As part of this evening’s AMD Capsaicin event (more on that later), AMD’s Chief Architect and SVP of the Radeon Technologies Group has announced a new Radeon Pro card unlike anything else. Dubbed the Radeon Pro Solid State Graphics (SSG), this card includes M.2 slots for adding NAND SSDs, with the goal of vastly increasing the amount of local storage available to the video card.

Details are a bit thin and I’ll update this later this evening, but in short the card utilizes a Polaris 10 Fiji GPU and includes 2 PCIe 3.0 M.2 slots for adding flash drives to the card. These slots are then attached to the GPU (it’s unclear if there’s a PCIe switch involved or if it’s wired directly), which the GPU can then use as an additional tier of storage. I’m told that the card can fit at least 1TB of NAND – likely limited by M.2 MLC SSD capacities – which massively increases the amount of local storage available on the card.

As AMD explains it, the purpose of going this route is to offer another solution to the workset size limitations of current professional graphics cards. Even AMD’s largest card currently tops out at 32GB, and while this is a fair amount, there are workloads that can use more. This is particular the case for workloads with massive datasets (oil & gas), or as AMD demonstrated, scrubbing through an 8K video file.

Current cards can spill over to system memory, and while the PCIe bus is fast, it’s still much slower than local memory, plus it is subject to the latency of the relatively long trip and waiting on the CPU to address requests. Local NAND storage, by comparison, offers much faster round trips, though on paper the bandwidth isn’t as good, so I’m curious to see just how it compares to the real world datasets that spill over to system memory.  Meanwhile actual memory management/usage/tiering is handled by a combination of the drivers and developer software, so developers will need to code specifically for it as things stand.

For the moment, AMD is treating the Radeon Pro SSG as a beta product, and will be selling developer kits for it directly., with full availability set for 2017. For now developers need to apply for a kit from AMD, and I’m told the first kits are available immediately. Interested developers will need to have saved up their pennies though: a dev kit will set you back $9,999.

Update:

Now that AMD’s presentation is over, we have a bit more information on the Radeon Pro SSG and how it works.

In terms of hardware, the Fiji based card is outfit with a PCIe bridge chip – the same PEX8747 bridge chip used on the Radeon Pro Duo, I’m told – with the bridge connecting the two PCIe x4 M.2 slots to the GPU, and allowing both cards to share the PCIe system connection. Architecturally the prototype card is essentially a PCIe SSD adapter and a video card on a single board, with no special connectivity in use beyond what the PCIe bridge chip provides.

The SSDs themselves are a pair of 512GB Samsung 950 Pros, which are about the fastest thing available on the market today. These SSDs are operating in RAID-0 (striped) mode to provide the maximum amount of bandwidth. Meanwhile it turns out that due to how the card is configured, the OS actually sees the SSD RAID-0 array as well, at least for the prototype design.

To use the SSDs, applications need to be programmed using AMD’s APIs to recognize the existence of the local storage and that it is “special,” being on the same board as the GPU itself. Ultimately the trick for application developers is directly streaming resources from  the SSDs treating it as a level of cache between the DRAM and system storage. The use of NAND in this manner does not fit into the traditional memory hierarchy very well, as while the SSDs are fast, on paper accessing system memory is faster still. But it should be faster than accessing system storage, even if it’s PCIe SSD storage elsewhere on the system. Similarly, don’t expect to see frame buffers spilling over to NAND any time soon. This is about getting large, mostly static resources closer to the GPU for more efficient resource streaming.

To showcase the potential benefits of this solution, AMD had an 8K video scrubbing demonstration going, comparing performance between using a source file on the SSG’s local SSDs, and using a source file on the system SSD (also a 950 Pro).

The performance differential was actually more than I expected; reading a file from the SSG SSD array was over 4GB/sec, while reading that same file from the system SSD was only averaging under 900MB/sec, which is lower than what we know 950 Pro can do in sequential reads. After putting some thought into it, I think AMD has hit upon the fact that most M.2 slots on motherboards are routed through the system chipset rather than being directly attached to the CPU. This not only adds another hop of latency, but it means crossing the relatively narrow DMI 3.0 (~PCIe 3.0 x4) link that is shared with everything else attached to the chipset.

Though by and large this is all at the proof of concept stage. The prototype, though impressive in some ways in its own right, is really just a means to get developers thinking about the idea and writing their applications to be aware of the local storage. And this includes not just what content to put on the SSG's SSDs, but also how to best exploit the non-volatile nature of its storage, and how to avoid unnecessary thrashing of the SSDs and burning valuable program/erase cycles. The SSG serves an interesting niche, albeit a limited one: scenarios where you have a large dataset and you are somewhat sensitive to latency and want to stay off of the PCIe bus, but don't need more than 4-5GB/sec of read bandwidth. So it'll be worth keeping an eye on this to see what developers can do with it.

In any case, while AMD is selling dev kits now, expect some significant changes by the time we see the retail hardware in 2017. Given the timeframe I expect we’ll be looking at much more powerful Vega cards, where the overall GPU performance will be much greater, and the difference in performance between memory/storage tiers is even more pronounced.

Source: AMD

POST A COMMENT

120 Comments

View All Comments

  • ddriver - Tuesday, July 26, 2016 - link

    So then how is it "incredibly innovative" if it doesn't seem to have any practical applications? Because it is new? In this regard putting just about anything on a GPU would be "incredibly innovative"... Let's put a bar code printer on the GPU so it can print diagnostics on paper when something goes wrong... or something like that LOL.

    To me this move from AMD doesn't spell out "innovation" but rather "desperation". Sadly, I think they wouldn't be doing stuff like that if they were confident in their future products. At this point it doesn't look like this is about anything more than some hype and making a few thousand dollars on those wonderfully priced developer kits, because somehow a 250$ GPU with added M2 slot more than justifies a 10k purchase, whose purpose would likely be to prove SSD on GPU is pointless, but hey at least it will generate more heat.
    Reply
  • Spunjji - Tuesday, July 26, 2016 - link

    Your reasoning doesn't make sense. A company experimenting with new products is not "desperate" unless they absolutely have to make them sell. This is a beta so it's not like they have bet the farm on it. Their customers will decide whether or not it makes sense for them. Reply
  • ddriver - Tuesday, July 26, 2016 - link

    It is not just a company, it is a company that struggles to compete. AMD have a long history of trying to make up for their hardware in various gimmicks, ranging from barely useful to entirely pointless.

    This particular "product" - aside from being applicable in a very, very narrow market niche, will also over very, very mediocre performance advantage, hopefully enough to be worthy the price premium and enough to pay for the RD it took. Knowing AMD, it will likely be yet another loss on a very tall mountain of losses.

    The very fact AMD are not showing any "charts" and making claims of improvement figures goes to show they literally have nothing, because if they did, they'd be more than willing to use it.

    Lastly - fast M2 SSDs are known at being very good and fast at hitting throttling temperatures where performance degrades significantly, plus imagine the effect on the GPU that already reaches 90 degree C would have when you slap two 90 degree C SSDs on the back of the PCB - that means you immediately lose a lot of performance, as the GPU will throttle EVEN MORE than it usually does, likely diminishing any tiny benefits you may get from putting SSDs on board.

    Finally - don't take my word for it, just look at AMD's track record, those guys - very good at shooting themselves in the foot, even when they have decent architecture as a foundation. How much confidence can one have in a company that struggles to stay in business and hasn't really made money in like forever?
    Reply
  • ddriver - Tuesday, July 26, 2016 - link

    I'd go for a neat marketing moniker for it - the Radeon Pro SSG with MAT technology, that is Mutually Assured Throttling, where the GPU and the SSD work in a delicate and mutually beneficial tandem - the GPU ensures the SSDs throttle sooner, the SSDs ensure the GPU throttles sooner and further. Plus I just bet being strapped onto a literal hotplate will do wonders for the reliability of the SSDs.

    But let's wait and see how AMD will capitalize on this pointless "yet-another-amazing-idea", they can't make money with good products that make sense, perhaps they will score big on stuff that doesn't make sense.
    Reply
  • MLSCrow - Tuesday, July 26, 2016 - link

    I now regret having responded to a previous post of yours. It's clear now that you simply do not like AMD. Whether or not you lost money in the past, having invested in them, I don't know, but again, your negative comments are overly excessive. It doesn't matter what AMD has done in the past, the people running the company now are not the same people that were running it before. This is a new company and what they are doing and have been doing is nothing short of awesome. I welcome Polaris with open arms and a thinner wallet, I welcome Zen equally. AMD is indeed an innovative company, that may have invented some things that didn't pan out well, but for the most part, they've invented things that have changed the world of computing forever. The first and still used 64-bit instruction set, which Intel had to strait up copy. The first multi-core CPU's, the first APU's, great GPU's over the history of their acquisition of ATi, Mantle, which Vulcan and DX12 may not have existed without, HSA, which is the right way to evolve processing. Compute performance, etc.

    In the past, AMD had incredibly poor management. Hector Ruiz almost saw to the destruction of the entire company. Predecessors weren't able to restore the strength of the company until recently and decision to take a chance on a new and different type of CPU architecture backfired (a great attempt at something different, which could have set them apart, perhaps if it wasn't rushed, over-hyped, and designed better out the gate, but you can't really hold it against them for trying), but with the leadership of Lisa Su as well as the decision to bring back arguably one of the greatest CPU architects of all time, giving him free reign to hire his own team and the freedom to design from the ground up, along with providing Raja with RTG to keep him around, are all paying off now.

    If not for AMD we wouldn't be where we are today or moving in the right direction like we are now toward superior technologies. Also, if not for AMD, we'd all be paying premium prices with unlimited ceilings due to monopolized markets, which their existence and products ensure we aren't. If anything, even haters of AMD should appreciate their existence and efforts simply in reducing prices of their more favored products. Whether you like em or hate em, it's good that they are around.
    Reply
  • ddriver - Tuesday, July 26, 2016 - link

    You are wrong - I'd like to see AMD up on its feet, not because I like it, but for the sake of competition. And does tend to offer better value for their products, but not because of noble aspirations, only because they are currently the underdog. But as good as their value might be, it doesn't help much when they can't compete in terms of absolute performance. And this includes power efficiency as well.

    Hopefully Zen won't be yet another flop, they desperately need to get back in the game, not that they did that much better back in the days of athlon, when intel could only offer garbage. But it is AMD after all, they've been making a lot of promises, they did waste RD budget on lots of uarch iterations, and they all sucked. It doesn't look like they have their priorities in tact. I have a theory there, that intel is secretly keeping amd "alive" just so that it looks less of a monopoly.

    It would be quite foolish for ANYONE to LIKE any CORPORATION, it is not that any of them are there to do any of us favors, they do what they do not for the sake of the product or the consumer, they do it for the sake of money. A lot of the "AMD" love is based on sympathy for an underdog, a lot of it is a product of AMD being forced in a more "giving" position, a lot of it is pure moronic fanboysm. But don't fool yourself that AMD is any better than intel, if they had the upper hand they'd be just as bad. It is the same with regular people - the rich can get away with anything, the poor NEED to be good because they don't have the money to afford to be bad.
    Reply
  • cocochanel - Tuesday, July 26, 2016 - link

    He doesn't hate AMD. He is just an Nvidia shareholder. There is a bunch of them patrolling the top tech websites out there causing so much misery. They just want to see AMD fold up, that's all there is to it. And they'll pull all the tricks out of the bag to do so. Reply
  • ddriver - Tuesday, July 26, 2016 - link

    I don't gamble, and that includes buying anyone's shares ;) Also, I have over 60 radeons under my roof running and crunching numbers 24/7, simply because they are the best bang for the buck for FP64 performance. In fact the only nvidia product I currently own is the GPU that came with a laptop as the only option. Reply
  • Samus - Tuesday, July 26, 2016 - link

    Your just now picking up on drivers anti-AMD trolling? He's been going at it awhile now, since at least Fuji, under the defense he likes to see competition while at the same time never being critical of nVidia and their anti competitive nature (dating back to 3Dfx) their price fixing, or their proprietary atmosphere (they like closed systems - Linux drivers being a rare exception) which caused them to lose consideration for any game console now and moving forward.

    Then there is the obvious fact AMD doesn't actually make bad products, their entry level gaming cards and their FireGL professional cards are all industry mainstays. Solid works in particular favors AMD GPU's over nVidia.
    Reply
  • ddriver - Tuesday, July 26, 2016 - link

    Yeah, let's hope I will someday get to reach the level of excellence of you fangirls and cheer at mediocrity. I have never criticized nvidia? Willing to bet on it?

    After a cursory search I was able to find those, there are probably plenty more:

    http://www.anandtech.com/show/7764/the-nvidia-gefo...

    LOL, epic? Crippling FP64 performance further from 1/24 to 1/32 - looks like yet another nvidia architecture I'll be skipping due to abysmal compute performance per $ ratio...

    http://www.anandtech.com/show/9516/nvidia-announce...

    Way to go, 8 times the difference is equal to almost no difference. I guess you almost won't care if your boss cuts your salary 8 times right?

    http://www.anandtech.com/show/7525/nvidia-33182-ga...

    I just wish they were not so botched ... And it is not like this is an isolated case, I've seen a ton of people complaining about the same issue to which nvidia remains indifferent. My next GPU will be a radeon...

    http://www.anandtech.com/show/7897/nvidia-announce...

    And now, a limited time offer, you can get TWO titans at the price of THREE!

    http://www.anandtech.com/show/8069/nvidia-releases...

    OpenCL + ATi gives you about 10 times better bang for the buck. Only people looking to waste money will go for nvidia compute solutions.

    http://www.anandtech.com/show/9779/nvidia-announce...

    It is too expensive even at educational pricing level. Cheap nvidia pushing for profit even on such a low volume market as dev platforms... way to go...
    Also, I don't see OpenCL mentioned anywhere, so thanks but no thanks!

    http://www.anandtech.com/show/9681/nvidia-announce...

    Once again, disappointing FP64 performance. FP64 is very important for workstation workloads.
    Sadly, in order to compete AMD is adopting the same strategy.

    Naturally, nvidia's product value is so low that it's been long since I even considered buying it, and thus logically, I care very little about what nvidia does. Do you bother criticizing stuff you don't care about? You are only confusing me having standards with trolling because you are a clueless wannabe who pretends to be competent while failing hard at it, a typical victim of consumerism.
    Reply

Log in

Don't have an account? Sign up now