will change the entire PC Gaming world if it was designed for gaming.

 

[obobski]

I apologize if this post created an argument among you guys. I was just curious when they associated it with super-computing, I was just curious, whether it will change the entire PC Gaming world if it was designed for gaming.

nah

because Tesla associated with supercomputing overall, and didn't really change the gaming world (look at the GF cards that its derived from), but did make large strides for supercomputing

as far as the change for supercomputing, yeah, thats huge, because you're upgrading from something like a Xeon or Opteron to a GeForce or Radeon, which is MASSIVELY faster (at certain applications), so its a big deal

but that tech started in the realm of rendering and gaming

in other words you're looking at the tech going the opposite direction