Page 257 of 372 FirstFirst ... 157207247255256257258259267307357 ... LastLast
Results 2,561 to 2,570 of 3718

Thread: Vega 10, Vega 11. What we know

  1. #2561
    Senior Member
    Join Date
    Jan 2010
    Posts
    3,487
    Quote Originally Posted by shagul View Post
    People are talking about it because this is what AMD chose to release. Also, I didn't see that AMD themselves confirmed that they were unable to produce gaming drivers for a card that is essentially a year late, but I'll take your word for it as none of this makes sense anyway. Maybe an extra month will result in 40-50% faster gaming performance, but in that case I think someone at AMD is taking this "fine wine" strategy to absurd levels.
    I find nothing odd in not wanting to package slightly unstable up-to-date drivers on a Pro card versus compatible-but-otherwise-legacy drivers that are stable.

    The fact that there are barely any differences to Fiji driver is very telling. It's not that AMD doesn't have RX gaming drivers, or that these RX gaming drivers don't use the new functionality. It is that AMD have chosen not to include them with the FE release for whatever reasons (likely to ensure the RX Vega launch looks good, and to ensure stability in the Pro drivers when you switch to game mode, which AMD have said is useful for game devs to switch to and from when developing).

    Because of this, you cannot determine anything but a baseline performance level, about the RX Vega performance from the FE Vega launch. RX could be +10% or +50%, we can't tell what the new features will provide.

    All of this is even more strange given that there is nothing fundamentally wrong with Polaris. I don't see why they couldn't have repeated the 290/390 strategy and just made a card with 100% more cores/bandwidth (keeping GDDR5). It might not have been optimal in some aspects but would at least have provided competition for Nvidia and given their driver developers more time to create drivers for the next generation.
    Yes, if Vega didn't meet expectations in simulation, then AMD would have done this, despite Polaris' clock scaling issues and other flaws that limit scalability.

    They haven't done this, despite knowing over a year ago, maybe up to two years ago, how Vega would perform in reality from simulations.

  2. #2562
    Senior Member
    Join Date
    Jan 2012
    Posts
    829
    Unfortunately amd has brought this on themselves, the product is released and is selling for 999$, it is in the same market as the Titan xp, therefore it can and should be judged to the same standards, which is a mix of compute and GAMING, titan is the top of the line nvidia gaming card with some compute functionality to appeal to the professionals.

    People are pretending the Titan is crap at gaming or its buyers don't use it to play games, nonsense that market is a gaming market.
    Also this shows the architecture in a bad light meaning anyone who has held back from buying a gtx 1080ti to see what Vega uarch offers for gaming will almost certainly just jump on the nvidia bandwagon.
    I get amd is seriously short on budget and has a lot of projects running concurrently but this is a mistake, it could be a small mistake if Rx Vega increases performance by 40% but is that really realistic? No, certainly not on launch, 1 year? More likely imo.

    IF amd has an uber driver they are cooking up that increases performance by 35-40℅ then it would be prudent of them to announce it, first impressions are paramount especially on a new architecture and it seems something has gone wrong here, either in design or marketing.

  3. #2563
    Senior Member
    Join Date
    Jan 2010
    Posts
    3,487
    No, AMD are pretty ************ clear that the FE is a Pro card. It comes with Pro drivers, all the marketing includes Pro apps, etc, etc.

    Just because Nvidia's similarly priced card is not solely a Pro card doesn't make AMD's card have to compete in gaming. How about comparing it with Nvidia's Pro cards which cost even more?

    I agree that AMD could do more to clarify the game mode driver on FE, and the future RX gaming driver's improvements, but they have clearly said it will be faster already.

  4. #2564
    Quote Originally Posted by JeeBee View Post
    Yes, if Vega didn't meet expectations in simulation, then AMD would have done this, despite Polaris' clock scaling issues and other flaws that limit scalability.

    They haven't done this, despite knowing over a year ago, maybe up to two years ago, how Vega would perform in reality from simulations.
    Nah, think about BD. Simple Thuban, directly shrunk to 32nm without any improvement would be a better chip. AMD clearly made tons of simulations, yet they failed to predict BD will fail that bad. Such stuff is always possible and Vega might have been in similar situation - looks great on paper, yet the thingy just doesn't work as it should on real device.

  5. #2565
    Senior Member
    Join Date
    Jan 2010
    Posts
    3,487
    Well let's wait for the gaming product release with gaming drivers.

    Yes, AMD can pull defeat from first place in a one-man race. It is a skill of theirs. But have they done it in this case (apart from their obvious marketing problems).

  6. #2566
    Quote Originally Posted by Zizy View Post
    Nah, think about BD. Simple Thuban, directly shrunk to 32nm without any improvement would be a better chip. AMD clearly made tons of simulations, yet they failed to predict BD will fail that bad. Such stuff is always possible and Vega might have been in similar situation - looks great on paper, yet the thingy just doesn't work as it should on real device.
    Well, I have found myself in similar situations a couple of times.
    When marketing and management decides to do (or not to) something, they overpower engineering quite easily and it doesn't matter whether you have data behind you or not, your points get rejected and you get fired if you persist. That's how the BD fiasco has happened.

    Hopefully, this is not the case with Vega. After all, it carries most significant refinement in GPU arch since the introduction of GCN and, if one remembers, Tahiti has had disastrous drivers for about a year since launch (before frame pacing and numerous game-specific optimizations), but in the end it turned out to be a great card.

  7. #2567
    Quote Originally Posted by muziqaz View Post
    Em, Raja did say RX will be faster than FE. What else do you want AMD to say after they release FE with working drivers and those drivers are performing great.
    RX will be faster and cheaper than FE, sure, but 1080Ti is faster and cheaper than Titan as well. What I (and I believe most people) expected from FE is AMD's equivalent of Titan. So, prosumer card, also good for playing games but overpriced for this task. Given its current benchmarks, FE is great for making games and pointless for playing games (vs competition). If RX is about as much ahead of FE as 1080Ti is ahead of Titan, RX doesn't look good at all.

    Quote Originally Posted by muziqaz View Post
    Edit: regarding 480. What planet are you living in? Who in their right mind ever promised 480 beating 1070? Not AMD for sure. Who in their right mind ever expected drivers to push 480 past 1070???
    Nah, AMD never promised anything, I don't blame them at all. AMD fans were hopeful that with DX12, card's pretty high theoretical flops will let it recover some ground vs 1070 which has similar theoretical performance and be way ahead of 1060, which should be worse in every manner. As devs will finally be able to code close to hardware and will be able to extract WAY more performance out of AMD cards, and their better scheduler etc etc...
    And this is exactly why this situation feels so similar to me. AMD isn't promising anything, fans are hoping. If AMD delivers 10% faster card for half the price, they will deliver exactly as they promised. Every sane gamer should indeed have waited for that instead of grabbing FE. Yet the card would be still meh compared to the competition.

    ---

    @many:
    Sure, we could have waited to see final RX instead of speculating, but hey, there were more than half of the pages written before we had ANY indication of card's performance. When was the launch of FE? Somewhere near 200-ish and we are not past 300 yet?
    Yet back then I didn't see that we should stop discussing potential performance, competitive landscape etc. So, why is this little data guessing any different? Because some of us are now guessing Vega will not be all that good for gaming, at least for the time it really matters (from release to +1-2 years).

    ---

    @nvr4d - I agree with most significant optimizations. However, are these optimizations really going to turn out important for games? List of changes AMD mentioned sounds helpful for VR, mixed workloads, professional usage etc, but not too many seem to be primarily for current DX11/12 games.
    As for the 79xx, NV conveniently just then shot them with FCAT and it took AMD a while to recover. And even that wouldn't even matter, AMD would still win, if only they launched at higher clocks which these cards were easily capable of. I agree, drivers helped AMD here a lot too, this is completely different card now than when it launched. But this was a gradual process over many releases, not a huge boost all at once.

  8. #2568
    Quote Originally Posted by Zizy View Post
    I agree with most significant optimizations. However, are these optimizations really going to turn out important for games? List of changes AMD mentioned sounds helpful for VR, mixed workloads, professional usage etc, but not too many seem to be primarily for current DX11/12 games.
    I think yes. Tiled rendering is the thing that can keep all those CUs busy and well-fed. Geometry culling significantly reduces the quantity of garbage going through geometry pipeline which is still not distributed.

    Quote Originally Posted by Zizy View Post
    As for the 79xx, NV conveniently just then shot them with FCAT and it took AMD a while to recover. And even that wouldn't even matter, AMD would still win, if only they launched at higher clocks which these cards were easily capable of. I agree, drivers helped AMD here a lot too, this is completely different card now than when it launched. But this was a gradual process over many releases, not a huge boost all at once.
    As for FCAT and NVidia, well, let's face the truth. Micro (and not so micro) stuttering WAS an issue with Tahiti which visibly ruined gaming experience, while Kepler-based GPUs rendered games way smoother. Game compatibility was another shot in the knee - RAGE was running perfectly fine on HD3870, on HD6970, but with Tahiti you were bound to play driver solitaire to even get it running, even after 5 months after game release. Skyrim was another 'great' example.

    So higher clocks were not going to save them, unfortunately. I was looking to upgrade my 6970, but ended up buying another one and staying that way for almost 2 years. Cayman and its drivers were the pinnacle of AMD's VLIW arch, so it was a pleasure to game on it, even in CFX, despite jet-like noise and overloading 1 kVA UPS. I still miss them - got to switch to 290X when they ran out of memory in Dragon Age Inquisition.

  9. #2569
    Senior Member
    Join Date
    Jan 2012
    Posts
    829
    Can we not just speculate and debate a released cards results and potential without stupid emotional crap? Is that too much to ask?
    For those complaining about Vega being compared to titan xp and not professional, AMD THEMSELVES used this comparison and also showed games playing on it so let's not start picking and choosing which results to use and what not to use, I don't care that amd says it's not for gaming, nvidia says the same thing for its prosumer card the Titan xp, yet it's great at games.

    Also nvidia professional quadro cards perform very well in games, only about 10% below consumer with professional drivers (when comparing similar cards) .
    Granted Vega is a brand new uarch and amd has little resources but to be expecting 35-40% performance improvement with Rx Vega is more than wishful thinking, it borders on religion.

    I suspect we might see 10-15% through a mixture of clocks and drivers, that won't be nearly enough on launch I'm afraid, but I do expect to see a good 30℅ improvement over the course of a year as I can't see this terrible performance as being representative of the uarch, that's impossible.

  10. #2570
    Senior Member
    Join Date
    Jan 2010
    Posts
    1,245
    Quote Originally Posted by Zizy View Post
    Nah, think about BD. Simple Thuban, directly shrunk to 32nm without any improvement would be a better chip. AMD clearly made tons of simulations, yet they failed to predict BD will fail that bad. Such stuff is always possible and Vega might have been in similar situation - looks great on paper, yet the thingy just doesn't work as it should on real device.
    I am on Thuban shrunk to 32nm :-) 2.3GHz/2.6GHz turbo, 35Watt.

Tags for this Thread

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  
WordPress Appliance - Powered by TurnKey Linux