If geeks love it, we’re on it

AMD unleashes the Cayman-based FirePro V5900 and V7900

AMD unleashes the Cayman-based FirePro V5900 and V7900

AMD Firepro V5900 and V7900

AMD FirePro V5900

The new AMD FirePro V5900

Today AMD unveils two new workstation GPUs aimed squarely at the mid- and high-end DCC and workstation market: the FirePro V5900 and V7900. The ATI moniker is fully dead, as these are branded AMD parts.

Not that it matters much—the entire ATI team responsible for previous FirePro cards is still in place, and they’ve been hard at work on these computational monsters.

The AMD FirePro V7900

The new AMD FirePro V7900

The two new GPUs are based on the Cayman GL architecture. The V5900 has 512 stream processors while the V7900 has 1280 stream processors. Each card has 2gb of GDDR5 memory with a 256-bit memory interface. The V5900 can push 1200 MTriangles/s and the V7900 an impressive 1450 MTriangles/s.

Here’s where things get cool, though: Each card is a single slot solution, and each supports Eyefinity. The V5900 sips power daintily at a mere 75W and the V7900 tops out at 150W. This is possible because of some frankly incredible advancements in power technology which we’ll get to in a moment. The V5900 has two DisplayPort 1.2 ports and 1 dual-link DVI and can run three displays with Eyefinity, while the V7900 has four DisplayPort 1.2 ports and can push four displays. This generation of FirePro are the first workstation GPUs to support the new DisplayPort 1.2 spec, as well.

The list price of the V5900 is $599 and the V7900 comes in at $999.

The V5900 replaces the V5800 as their mid-range workstation GPU and the V7900 replaces the V7800. Here’s how the 2011 FirePro lineup stacks up:

AMD FirePro 2011 lineup

New Tech

The V5900 and V7900 represent a big leap for FirePro in terms of raw performance, but more exciting is the flexibility and productivity enhancements made possible by the low-power design and Eyefinity support. Over the next few weeks we’ll be going in-depth into dissecting these exciting technologies, but for now, let’s put a name to them: First up we have PowerTune, which at first seems mundane, but once you find out exactly how they developed this power technology and what it makes possible, you’ll change your opinion just as quickly as I did. Second, we have a new anti-aliasing mode called EQAA or Enhanced Quality AntiAliasing. Third, we have a technology that basically revolutionizes the way workstation GPUs process data called GeometryBoost. On a side note, GeometryBoost was literally named by Icrontic. Bobby “UPSLynx” Miller and Eli “Pragtastic” Robbins were discussing it one night over beers and they came up with the name. Since Bobby also happens to be the Product Marketing Manager for FirePro graphics, well… boom, headshot.

Geometry Boost can be summarized thusly, in a RealTalk™ quote from FirePro Product Marketing Manager Bobby Miller. The exchange went something like this:

Brian: Bobby, give me a one-sentence summary of GeometryBoost.

Bobby: Do you want the official AMD definition written by yours truly? Short version: GeometryBoost technology provides extremely fast geometry performance and ensures smooth handling of complex models.

Brian: No, I want the RealTalk™ version

Bobby:  How about this: GeometryBoost employs dual graphics engines which process two primitives per clock cycle, allowing much faster processing of complex geometry.

Let’s get that further simplified: According to Miller, GeometryBoost technology means smoother handling of large models and geometry thanks to the graphics card being able to draw two geometric objects per clock cycle. Two is like, twice as many as one, which is what other cards can draw per clock cycle.

GeometryBoost is revolutionary, no doubt, but there are still more exciting developments in FirePro world.

Next up we have EQAA, which is a new form of anti-aliasing. This doesn’t have a great deal of uses for DCC professionals, but for digital signage and other display-sensitive tasks, EQAA promises to improve anti-aliasing performance and aesthetics with a lower impact than other forms of AA. From the marketing materials:

AMD FirePro EQAA

EQAA is a new anti-aliasing option available on the latest series of AMD FirePro™professional graphics cards.This technique offers advanced smoothing of aliased edges without requiring additional video
memory, and with a minimal performance cost. It offers enhanced quality over standard Multi-Sample Anti-Aliasing (MSAA) modes by doubling the number of coverage samples per pixel, while keeping the same number of color/depth/stencil samples.

Cool, cool.

Now we get to my personal favorite, PowerTune. Friends, I’m not exaggerating when I say that at first glance, PowerTune seems to be alien technology. Wait, let’s go back a bit. At first glance, PowerTune seems very boring. “Oh,” they say, “What’s the big deal? The card gets hot and it clocks down. Yawn.”. They say that, I’ve seen it. Hell, I said it. Then Eric Demers, who is the CTO of  Graphics for AMD, got up in front of a room full of journalists and wowed us with what’s really going on here.

PowerTune is not analog technology. That is to say, it doesn’t employ something as simple as a thermometer and a signal that says “Hey, it’s hot, turn that thing down.” That’s amateur hour, and it’s a very basic technology that gives extremely coarse control over temperatures to keep the card within TDP (Thermal Design Power. Here’s an awesome primer on TDP if you don’t know what it means).

PowerTune changes the entire game. These GPUs were put into a laboratory and in a process which I can only describe as utterly thrilling, were thermally mapped out in a variety of real-world usage scenarios. Scripted tests like “What parts of the GPU die get hotter during spinning a model in the viewport on 3DS Max?” and “What happens while we’re rendering MPEG-4 video?” and “Where’s the heat during OpenCL computations?” were run over and over again. The data was collected and parsed, and a digital map of the heat signature of the GPU was created. Next, they built logic into a microcontroller that recognizes what the GPU is currently doing, and what it’s likely about to do, and is able to prepare in advance for what the heat output is going to look like in the future. This way, it can intelligently clock down various components in real time in order to keep the card within it’s already absurdly low TDP of 75W (V5900) or 150W (V7900). This is precisely why these cards are able to be single-slot models. In fact, the V5900 doesn’t even require an external PSU cable. It’s freaking bus powered!

PowerTune is so awesome, I’m going to write an in-depth article about it soon. For now, suffice to say that it is something revolutionary, that the competition does not possess, and is a significant factor in the vast improvements that these cards represent over previous iterations.

Productivity Enhancement

These cards will essentially enable an entirely new workflow for certain people. Eyefinity is extremely awesome for consumers and gamers, sure… but the workstation market is who is really going to benefit from this technology.

Once you see a workflow with three or four monitors, you cannot unsee it. You cannot imagine going back to a single- or even dual-monitor setup. Content creation is done in suites now—from Adobe CS to a Catia workflow, professionals use a multitude of apps at once in the content creation process. Having each app in its own display makes everything flow much faster and more efficiently, which translates into time saved and thus, money saved. You can get multi-monitor workflows with NVIDIA as well, but… well,

AMD FirePro Eyefinity cost advantage

Yeah, it kind of speaks for itself...

The benchmarks

Because of the unique nature and usage scenarios of these workstation GPUs, we are not going to break down dozens of different benchmarks and give you Crysis numbers and all of that—it’s ultimately pointless, and people who are the target market for this class of hardware don’t really need to know how they perform in 3DMark at any rate. Our test system was an AMD Phenom X4 965 at 3.4ghz and 4gb of RAM on Windows 7 64-bit. We’ll do a follow-up article with the charts and graphs that the more pedantic among you expect, along with some interesting comparisons to other products, but in the meantime, I will summarize it with this: In SpecViewperf 11, the V7900 is about neck-and-neck with the $4000 NVIDIA Quadro 6000, and in some tests exceeded the legendary Q6000. This is a bull in a china shop, this card. If you’re interested in Linux benchmarks for the V5900 and V7900, our friends at Phoronix have you covered in that area. If you are interested in synthetic benchmark numbers, PC Perspective has them.

The Icrontic Golden FedoraFor now, considering the performance, new technologies, incredible advancements in power usage, and very compelling price, I have no doubts at all about presenting AMD our coveted Golden Fedora award for this groundbreaking release. This generation of cards is a very, very serious game changer for a great many people in a lot of different industries, and if you are at all serious about DCC, signage, science, imaging, oil & gas exploration, and other workstation GPU tasks, you absolutely must take a long and hard look at these cards, because they may vastly improve your work.

Disclaimer: AMD flew me out to Sunnyvale to show me these cards, bought dinner for me, and I’ll be honest, I had some drinks and lulz with Bobby Miller. They also sent us a V5900 and V7900 for review. Of course, you know this stuff doesn’t color our reviews. This shit is legit. Alien technology, seriously.

Comments

  1. pragtastic
    pragtastic The original term was "geohancement" I think. Either way, I expect AMD to be calling with a job offer any moment now.
  2. BuddyJ
    BuddyJ I approve of the RealTalkâ„¢ review. Thanks for talking about the real world benefits of the card and not just puking up bar graphs. Bar graphs have their place, but they don't showcase just how cool these new FirePros are.
  3. Thrax
    Thrax I wish more people understood the value of AMD PowerTune. It's not like other power management technologies that simply cut your clockspeed in half when the GPU gets too hot. In fact, it's a power management technology that's actually designed to <i>raise</i> the average level of performance.

    When you go to design a GPU, you have a specific thermal power draw you want to target. Let's say it's 300W. All the components on the board have to be orchestrated in just such a way that the board doesn't draw more than that in most cases, but applications like Furmark and OCCT are an engineer's worst nightmare: they easily push any GPU without powertune past their envelope.

    They push things past their envelope because they do something that no game would ever do: run the clockspeed up to maximum, then run every single shader/SIMD at maximum capacity. It's completely unrealistic, but people use these tools, and you have to deal with them.

    So, in the past, you really had two options:
    1) Design the card with clockspeeds so low that apps like OCCT and Furmark don't push the GPU past its envelope, or...
    2) Overengineer the shit out of the card so it can survive prolonged operation well past the targeted envelope.

    #1 robs the consumer of performance in games, and #2 would make graphics cards incredibly expensive. So you have to strike a balance between under-specing and over-engineering, and hope that the public doesn't design an app that rakes your gamble over the coals.

    In sweeps PowerTune, which intelligently locks the card at its targeted envelope. What happens when you can now guarantee--with very little additional hardware engineering--that your card is safely protected from long stints beyond spec?

    You can raise the maximum clockspeed! This means everyday games and applications can benefit from a <i>much</i> higher maximum clockspeed. Does that performance come at the expense of soul-crushing apps like Furmark and OCCT? Yes, it does. But if you bought a GPU to run Furmark all day, maybe you need to reconsider your priorities.

    Let me give you a real scenario:
    The AMD Radeon HD 6990 shipped with an 830MHz clockspeed. If AMD didn't have PowerTune to protect the card from those outlier apps, then we would have had to ship the card in the low 600MHz range to make sure that the thermal output of the card at full blast didn't damage the hardware.

    Now look at the GTX 590, which still uses the "hurrdurr board hot cut clockspeeds" method. What frequency did it ship at? 607MHz.

    PowerTune was directly responsible for giving more performance to the gamer in the 6900 Series, and now AMD's FirePro lineup can do the same for DCC users.

    It's awesome.
  4. primesuspect
    primesuspect Thank you for posting that. RealTalkâ„¢
  5. RyanFodder
    RyanFodder
    pragtastic wrote:
    The original term was "geohancement" I think. Either way, I expect AMD to be calling with a job offer any moment now.


    I seem to remember this conversation.
  6. UPSLynx
    UPSLynx
    I seem to remember this conversation.


    haha, yep. Discussed over beers and a game of Dominion in the ICHQ garage.

    The flow was something like this:

    Geohancement>GeoBoost>GeometryBoost.
  7. BuddyJ
  8. Thrax
    Thrax That shit's super effective, son.
  9. lezlow
    lezlow anything from amd has got to be a plus
  10. lezlow
  11. Loklo Media Do you know if there will be a 4900? I'm torn between the benefits, features, power, performance of the AMD cards, vs. Cuda for 3DS Max... arrgh
  12. Dave @Brian (primesuspect) - First before you cast stones, maybe you should look in the mirror, kiddo. See, the deal with multi-page reviews is that we actually provide useful data, instead of random miscellany and snarky drivel-fluff. But before you go that route, you actually have to register on the radar of major manufacturers so you can actually get product to test. Wait, let me snap a few slides from a PDF deck that I didn't create... that's content and I can take shots at the guys with real data in the process. Yeah, I'm cool. All good. :)
  13. primesuspect
    primesuspect LOL Y U MAD THO

    ic4IxW.jpg

    Thanks for the feedback, Dave!
  14. csimon
    csimon I could really put (even) the V5900 to good use. We're currently running quadro fx4880's, and for what we do it just isn't enough.

    Did you say you'd run SPEC benches? Or have they already been done elsewhere?

    If this card did game worth a hoot I would use it at home as well. :bigggrin:
  15. Thrax
    Thrax Benchmarks forthcoming.
  16. Bandrik
  17. _k
    _k It might be the fact that you said their website looked old and junk as you linked to them. :tim:
  18. UPSLynx
    UPSLynx bwahahahahaha!
  19. primesuspect
    primesuspect Yeah, it's all good. We hugged it out via email. Friends again.
  20. photodude
    photodude One V7900 vs two Quadro 4000 cards sounds nice....now if Adobe will just rewrite the mercury playback engine to run using OpenCL rather then living on the island of CUDA forever (that or someone needs to write a CUDA emulator for OpenCL video cards...I heard there was one in development but there was legal issues that still have not been resolved)
  21. primesuspect
    primesuspect AMD told us that Adobe is working on cranking out OpenCL support for MPE. It's not hard to figure out that we'll probably see OpenCL in the next version of CS (I don't anticipate it as a patch to CS 5.5, but who knows? That would be awesome +)
  22. photodude
    photodude sweet....I see head to head beanchmarks for MPE with Quadro vs FirePro once Adobe releases MPE with OpenCL support.

    I question whether Adobe will drop CUDA support for MPE once they release OpenCL support, since Quadro Cards also support OpenCL; or if they will give people with CUDA supported cards the option to choose to between CUDA and OpenCL.
  23. UPSLynx
    UPSLynx Adobe is pretty excited about OpenCL and are working with AMD to implement. It will certainly happen sooner than later.

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!