Handicapping AMD’s Fiji with David Kanter

AMD's upcoming Fiji GPU, rumored to be called the Radeon R9 390X, will aim to take the crown for fastest graphics card, aided by a brand-new memory technology known as HBM. David Kanter and I considered Fiji's prospects in some depth on our recent live stream. Here's the portion of our discussion related to Fiji.

The rest of the session is available here. Consider subscribing to our YouTube channel or following us on Twitch for more in the future.

Comments closed
    • NoOne ButMe
    • 8 years ago

    I disagree. I believe Fiji will be within the margin of error to plus about 10% in non-NVidia-blackbox games. Based off of the few supposed leaks of performance a while back.

    Now, that performance could be higher than Fiji will be, it could be lower. However, I think that in gaming, the difference shouldn’t be to large. It’s in bandwidth bound applications and compute that Fiji will be king.

    • NoOne ButMe
    • 8 years ago

    thanks!

    • moose17145
    • 8 years ago

    Not now, Ralph…

    • dkanter
    • 8 years ago

    The big change is in the PHY that is responsible for talking to the DRAM. That could easily be 20% of the die.

    • Knee Dragger
    • 8 years ago

    My cat’s breath smells like cat food.

    • Amgal
    • 8 years ago

    He said “better work”.

    • Soossi
    • 8 years ago

    Handicapping Fiji? At least trying so. Grey guy commented: “big disadvantage for HBM DRAM is the support of ‘only’ 4 Gigs of RAM (?, I’m happy with my 1Gb board) that might impact on whatever theoretical 4K rendering stuff and can’t match the competition (nvidia of course)”. The other guy was like: “Yeeeaa…”. And quickly the questioner changed topic. Another thing was the water cooling. What is bothering with that? I’d love to have 0 fan noise in my high end GPU board if I’d ever need one, and not one of those “puppy” fans from Fermi.

    Fiji will probably wipe the floor on everything nvidia has unless the test cases are cherry-picked in a biased way.

    • Redundant
    • 8 years ago

    Great a video. Getting so tired of reading articles

    • MadManOriginal
    • 8 years ago

    I was interested in reading an article. I have no interest in watching a video about a topic, and it’s a terrible way to be able to reference information later.

    • jihadjoe
    • 8 years ago

    D’oh!
    My bad.

    • NoOne ButMe
    • 8 years ago

    “David Kanter and I considered Fiji’s prospects in some depth on our recent live stream. Here’s the portion of our discussion related to Fiji.”

    Not sure how to help you…?

    If you mean the “standard” Techreport podcast, you’re also wrong. Did they make a special one for the Kanter? I don’t use podcasts personally.

    • jihadjoe
    • 8 years ago

    I’d like to have been told that this was an excerpt from the podcast, and not new content. I actually started the video and it wasn’t until a few minutes in that I realized I had already seen it.

    • esterhasz
    • 8 years ago

    Good points.

    I think whether the move is beneficial or not depends a lot on whether the company has to hire/train/manage new work force or can rely on existing structures and employees to shoulder the new business.

    AMD has let go so many people over the last years that the functioning of certain business units has come under heavy pressure. Downsizing is incredibly hard from a management standpoint. This is really what higher cashflow / same benefits can mean: keeping business units at functioning size and reducing the devastating effects of continuous shrinking. I’d therefore say it’s a good move, but AMD still needs to deliver a real smash product over the next 2-3 years to stabilize the company.

    • Anomymous Gerbil
    • 8 years ago

    It’s almost unwatchable at times /slight exaggeration.

    Stop interrupting David!

    • Anomymous Gerbil
    • 8 years ago

    Yep. People get so wrapped up in metrics like margins, that they can lose sight of the bigger picture. Not that margins (or any other metric) aren’t important, but you need to understand the context.

    • NoOne ButMe
    • 8 years ago

    I think going from GDDR5 to HBM would reduce it even further, but, Hawaii has a pretty small memory controller to start. I suppose. And, it is a massive bus increase.

    • NoOne ButMe
    • 8 years ago

    True, however, AMD does have a lot of room to make a decent air cooler, and, has shown to be willing to make water cooled cards.

    [[to clarify, I’m stating that AMD’s stock air coolers have all been cr*p. AIB way better coolers. They can merely use of those and be decent with cooling on Fiji (stock) I believe]]

    Look at how much better the “custom” coolers do for the 290(x) for example. Honestly, why didn’t they just put the 7990 cooler on Hawaii. Slightly overkill, but, AMD needed the overkill badly.

    So, I don’t think the power density will cause an issue with Fiji. Once we drop down to 14/16FF I think it will cause them quite a bit of trouble if they try to make a big card.

    Of course, from what I have heard, TSMC’s yields at over 200-250mm^2 at 16FF is terrible. So, that issue for both companies (certainly more an issue for AMD based on GCN 1/1.1/1.2/1.x against Kepler/Maxwell) could be pushed back further.

    added after OP:
    On a side note, given that the yield situation doesn’t improve on 14/16FF (possible, given, well… The largest chip I’ve seen Intel product is well under 150mm^2. And, they are the most advanced foundry I know of currently) that AMD will have an advantage in making a “large” GPU due to having more experience with MCM, interposer and a lot of other technologies. Of course, NVidia will have a lot more money to explore multiple solutions, or eat high costs in the short term.

    Unless AMD wins a lot of lotteries XD. Man. Imagine if AMD won like 50 ~100 million lotteries from 50 tickets. That would be hilarious “GAAP income is over 2 billion dollars” due to winning multiple lotteries and choosing the lump sum option. That would be amusing. I should probably stop typing now.

    • NoOne ButMe
    • 8 years ago

    AMD needs more cash flow, even if it’s at the cost of their margins in percentage points.

    15% of $200 is more than 20% of 100. THOSE NUMBERS ARE MADE UP FOR THE POINT OF EXAMPLE.

    Even if they end up with the same net profit due to lower margins, higher cashflow is still better than lowercash flow, at least, in AMD’s position. IMHO, ofc.

    • Melvar
    • 8 years ago

    Kneecapping AMD’s Fiji with David Kanter

    • chuckula
    • 8 years ago

    [quote<]Handicapping AMD's Fiji with David Kanter[/quote<] Oh, I'm sure Nvidia would just LOVE for you two to handicap Fiji! They need all the help they can get!

    • tviceman
    • 8 years ago

    “AMD is not going to do something to really spite themselves.”

    Hahahaha ARE YOU SURE ABOUT THAT ONE?

    • NeelyCam
    • 8 years ago

    Seconded, or thirded, or… tenthed? Totally agree/

    • WaltC
    • 8 years ago

    Kanter is really a great guy…;) Very personable and down to earth. I only ever had one “serious” argument with him a few years ago on his site–seems really silly now–and that is we were were “arguing” about how long AMD held Intel’s family jewels in its hand and squeezed on the CPU side of the house (Pentium vs. A64, etc.) I said it was two years…Kanter said it was more like one…Who knows…maybe they can do it again with Zen…

    Nice info from Tech Report…!

    • MathMan
    • 8 years ago

    One problem with that is that it’s bad for your gross margin. Nvidia used to sell the DRAM that came with their GPUs as well, and each conference call, they had to add a note that margins were lower than they could have been because of that.

    This obviously doesn’t make sense from a pure money making point of view, but this is how Wall Street values companies and a dwarf like AMD isn’t in the position to change that.

    • pranav0091
    • 8 years ago

    It takes even less time to turn off CAPS lock you know..

    • jjj
    • 8 years ago

    Depends how you define big and AMD was already pushing clocks too far so if they go from that to more realistic clocks that can be reached by all units ,the gap could be rather small.
    Besides,this would be an indirect benefit , ideally the big gain should come from gaining memory BW but remains to be seen how big that is.
    And it’s not certain that the chip can clock higher because of better temps to begin with , you bring the RAM much closer to the big die, plus (something mentioned in the vid) you shrink the PCB so cooling can become more difficult.

    • mesyn191
    • 8 years ago

    Good discussion but you gotta work on not interrupting eachother!!!

    • Pwnstar
    • 8 years ago

    I heard a rumor that redesigning the Integrated Memory Controller for HBM made it 20% smaller. That leaves more room for Stream Processors.

    • cmrcmk
    • 8 years ago

    One thing I think AMD gets out of HBM memory: being the vendor of the memory on the GPU. If i understand this tech correctly (though there’s a reasonable chance I don’t), it sounds like HIS, Sapphire, etc. will now be buying not just a Radeon GPU, but also it’s memory simultaneously from AMD instead of separately from Samsung or Hynix. Even if AMD is just buying Hynix RAM and integrating it, they would still get a small markup in the process.

    • VincentHanna
    • 8 years ago

    This video has 15 comments. 11 of them are on the topic of whether TR is being dishonest/bait and switch/clickbait with bad practices.

    That alone both answers the question, and tells me all I need to know about “why” TR is not changing their practices.

    • Generic
    • 8 years ago

    Tricky. Tricky. 😉

    For a while there, one could rely on the word “we” to indicate a video was behind the link.

    • NoOne ButMe
    • 8 years ago

    The big gain currently will be that it will use less power, allowing you to clock the cores higher in a set power envelope.

    Or increase perf/watt. Or perhaps even both in some cases.

    • NoOne ButMe
    • 8 years ago

    David Kanter currently has two website he runs/works for.

    Real World Tech (www.realworldtech.com) and Microprocessor report.

    RWT is generally diving deep into products, or, some analysis on what the future will bring. Or short analysis on new information. Examples are:
    Diving deep into CPU/GPU architecture (Bulldozer, Sandy Bridge, Fermi)
    Theorizing that Intel will use SomeFancyStuffIForgetNameOf at 10nm
    A one page article on adaptive clocking in AMD’s Steamroller APU.

    Hence, no place for talking about what Fiji could be. I would expect/hope for a short analysis of HBM on it once Fiji comes out from Kanter, however. 🙂

    ADDED AFTER OP:
    Microprocessor report does papers and such. IT’s really expensive. I wish I could afford it. The sample’s they have are amazing. grrrr

    • sweatshopking
    • 8 years ago

    Plus one for saying I had best work

    • derFunkenstein
    • 8 years ago

    Who said I didn’t want to to say that? I’ve agreed in the past it’d be best. But at the same time, it doesn’t take a genius; just a little observation.

    • Melvar
    • 8 years ago

    Yes. I look at TR every day, and I’ve learned to pick out these videos now. They really annoyed me at first before I learned to recognize them though. Not because I don’t want videos posted, but because I was expecting a written article and got something else. It really doesn’t bother me anymore, but it looks sloppy to have two main types of content and not distinguish between the two.

    Now you justify not wanting it to say “Video: ” in front of videos even though it’s blindingly obvious that some portion of TR’s readers want it. Maybe start with “it will make the site worse by adding an unreasonable workload and no non-corporate site could possibly keep up”.

    • derFunkenstein
    • 8 years ago

    It was “with David Kanter” and recently on the front page was…you know…the podcast with David Kanter.

    • TopHatKiller
    • 8 years ago

    Interesting and I suppose worthwhile as a discussion but… without any actual real facts is this not just adding to the ghastly rumourmill? [I so sick of the rumourmill, now.]
    Kanter is a knowledgeable fellow, there’s no doubt, but any real insights that he might have, based on his contacts in the industry – wouldn’t he be more inclined to put those on his site? Rather than having a nice chat here?
    I still applaud, without hesitation, nice chats.
    And still stipulate, by the by, that we should all wait for 14/16 products from amd and nv, ignore the current shenanigans -[desperately so if you want quad-hd resolution] and see what happens when both companies have access to the new nodes.

    • ImSpartacus
    • 8 years ago

    Yeah, I’m trying to think of a reason why they can’t just add a simple tag to these kinds of articles.

    • Melvar
    • 8 years ago

    You don’t care, and other people do. Nothing will be lost for you other than an opportunity to troll, and frankly you do better work when it’s a more contentious subject.

    Edit: Also, it would literally take less time for a single staff member to accommodate this request than it takes someone to click, see it’s a video, and click back, or close tab.

    • sweatshopking
    • 8 years ago

    OMG I CLICKED ON SOMETHING AND LOST TWO SECONDS ILL NEVER GET BACK

    • Melvar
    • 8 years ago

    This keeps coming up, and I agree. Just have an icon next to the videos, or do it like
    “Video: Handicapping AMD’s Fiji with David Kanter”.

    • ImSpartacus
    • 8 years ago

    I would’ve liked to see that this is a video.

    While I enjoyed the video, this feels like a bait & switch. Can we just have a “Video: ” at the beginning of each of these kinds of articles?

    • jjj
    • 8 years ago

    Keeping my expectations low, sure been waiting for years for interposers to become more relevant but not sure there will be a huge perf gain from the extra memory BW now on 28nm. Ofc will depend a lot on size but maybe this gen is more about server/workstation perf and power not about gaming. Next year on 16ff they should be able to take better advantage of the BW.
    As for size of the memory , isn’t part of the goal to be able to use less, at least in some segments where the target is not the ultimate perf. and they just need w/e is enough for the targeted perf.
    For now looking at this as a stop-gap product in gaming (and likely aimed at other segments), at least before we have more info.

Pin It on Pinterest

Share This

Share this post with your friends!