Second Generation Maximus

NVIDIA’s second announcement of the day directly ties into the Quadro announcement, which is the announcement of NVIDIA’s second generation Maximus technology. Strictly speaking this isn’t a new technology in and of itself – it’s the same concept of using a Quadro card and a Tesla card together in a single system and is based on the same software – but for marketing purposes NVIDIA is going to be separating Maximus based on the generation of the video cards involved.

From here on NVIDIA’s first generation Maximus technology will refer to pairing together Fermi cards, e.g. Quadro 6000 paired with a Tesla 2075. Second generation Maximus on the other hand will refer to pairing together Kepler cards, specially a Quadro K5000 with a Tesla K20.

The significance of this announcement isn’t in the branding itself, but rather what it represents for NVIDIA. Because NVIDIA is not launching a GK110 Quadro card (or at least not yet) Maximus’s importance in NVIDIA’s ecosystem has been greatly elevated virtually overnight. For the Fermi generation all of NVIDIA’s high-end Quadro cards were capable of both high performance rendering and high performance compute, with the Quadro 6000 in particular being a superset of both the Quadro and Tesla families by offering NVIDIA’s best graphics performance combined with their best compute performance. Maximus improved the experience by offloading compute onto a Tesla card – and thereby bypassing the performance penalty of frequently context switching – but Maximus wasn’t strictly necessary for high performance compute with Quadro.

Second generation Maximus on the other hand sees Maximus become essential to compute performance. Because of GK104’s limited compute performance, Quadro K5000 cannot achieve high compute performance on its own; at best it’s a great graphics product and a decent single-precision compute product. Instead Quadro K5000 needs to be paired with a GK110 Tesla card in order to achieve both high graphics performance and high compute performance. Second generation Maximus in turn is the technology and branding glue that brings all of this together.

As we mentioned in our quick look at the Quadro K5000, NVIDIA going this route isn’t particularly surprising given the fact that GK104 is such a strong performer at graphics tasks and is far more available than GK110, but it means that NVIDIA is now very reliant on Maximus in all situations that call for a Quadro video card with strong compute performance. It also means that NVIDIA is at the mercy of application developers to a certain extent as developers need to specifically accommodate Maximus. To NVIDIA’s credit the number of programs supporting Maximus has more than tripled in the last year, but that’s still a far cry from the hundreds of applications professionals typically use.

For this reason we wouldn’t be shocked to eventually see a GK110 Quadro card, but until that time this is NVIDIA’s flagship graphics + compute solution.

Quadro K5000
POST A COMMENT

29 Comments

View All Comments

  • unidntifiedbones - Wednesday, August 08, 2012 - link

    Look again, Gee Bee R1, not I-16.

    Bit of a silly image really.
    Reply
  • HisDivineOrder - Wednesday, August 08, 2012 - link

    "Meanwhile Quadro K5000 also brings with it all of the major Fermi family features that we first saw with the GeForce GTX 680, including support for DisplayPort 1.2, 4 display controllers per GPU, PCI-Express 3.0 support, the NVENC H.264 video encoder, and even bindless textures."

    You said "major Fermi family features," but did you mean Kepler?
    Reply
  • Ryan Smith - Wednesday, August 08, 2012 - link

    Yes, I did. Thank you for that. Reply
  • shawkie - Wednesday, August 08, 2012 - link

    Any idea how they've managed to make PCI-Express GEN3 work with SNB-E on this board and not on the GTX 600 series? Have they managed to respin the GK104 already? Or have they used a bridge chip like on the GTX 690? Or is it just down to the track layout on the PCB?

    BTW, I'm aware that its possible to force the driver to run in GEN3 with SNB-E even with the GTX 600 series but in my experience this doesn't actually work properly (slow transfer speeds and system instability) so I can't imagine nVidia would suddenly claim it was supported unless they've actually changed something.
    Reply
  • Ryan Smith - Wednesday, August 08, 2012 - link

    While I don't have the full details, it doesn't look like NVIDIA has actually changed anything. Rather Kepler and SNB-E working correctly with PCIe 3.0 is dependent on some (unknown to us) hardware factor on the host machine. NVIDIA knows what that factor is, and will be qualifying workstations for PCIe 3.0.

    As it stands we know that PCIe 3.0 works perfectly fine on some SNB-E systems (namely: ours), so this isn't all that hard to believe.
    Reply
  • shawkie - Wednesday, August 08, 2012 - link

    So logically if a workstation is qualified by nVidia for PCIe 3.0 with the K5000 then we should expect that workstation to also work (in GEN3 mode) with the GTX 600 series...

    So far nVidia have repeatedly refused to publish any kind of list of qualified PCIe 3.0 motherboards claiming that results also vary by CPU (among other factors?).
    Reply
  • Ryan Smith - Wednesday, August 08, 2012 - link

    Correct, if it were to be qualified with K5000 then it should also work with a manually enabled GTX 600 card. Reply
  • pvrvideoman - Wednesday, August 08, 2012 - link

    Reading these articles on Kepler products makes me think that sticking to the Fermi products would be a better, less expensive choice. They are well established, and are valuable in many applications. Check out RenderStreamTV on Youtube. This isn't the only video where the power of the mainstream GPU is utilized in an app, but it's a nice one. The consumer and the manufacturer have such different goals. AMD seems to have the consumer in mind a bit more than NVidia. Reply
  • Rictorhell - Wednesday, August 08, 2012 - link

    What I am trying to find out is if there is the possibility of Nvidia releasing a newer, more powerful graphics card later this year or early next year, or if the GTX 690 is it, as far as consumer cards, for the near future?

    I haven't purchased a new graphics card for a few years, so I want the best, but most power efficient card that I can buy. The impression I have is that the 680 and the 690 are great cards, but that they don't really provide a "huge" power boost over the previous generation cards or the cards just prior to that generation.

    If they are planning to release something that is significantly faster then the 680, I would definitely be interested in that, if it's coming.
    Reply
  • Rictorhell - Wednesday, August 08, 2012 - link

    The last card that I purchased was the Geforce 8800 GTX. Is there anyone here that had that card or has that card, that can tell me, roughly, how the performance of that card would compare to the peformance of the GTX 680? Reply

Log in

Don't have an account? Sign up now