Page 25 of 45 FirstFirst ...
15
23
24
25
26
27
35
... LastLast
  1. #481
    Quote Originally Posted by Amalaric View Post
    If AMD doesn't want to best Intel / nVidia they can just fuck off for all I care.
    They kinda did tbh. In terms of data crunching (aka multi-threaded performance, encoding/decoding etc.), Ryzen 1700 is miles a head of i7. Not sure about the Ryzen 1700's performance against Intel xeon processors tho.

    This is the primary reason why I prefer Ryzen 1700 over i7. I will be doing stuff other than gaming.
    Last edited by Kuntantee; 2017-07-09 at 10:47 AM.

  2. #482
    Warchief Zenny's Avatar
    10+ Year Old Account
    Join Date
    Oct 2011
    Location
    South Africa
    Posts
    2,171
    Review of the water cooled version:

    https://www.pcper.com/reviews/Graphi...-Cooled-Review

    Decent gains, roughly equals a stock 1080 now, albeit with a 50% larger die and double the power usage.

    Assuming AMD does it's post launch driver optimization, we are looking at a card that can be around 10% over the 1080 in best case scenarios.

    Hope they price the RX version for $449 at most.

  3. #483
    Quote Originally Posted by Zenny View Post

    Hope they price the RX version for $449 at most.
    That watercooler desing alone is so expensive, it will not be near $449, if it has the watercooler. Then again RX version might launch with a different design.

  4. #484
    Warchief Zenny's Avatar
    10+ Year Old Account
    Join Date
    Oct 2011
    Location
    South Africa
    Posts
    2,171
    Quote Originally Posted by mrgreenthump View Post
    That watercooler desing alone is so expensive, it will not be near $449, if it has the watercooler. Then again RX version might launch with a different design.
    I assume a triple 8-pin design with a really hefty air cooler would get the job done, in terms of cooling and power needed. The problem is RX Vega needs to retail for lower then a 1080 in price due to it being very similar in performance whilst having some rather large drawbacks. The 1080 can be found on specials all the time now due to it being so long on the market.

    So, under $500 is basically a requirement if they want to ship any significant numbers.

  5. #485
    Quote Originally Posted by Zenny View Post
    I assume a triple 8-pin design with a really hefty air cooler would get the job done, in terms of cooling and power needed. The problem is RX Vega needs to retail for lower then a 1080 in price due to it being very similar in performance whilst having some rather large drawbacks. The 1080 can be found on specials all the time now due to it being so long on the market.

    So, under $500 is basically a requirement if they want to ship any significant numbers.
    The underlined part is the one that is really important.

    RX Vega looks to be the top-end of that design. That's all their going to get out of that chip.

    And in about 3 months from now (2 1/2-ish from RXV launch) professinal Volta (V100) is going to start shipping. (Early Q3 this year).

    Expect consumer Volta right after that. (Another 2-ish to 3-ish months - late Q4 or early Q1 2018 at the outside).

    So, within just a few months of launch, at best, AMD is going to be back to competing for the mid-range, with nothing to show at the top end at all, as all reports coming out of people who have engineering samples of Volta is that it is as much of a leap over Pascal as Pascal was over Maxwell.

    So, the RX Vega will end up being about as powerful as a 2060/1160 (whatever we call them) that is going to end up retailing for ~260.

    AMD needs to get the lead out.

    They showed with Ryzen that they can still innovate. Lets see some of that in the GPU space. I'd like some options in the future when it is time to rebuild.

  6. #486
    Quote Originally Posted by Kagthul View Post

    Expect consumer Volta right after that. (Another 2-ish to 3-ish months - late Q4 or early Q1 2018 at the outside).
    If Vega doesn't outperform the 1080 by a fair margin, we won't see consumer volta(outside Titan) until spring 2018.

  7. #487
    The Lightbringer Evildeffy's Avatar
    15+ Year Old Account
    Join Date
    Jan 2009
    Location
    Nieuwegein, Netherlands
    Posts
    3,772
    Except that Volta for the consumer market will not come out in 2017 since GDDR6 won't enter mass production untill "early 2018" which can be the entirety of Q1 2018.

    Volta will use GDDR6 from any PoV we've heard so far, so unless you have sources that tell me that Volta will continue on GDDR5X and that the V100 is a completely different animal to the consumer variants I'm going to keep it at Volta very likely Q2 2018.

    Since Micron is the only one to produce GDDR6 for now ... it won't be earlier than Q2 2018.

    Also it is unwise to even remotely assume an all new architecture performance.

  8. #488
    hmm, so pretty much liquid-cooling and/or also bumping up the power target to 125% (thus going over 400W) was needed for the FE silicon to hold a solid 1600 Mhz boost

    and at 1600 Mhz its equal to stock GTX 1080


    from gamersnexus hybrid liquid-cooled modding of FE they got a max OC of 1700 Mhz I believe - that could be on par with 1900-2000 Mhz 1080



    question is - will air-cooled aftermarket Vega cards reach any of these (1600 stable, much less 1700) or will that be purely AIO territory ?
    Last edited by Life-Binder; 2017-07-17 at 07:32 PM.

  9. #489
    I was going to post something in response to Evildufus' idiocy, but thought better of it.

    Why is it so hard for people to actually keep up on things that have changed since they were first talked about nine months ago?

    /picardfacepalm

    User was infracted.
    Last edited by noteworthynerd; 2017-07-27 at 09:27 PM.

  10. #490
    It needs to significantly undercut the 1080 in terms of pricing to be remotely a competitive option. It was being benched in the article against Nvidia cards at Founders Edition clock rates, so it's probably already 5-10% slower than AIB 1080s. Plus, if AMD already had to put on a water cooling solution to reach those performance levels, it's a pretty good sign that there isn't near as much head room for AIB manufacturers to improve on the performance (certainly not as much as there is for 1080 AIB cards to move up from the Founders Edition cooler). In other words, if it's only matching reference card 1080 performance, it's already basically behind the 1080.

    It's also consuming a full 200 watts more power than a 1080 (and 100 watts more than a 1080 Ti) based on those benchmarks. That is pretty brutal, and is going to add to another $50+ to the effective price of the GPU, because while a 1080 would be fine on a quality ~550w EVGA or Seasonic PSU that can be had for $40-50, this thing is probably going to need a 750+ W PSU for $100+. The price has to reflect the extra PSU requirements. On top of that, even if everything else was equal, by default, I wouldn't go with an AMD GPU over an Nvidia GPU because of the inferior drivers and frequency of driver updates. That factor also needs to be compensated.

    Therefore, I'd make the following conclusions off that article.
    - It's probably 5-10% slower than a decent AIB 1080 if it's only matching a Founder's.
    - It's going to require that you spend $50-$75 more on your PSU
    - If you're using it for 6 hrs a day for heavy gaming, it's going to cost you ~$50 extra a year in electricity expenses (assuming 12c/kWH).
    - It has a weaker driver infrastructure/stability/update rate than Nvidia cards

    Assuming the baseline price of the 1080 stabilizes to ~$500, I really can't see this thing being competitive at anything more than $350.

    - - - Updated - - -

    Quote Originally Posted by Kagthul View Post
    The underlined part is the one that is really important.

    RX Vega looks to be the top-end of that design. That's all their going to get out of that chip.

    And in about 3 months from now (2 1/2-ish from RXV launch) professinal Volta (V100) is going to start shipping. (Early Q3 this year).

    Expect consumer Volta right after that. (Another 2-ish to 3-ish months - late Q4 or early Q1 2018 at the outside).

    So, within just a few months of launch, at best, AMD is going to be back to competing for the mid-range, with nothing to show at the top end at all, as all reports coming out of people who have engineering samples of Volta is that it is as much of a leap over Pascal as Pascal was over Maxwell.

    So, the RX Vega will end up being about as powerful as a 2060/1160 (whatever we call them) that is going to end up retailing for ~260.

    AMD needs to get the lead out.

    They showed with Ryzen that they can still innovate. Lets see some of that in the GPU space. I'd like some options in the future when it is time to rebuild.
    And, this GPU is not going to be mid-range viable unless they can work miracles with the power utilization. Mid-range buyers are not typically going to put up with huge PSU requirements, and exotic water cooling setups and/or the fan noise/heat/power consumption this design is going to bring. The RX 400 series was successful in the mid range in part because it had reasonable power consumption.

  11. #491
    Warchief Zenny's Avatar
    10+ Year Old Account
    Join Date
    Oct 2011
    Location
    South Africa
    Posts
    2,171
    The 1080 shipped months before the time Micron stated GDDR5x would be in mass production. Heck Nvidia confirmed the 1080 was using GDDR5x even before Micron did the announcement.

    For the last 5 generations Nvidia has been releasing xx80 class products on a roughly 18 month schedule, give or take a couple months. The longest gap being 20 months. Assuming we are looking at another 20 month gap gives us a launch date of Jan 2018, but it could be sooner. GDDR5x has just hit speeds of 16GBps so Volta can launch with that if needed, that would give a 256bit card 512Gbps, which is plenty.
    Last edited by Zenny; 2017-07-17 at 08:03 PM.

  12. #492
    The Lightbringer Evildeffy's Avatar
    15+ Year Old Account
    Join Date
    Jan 2009
    Location
    Nieuwegein, Netherlands
    Posts
    3,772
    Quote Originally Posted by Kagthul View Post
    I was going to post something in response to Evildufus' idiocy, but thought better of it.
    Look at that ... you're still an asshole, good thing that hasn't changed I see.

    Quote Originally Posted by Kagthul View Post
    Why is it so hard for people to actually keep up on things that have changed since they were first talked about nine months ago?

    /picardfacepalm
    http://wccftech.com/sk-hynix-gddr6-n...-gpu-gtc-2017/

    9 months ago .. hmm cool, I didn't realize Computex was held in Oktober 2016 last year... could've sworn we had one 2 months ago.

    Or are you referring to that singular rumour of fudzilla and fudzilla only that Volta will use GDDR5X instead?
    Hmmm ... Announcement from SK Hynix to be believed or Fudzilla's rumour mill who's been wrong more times than I care to count?...

    If you're going to act all smug and superior you may want to back up your claims instead of responding like that.

    So either respond like a normal person or don't respond at all.

    - - - Updated - - -

    Quote Originally Posted by Zenny View Post
    The 1080 shipped months before the time Micron stated GDDR5x would be in mass production. Heck Nvidia confirmed the 1080 was using GDDR5x even before Micron did the announcement.

    For the last 5 generations Nvidia has been releasing xx80 class products on a roughly 18 month schedule, give or take a couple months. The longest gap being 20 months. Assuming we are looking at another 20 month gap gives us a launch date of Jan 2018, but it could be sooner. GDDR5x has just hit speeds of 16GBps so Volta can launch with that if needed, that would give a 256bit card 512Gbps, which is plenty.
    Quite possibly but I'm remembering mass production dates of GDDR5X being after the availability launch, not during it's paper launch.

    Also prototyping != Mass Production.

    Clarification:
    https://www.digitaltrends.com/comput...5x-production/ <-- Mass Production in started in early May 2016.
    http://wccftech.com/micron-gddr5x-me...ss-production/ <-- Confirmation of mass production way before.

    GTX 1080 purchase capability officially March 10th 2016... well how many could you buy for real until about July?
    Availibility in the NL was only 40 cards of the 1080/1070 till July hit in total from March.
    You do the math why.

    User was infracted.
    Last edited by noteworthynerd; 2017-07-27 at 09:28 PM.

  13. #493
    Quote Originally Posted by Evildeffy View Post
    9 months ago .. hmm cool, I didn't realize Computex was held in Oktober 2016 last year... could've sworn we had one 2 months ago.

    Or are you referring to that singular rumour of fudzilla and fudzilla only that Volta will use GDDR5X instead?
    Hmmm ... Announcement from SK Hynix to be believed or Fudzilla's rumour mill who's been wrong more times than I care to count?...

    If you're going to act all smug and superior you may want to back up your claims instead of responding like that.

    So either respond like a normal person or don't respond at all.
    They will use whatever is available. It's not like there is much of a difference between the two in terms of performance (at least initially).

    Quote Originally Posted by Evildeffy View Post
    Quite possibly but I'm remembering mass production dates of GDDR5X being after the availability launch, not during it's paper launch.

    Also prototyping != Mass Production.

    Clarification:
    https://www.digitaltrends.com/comput...5x-production/ <-- Mass Production in started in early May 2016.
    http://wccftech.com/micron-gddr5x-me...ss-production/ <-- Confirmation of mass production way before.

    GTX 1080 purchase capability officially March 10th 2016... well how many could you buy for real until about July?
    Availibility in the NL was only 40 cards of the 1080/1070 till July hit in total from March.
    You do the math why.
    GTX 1080 launched at the end of May 2016 (and that was FE), what are you talking about? You could definitely buy a partner card at the end of June in the US, in the EU they mostly started shipping at the same time as GTX 1070 (which was beginning of July, when I got mine). You could preorder a FE and get it at the day of launch (I know a handful of people who did it), but buying an overpriced FE where you also had to pay for the shipment from the US is pretty stupid.
    R5 5600X | Thermalright Silver Arrow IB-E Extreme | MSI MAG B550 Tomahawk | 16GB Crucial Ballistix DDR4-3600/CL16 | MSI GTX 1070 Gaming X | Corsair RM650x | Cooler Master HAF X | Logitech G400s | DREVO Excalibur 84 | Kingston HyperX Cloud II | BenQ XL2411T + LG 24MK430H-B

  14. #494
    I received my Gigabyte 1080 G1 in the first half of June 2016 in EU


    although I live in a smaller EU country where not everything gets sold out or pre-ordered (you can still buy cards even in mining craze peaks and we had aftermarket Pascal cards available shortly after official launch)

    not like ~Germany or UK

  15. #495
    The Lightbringer Evildeffy's Avatar
    15+ Year Old Account
    Join Date
    Jan 2009
    Location
    Nieuwegein, Netherlands
    Posts
    3,772
    Quote Originally Posted by Thunderball View Post
    They will use whatever is available. It's not like there is much of a difference between the two in terms of performance (at least initially).
    It is required in controller design in the silicon, can't switch between controllers on a die.
    The difference is more than in performance, it's also in power use, GDDR5X uses 1.8V where-as GDDR6 uses 1.35V as an example.
    So it's either designed with GDDR6 or GDDR5X, can't be both and seeing as how SK Hynix pretty much confirmed it being used in the next generation graphics cards ... well you get the point.

    Quote Originally Posted by Thunderball View Post
    GTX 1080 launched at the end of May 2016 (and that was FE), what are you talking about? You could definitely buy a partner card at the end of June in the US, in the EU they mostly started shipping at the same time as GTX 1070 (which was beginning of July, when I got mine). You could preorder a FE and get it at the day of launch (I know a handful of people who did it), but buying an overpriced FE where you also had to pay for the shipment from the US is pretty stupid.
    I think you're misunderstanding something here.

    The mass production of GDDR5X was already going and announced by Micron when the cards for sale, official sale went up 10th of May, but even then availability was scarce, was from nVidia itself and retail didn't receive anything till 27th of May, but you could buy them (first batch) directly.
    The FE edition you could get but again that was only ~40 cards in total for the entirety of the Netherlands for almost 2 months when in early July the partner cards were finally starting to get introduced into the market.
    (Pre-ordering an nVidia GTX 1080 FE to NL was not possible btw, so we had to make due with what nVidia shipped us)

    I did specifically state "Availability in the NL", NL being Netherlands, a country in the EU and also one with the highest tech throughput in Europe.

    My point with this was that Mass Production was announced before and confirmed to start before the GTX 1080 was being sold.
    Since I highly doubt AMD's Navi will be the sole user of GDDR6 in 2018 and since AMD has bet it's chips on HBM... nVidia is the only other graphics manufacturer present which brings me yet again to my earlier point:
    If nVidia's Volta cards will use GDDR6 then that means it will not launch untill Q1 - Q2 2018 at the earliest.
    I will state again that you cannot design a silicon base with both GDDR5(X) and GDDR6 controller, it is not feasible.

    GDDR5 was an exception to this rule as it was specifically designed to be for the same controller but GDDR5(X) and GDDR6 are different generation and mix-matching isn't going to work.

    To expand on that with the "GV100" card ... that will use HBM2 and is an entirely different beast much like "GP100" was, they are not comparable in the least with the consumer line of cards.

    Assuming Volta will launch when it's going to use GDDR6 within this year (very early Q4 according to some people's estimation here) when mass production won't be till "early 2018" is presumptuous and stupid.
    If all memory manufacturers (Micron, SK Hynix and Samsung) come forward tomorrow that Mass Production of GDDR6 has started right now... then it'd be a possibility but do you see that happening since 2 months ago those very same memory manufacturers stated "early 2018" for that mass production date?
    Nope? I don't either... that's my point.

    - - - Updated - - -

    Quote Originally Posted by Life-Binder View Post
    I received my Gigabyte 1080 G1 in the first half of June 2016 in EU

    although I live in a smaller EU country where not everything gets sold out or pre-ordered (you can still buy cards even in mining craze peaks and we had aftermarket Pascal cards available shortly after official launch)

    not like ~Germany or UK
    You were lucky, trying to get a GTX 1080 was almost impossible here, like I said ... NL was a high throughput country.
    The fact we only got 40 cards for the entire damned country was stupid as well (contacts @ suppliers).

  16. #496
    Quote Originally Posted by mrgreenthump View Post
    If Vega doesn't outperform the 1080 by a fair margin, we won't see consumer volta(outside Titan) until spring 2018.
    I was waiting to buy a new video card but im worried the gtx 11 series (volta or w.e. its gonna be called) wont release till 2018...would be funny for nvidia to see how vega turns out and just release 11 series a month after that demolishs vega

  17. #497
    Quote Originally Posted by Evildeffy View Post
    I think you're misunderstanding something here.

    The mass production of GDDR5X was already going and announced by Micron when the cards for sale, official sale went up 10th of May, but even then availability was scarce, was from nVidia itself and retail didn't receive anything till 27th of May, but you could buy them (first batch) directly.
    The FE edition you could get but again that was only ~40 cards in total for the entirety of the Netherlands for almost 2 months when in early July the partner cards were finally starting to get introduced into the market.
    (Pre-ordering an nVidia GTX 1080 FE to NL was not possible btw, so we had to make due with what nVidia shipped us)

    I did specifically state "Availability in the NL", NL being Netherlands, a country in the EU and also one with the highest tech throughput in Europe.

    My point with this was that Mass Production was announced before and confirmed to start before the GTX 1080 was being sold.
    Since I highly doubt AMD's Navi will be the sole user of GDDR6 in 2018 and since AMD has bet it's chips on HBM... nVidia is the only other graphics manufacturer present which brings me yet again to my earlier point:
    If nVidia's Volta cards will use GDDR6 then that means it will not launch untill Q1 - Q2 2018 at the earliest.
    I will state again that you cannot design a silicon base with both GDDR5(X) and GDDR6 controller, it is not feasible.

    GDDR5 was an exception to this rule as it was specifically designed to be for the same controller but GDDR5(X) and GDDR6 are different generation and mix-matching isn't going to work.

    To expand on that with the "GV100" card ... that will use HBM2 and is an entirely different beast much like "GP100" was, they are not comparable in the least with the consumer line of cards.

    Assuming Volta will launch when it's going to use GDDR6 within this year (very early Q4 according to some people's estimation here) when mass production won't be till "early 2018" is presumptuous and stupid.
    If all memory manufacturers (Micron, SK Hynix and Samsung) come forward tomorrow that Mass Production of GDDR6 has started right now... then it'd be a possibility but do you see that happening since 2 months ago those very same memory manufacturers stated "early 2018" for that mass production date?
    Nope? I don't either... that's my point.
    Official FE launch was May 27th. Nvidia themselves stated that they expect partner cards to start shipping in two weeks. That's very close to what actually happened (at least in the US). Everything that happens overseas is mostly due to specific agreements they have with partners locally, so that's not at all indicative of what comes out of the factory. I dont think memory availability was the problem with GTX 1080 numbers, as GTX 1070 had the same problem, and those dont use GDDR5X.

    I dont see any point to specifically use GDDR6, it doesnt offer any significant advantages over GDDR5X right now. They could very well use GDDR5X in "2080" and then use GDDR6 in "2080Ti". Obviously once the new manufacturing process matures higher frequencies would be possible, that combined with a dual channel architecture will offer some performance benefits, but right now it doesnt look like there are any.

    In the end I dont see any of these making any impact. Vega very likely wont be a contenter (it might get there in performance, but price and power consumption are too crazy for an average gaming PC user), Navi likely wont come before 2019 (if they still plan to use 7nm fabrication process, which looks like they do), so unless it's someone groundbreaking Nvidia is going to have a lot of happy time.
    R5 5600X | Thermalright Silver Arrow IB-E Extreme | MSI MAG B550 Tomahawk | 16GB Crucial Ballistix DDR4-3600/CL16 | MSI GTX 1070 Gaming X | Corsair RM650x | Cooler Master HAF X | Logitech G400s | DREVO Excalibur 84 | Kingston HyperX Cloud II | BenQ XL2411T + LG 24MK430H-B

  18. #498
    AMD has been ahead of NVidia in pure tflops for a long time now. Vega will be once again tflop king until Volta comes around, and Volta is a monstrous chip, almost double the size of Vega, for a small lead in tflops for insane prices. (We won't see the full Volta chip in geforce cards for a long time to come). AMD's big problem is that for some reason they can't translate their tflop lead into gaming performace very well. Their other problem (which they are working on) is CUDA, which became the de-facto gpgpu language for many applications.

  19. #499
    The Lightbringer Evildeffy's Avatar
    15+ Year Old Account
    Join Date
    Jan 2009
    Location
    Nieuwegein, Netherlands
    Posts
    3,772
    Quote Originally Posted by Thunderball View Post
    Official FE launch was May 27th. Nvidia themselves stated that they expect partner cards to start shipping in two weeks. That's very close to what actually happened (at least in the US). Everything that happens overseas is mostly due to specific agreements they have with partners locally, so that's not at all indicative of what comes out of the factory. I dont think memory availability was the problem with GTX 1080 numbers, as GTX 1070 had the same problem, and those dont use GDDR5X.
    You could buy the first batch of FEs during the reveal event, which was very limited, way before May 27th.
    More batches and up would use the 27th date and up, hence why I counted it along.
    And no it's not with what they agreed with partners it's what they could supply to Europe and nVidia simply hadn't prioritized any other country but the US for GPU sales, which is why they got burned with their rushed launch.
    That said ... like I said before GDDR5X is an exception as it uses the same controller as GDDR5, GDDR6 won't and you won't be seeing a card that can do either.
    So you won't see a 1180/2080 with GDDR6 and then an 1170/2070 with GDDR5X unless it's a completely different silicon.

    Quote Originally Posted by Thunderball View Post
    I dont see any point to specifically use GDDR6, it doesnt offer any significant advantages over GDDR5X right now. They could very well use GDDR5X in "2080" and then use GDDR6 in "2080Ti". Obviously once the new manufacturing process matures higher frequencies would be possible, that combined with a dual channel architecture will offer some performance benefits, but right now it doesnt look like there are any.
    Smaller physical size, reduced memory controller size on silicon, dual channel architecture for increased bandwidth with lower bus sizes, CONSIDERABLY less voltage/wattage requirements... just to name a few.
    There are more benefits to it than pure speed.

    Quote Originally Posted by Thunderball View Post
    In the end I dont see any of these making any impact. Vega very likely wont be a contenter (it might get there in performance, but price and power consumption are too crazy for an average gaming PC user), Navi likely wont come before 2019 (if they still plan to use 7nm fabrication process, which looks like they do), so unless it's someone groundbreaking Nvidia is going to have a lot of happy time.
    It depends, how things look like it's not good for AMD's graphics division ... that said we are assuming things before we see them.
    I'm not expecting much but I'll await the reveal and reviews before making judgement.
    Power consumption is less of an issue than you think unless the RX Vega gobbles down 400W though, but we'll see if it's worth a damn in performance.

    Like you said don't expect anything too much from RX Vega but also don't expect nVidia to break the laws of time and physics and expect Volta this year when it's memory subsystem require 2018 to begin to being mass produced.
    And just saying this once again... you cannot use a GDDR5(X) memory controller to run GDDR6 memory, so the silicon either needs 2 separate IMCs or nVidia spends double the amount on R&D to create 2 different silicons for each different tier of graphics cards, which I do not see happening.

  20. #500
    Warchief Zenny's Avatar
    10+ Year Old Account
    Join Date
    Oct 2011
    Location
    South Africa
    Posts
    2,171
    Quote Originally Posted by Evildeffy View Post
    Clarification:
    https://www.digitaltrends.com/comput...5x-production/ <-- Mass Production in started in early May 2016.
    http://wccftech.com/micron-gddr5x-me...ss-production/ <-- Confirmation of mass production way before.

    GTX 1080 purchase capability officially March 10th 2016... well how many could you buy for real until about July?
    Availibility in the NL was only 40 cards of the 1080/1070 till July hit in total from March.
    You do the math why.
    As per your own links the Geforce 1080 launched over a month earlier when Micron indicated it should start mass production, they only actually confirmed May after Nvidia announced the 1080.

    The product wasn’t expected to become available until this summer, and the standard wasn’t published by the JEDEC Solid State Technology Association until this past January
    Based on that it's perfectly possible for Volta to launch in January with GDDR6. But that being said, Volta is not dependant on GDDR6, as a hypothetical Geforce 2080 can launch with GDDR5x memory and still have 60% increased bandwidth over the 1080.

    - - - Updated - - -

    Quote Originally Posted by zox2 View Post
    AMD has been ahead of NVidia in pure tflops for a long time now. Vega will be once again tflop king until Volta comes around, and Volta is a monstrous chip, almost double the size of Vega, for a small lead in tflops for insane prices. (We won't see the full Volta chip in geforce cards for a long time to come). AMD's big problem is that for some reason they can't translate their tflop lead into gaming performace very well. Their other problem (which they are working on) is CUDA, which became the de-facto gpgpu language for many applications.
    Consumer Volta won't actually be as monstrous as the Tesla variant, all those Tensor and FP64 cores will get removed, drastically shrinking the die size. As for TFLOP, the difference is slight at the high end. Most 1080ti cards can boost to over 14TFLOP with a easy overclock. It takes a Vega card clocked at almost 1750Mhz to equal that.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •