|
256 Maxwell cores is HALF of a desktop GTX 750. Xbone has a GPU roughly equivalent to a Radeon HD 7770. A GTX 750 (not Ti) is 5-10% faster than a 7770. Assuming the Tegra Maxwell-based GPU operates at desktop clockspeed (it won't), NX's GPU will be about half as fast as non-Scorpio Xbox One. That's without getting into the CPU disadvantage it will have.
|
# ? Aug 24, 2016 22:49 |
|
|
# ? May 21, 2024 13:19 |
|
I also find it impressive that Nintendo is re-releasing N64/Gamecube/Wii games onto the 3DS. What were once console games are now handheld, and also running faster in some cases. Although they probably have source code, which would let them do high-level emulation rather than low-level hardware emulation.
|
# ? Aug 24, 2016 23:24 |
|
Paul MaudDib posted:I also find it impressive that Nintendo is re-releasing N64/Gamecube/Wii games onto the 3DS. What were once console games are now handheld, and also running faster in some cases. Uh, what? They ported some N64 games sure (they also ported some N64 games to the original DS), but they sure don't have straight up N64 games to buy on the 3DS or New 3DS, and they definitely don't have GameCube or Wii games on there either, again outside complete recodes/ports. I've really got no idea where you got this notion from. Did you mishear a rumor about what the NX might be able to do? Like honestly I can't even find popular rumors about the 3DS/new 3DS running GameCube and Wii games which aren't just wild speculation from before it released. There simply isn't enough power in the thing to handle it. fishmech fucked around with this message at 23:34 on Aug 24, 2016 |
# ? Aug 24, 2016 23:29 |
|
fishmech posted:Uh, what? They ported some N64 games sure (they also ported some N64 games to the original DS), but they sure don't have straight up N64 games to buy on the 3DS or New 3DS, and they definitely don't have GameCube or Wii games on there either, again outside complete recodes/ports. I've really got no idea where you got this notion from. "Recodes and ports" are exactly what I'm referring to. There's Starfox 64, Super Mario 64, and both Zelda games from N64, Splinter Cell from GC, and apparently there's a Donkey Kong game and some FF-esque RPG from the Wii. http://nintendo.wikia.com/wiki/List_of_games_re-released_onto_Nintendo_handhelds#N64
|
# ? Aug 24, 2016 23:57 |
|
Paul MaudDib posted:"Recodes and ports" are exactly what I'm referring to. So nothing to do with actually putting games from other systems on there. Again: the regular old DS was full of N64 ports too. And half the Wii games they list are extremely multi-platform games that were also released on PSP or Vita, 360, PS3, PC. It's like being impressed that they poo poo out the yearly Madden games to every system. And there's absolutely 0 emulation, low level or high level, involved with these ports, despite what you originally claimed in your post - doing that would require quite a bit more processing power than the platform provides! The Nintendo 3DS is a very weak system, and the New 3DS really doesn't improve matters much. Edit: Another aspect is that the DS and 3DS are very low resolution systems: the DS is 256x192 on both screens, the 3DS is 400x240 (optionally 800x240 when the 3D mode is on, but still rendering the same aspect ratio) on the top screen and 320x240 on the bottom screen. And most of the time the bulk of 3D rendering is on only one screen, due to the hardware limitations, the other screen then being dedicated to very low intensity 2D stuff. Most of the games ported from the regular consoles to it thus only need to run at a reduced resolution compared to the original console, and even when it's the same resolution (many N64 games rendered at 320x240) you can get away with reduced detail in many aspects because they aren't noticeable on the small screens regardless. fishmech fucked around with this message at 03:05 on Aug 25, 2016 |
# ? Aug 25, 2016 00:12 |
|
Kyonko is right, they're completely native games, hence why they can run at all. If you want to talk emulation, the original ds does like 3fps in ps1 games. New 3ds does about 30. PSP could get 20-30 fps in some n64 games including star fox.
|
# ? Aug 25, 2016 02:40 |
|
More AMD stuff from Hot Chips, may be more useful for someone more chip design literate than I am, it's pretty technical: http://www.anandtech.com/show/10591/amd-zen-microarchiture-part-2-extracting-instructionlevel-parallelism
|
# ? Aug 25, 2016 21:15 |
|
http://techreport.com/news/30600/amd-takes-a-335m-one-time-charge-for-more-sourcing-flexibility TL;DR: We paid GloFo to get this loving albatross off our neck. The move is ostensibly a good one for AMD, because their balls aren't going to be stapled to GloFo's janky rear end anymore, but: * GloFo gets a bunch of cash they can blow on hookers and cocaine instead of sorting their poo poo out * It frees up manufacturing capacity for GloFo to sell to other companies. Which makes it kind of a wash for AMD. =( SwissArmyDruid fucked around with this message at 21:26 on Sep 1, 2016 |
# ? Sep 1, 2016 21:23 |
|
Is there going to be an RX480 Not poo poo edition now, with 20% higher clocks at lower power usage?
|
# ? Sep 1, 2016 21:29 |
|
Who's going to make it for them, Intel?
|
# ? Sep 1, 2016 21:53 |
|
mobby_6kl posted:Who's going to make it for them, Intel? Samsung is likely.
|
# ? Sep 1, 2016 22:06 |
|
mobby_6kl posted:Who's going to make it for them, Intel? Samsung most likely, but I expect some of the I swear, if it weren't for that loving 15 year contract.... (Was it Dirk Meyer? I think it was Dirk Meyer.) SwissArmyDruid fucked around with this message at 22:40 on Sep 1, 2016 |
# ? Sep 1, 2016 22:37 |
|
Looks like they have to pay an additional fee to GloFo to source from other people anyway, so hopefully that doesn't crank up prices too much. Is this is time to tape out for Vega? I think that would be a pretty tight time line, but Samsung I think already might have access to the details of the architecture?
|
# ? Sep 1, 2016 23:47 |
|
I think they already ran some test parts over at Samsung, yes.
|
# ? Sep 1, 2016 23:48 |
|
Ah neat, I couldn't remember if that was for Polaris or Zen or even both.
|
# ? Sep 2, 2016 00:13 |
|
If Vega is coming in 2017 as most suspect, I highly doubt it's going to be a part of the 400 series. If there really is a substantial difference between GloFo and Samsung, there is no reason not to move Polaris over as well for a 500 series release. If GloFo's problem is they can't scale performance up that still leaves GloFo Polaris for mobile where even the bad ones can sip power at .5v to .7v @ 600-900Mhz, while Samsung delivers desktop parts. I'm still going to stick to my theory that AMD was pressured into releasing Polaris before they were ready to having a competing product with Nvidia, and the original release was only intended for mobile. RX500 series in March?
|
# ? Sep 2, 2016 13:35 |
|
Isn't vega pretty much confirmed to use the fury naming?
|
# ? Sep 2, 2016 13:41 |
|
Truga posted:Isn't vega pretty much confirmed to use the fury naming? No? Its' expected, but there isn't just Vega 10 now, there is a Vega 11. Further, it appears AMD has dropped the Nano and NonX descriptors. There really doesn't seem to be room to include both Vegas in a 400 series line up, so I'd say it makes more sense to move to the 500 series. RX Fury: Vega 11 XT, rumored 6144SP RX 590: Vega 11 Pro, rumored 5376SP RX 580: Vega 10 XT, rumored 4096SP RX 570: Vega 10 Pro, rumored 3584SP RX 560: Polaris 10 XT, 2304SP RX 550: Polaris 10 Pro, 2048SP R 540: Polaris 11 XT, 1024SP R 530: Polaris 11 Pro, 896SP Makes a lot of sense when matched up to Nvidia's line up and relieves pricing compression on the product stack. The other theory is that Vega 11 and Vega 10 are basically the same thing just for different markets, consumer vs enterprise/semicustom.
|
# ? Sep 2, 2016 14:06 |
|
SwissArmyDruid posted:Samsung most likely, but I expect some of the If there's an extremely dumb management decision, it's probably Hector Ruiz. Buying Ati at an inflated price, cutting R&D funds for Barcelona, oversaw the AMD GloFo split landing himself a sweet 3 mil bonus with some insider trading on the side.
|
# ? Sep 5, 2016 09:06 |
|
AMD just diluted it's shares and is reaping just north of 1 billion from doing so, apparently to pay off half of it's 2 billion debt. This combined with renegotiating the wafer deal signals to me that AMD is concerned about rollout speed of Zen, similar to issues with getting ye olde Athlon moving to nab the necessary marketshare. If Zen takes too long to start generating enough revenue, it could possible that the need to overcome debt would pale to the need to buy wafers, to generate more money? It's likely AMD is moving to Samsung fabs then, as porting from 14nmLPP to 16nmFF+ would be too costly. There is also some rumor AMD will utilize Samsung 10nm instead because it's apparently easy to port 14nmLPP to 10nmLPP? Whatever the case, GloFo just got the "YOU HAVE FAILED ME FOR THE LAST TIME!" speech.
|
# ? Sep 7, 2016 17:23 |
|
A leapfrog to 10nm would probably be a huge boon to mobile and server oriented SKUs, wouldn't they? Those power savings just on a die shrink would be pretty enormous for graphs comparing them to Excavator and really help with current Intel product comparisons. This is pretty exciting. I would love it if AMD had albatross for dinner this Christmas.
|
# ? Sep 7, 2016 19:48 |
|
Again it's entirely speculation that AMD is moving to 10nmLPP and it's a bit strained in my opinion. Obviously moving to 10nm would probably allow AMD to catchup with Intel and Nvidia more effectively but it'd be costly and I can only see AMD doing it if they think 7nm is going to be delayed to whether it's useful to do so.
|
# ? Sep 7, 2016 21:48 |
I'm a bit skeptical AMD will quickly move to 10nm due to the astronomical design costs involved for 10nm. It's already like a quarter billion to design for 14nm (for a completely new device) and 10nm is expected to be almost half a billion. Honestly, economics might be the limiting factor here rather than physics.
|
|
# ? Sep 8, 2016 04:39 |
|
How is it hard anyway, jus ttake your 14nm design and scale it 30% smaller BOOM you have a 10nm design
|
# ? Sep 8, 2016 04:57 |
|
Watermelon Daiquiri posted:I'm a bit skeptical AMD will quickly move to 10nm due to the astronomical design costs involved for 10nm. It's already like a quarter billion to design for 14nm (for a completely new device) and 10nm is expected to be almost half a billion. Honestly, economics might be the limiting factor here rather than physics. Why is it so much more expensive? That's a mind-boggling delta.
|
# ? Sep 8, 2016 07:16 |
|
We are presently at a point where the smallest features on a chip are 7,000 times smaller than a human hair. We are fundamentally beginning to reach the limits of not only what bulk silicon can accomplish, but what silicon-based semiconductors as a whole can achieve. After the 4nm/7nm/10nm node, depending on what kind of dick-measuring benchmark that marketing is presently using, in order to have semiconductors continue to advance in terms of shrinking process, we will have to move to other, more exotic materials, and obviously more expensive materials. Part of the reason why we've gotten as far as we have on silicon alone is because of silicon's utter ubiquity, in the fact that silicon-containing minerals comprise somewhere north of 90% of our planet's crust, which means it's bleeping cheap to source. Frankly, it's a bit of a miracle that we've managed to continue to skate by on silicon as long as we have. SwissArmyDruid fucked around with this message at 07:33 on Sep 8, 2016 |
# ? Sep 8, 2016 07:31 |
Ha, I know, right? I mean, we aren't even strictly using just doped silicon anymore, we add extra stuff (not standard hole electron donors) to the transistors to even get the mobility we need. (I'm not sure how much is public knowledge) Considering there are only 4 companies in the world that do sub 20nm nodes (like ten actual fabs), and the tools last a decade+, there isn't alot of demand for the super high end photo tools (which costs 9 figure!). Hell, I don't even know if 450mm wafers will even be a thing unless a whole new fab gets built somewhere since they'd have to replace or greatly modify almost every single tool and transport mechanism to accept the larger wafer size. That will take literally billions of dollars to accomplish.
|
|
# ? Sep 8, 2016 18:42 |
|
Sounds like an opportunity for a massive government grant to me. Lets make silicon great again! (But no seriously you could make the argument that 450mm wafers and EUV constitute a deciding edge in national defense and security)
|
# ? Sep 8, 2016 19:35 |
|
What would manufacturing chips on bigass 450mm wafers accomplish?
|
# ? Sep 8, 2016 19:55 |
|
Potato Salad posted:What would manufacturing chips on bigass 450mm wafers accomplish? I'm assuming that it would make both Intel and the cloud hyperscalers (Facebook, Google, MS, Amazon) really happy because the millions of Xeons they soak up would get a bit cheaper.
|
# ? Sep 8, 2016 20:04 |
SwissArmyDruid posted:We are presently at a point where the smallest features on a chip are 7,000 times smaller than a human hair. We are fundamentally beginning to reach the limits of not only what bulk silicon can accomplish, but what silicon-based semiconductors as a whole can achieve. I've heard or seen this at least 20 times since the 80s, guess what it's bullshit every time. Somehow we were approaching the "fundamental size limit of the atom" or some such utter horseshit back in 1989 and and every year since then. Somehow we manage. Once again, the only reason things have slowed down is because AMD has been incompetent for many years, it has absolutely nothing to do with physics or engineering no matter how many times you read that rubbish in wired and verge.
|
|
# ? Sep 8, 2016 20:14 |
|
Pryor on Fire posted:I've heard or seen this at least 20 times since the 80s, guess what it's bullshit every time. Somehow we were approaching the "fundamental size limit of the atom" or some such utter horseshit back in 1989 and and every year since then. Somehow we manage. It's not about size of an atom, but we really are reaching fundamental physics and engineering limits of what you can do with silicon. http://arstechnica.com/information-technology/2016/02/moores-law-really-is-dead-this-time/ Sorry dude.
|
# ? Sep 8, 2016 20:21 |
|
If you want a classical matter wave approach to the issue, note that the deBroglie wavelength of an electron with standard rest mass and 1eV of kinetic energy is 1.23nm.
|
# ? Sep 8, 2016 21:06 |
|
Rastor posted:It's not about size of an atom, but we really are reaching fundamental physics and engineering limits of what you can do with silicon. That says Intel is dropping silicon for 7nm. So I guess this means GloFlo and SS/TSMC will be doing the same? The Age of Silicon is basically over already?
|
# ? Sep 8, 2016 21:32 |
|
Pryor on Fire posted:I've heard or seen this at least 20 times since the 80s, guess what it's bullshit every time. Somehow we were approaching the "fundamental size limit of the atom" or some such utter horseshit back in 1989 and and every year since then. Somehow we manage. You do accept that there are physical limits that we're approaching though, right? I mean, the atom does have a defined size and we are approaching it, this is not a matter of opinion. The fact that we've always managed to transcend the limits that we saw ahead us before does not mean that the concept of a fundamental limit which cannot be transcended is necessarily invalid. I don't see us getting to atom-sized transistors and going "OK no problem, we'll make them out of unbound quarks next!" but maybe I'm too pessimistic. On the competition topic, Intel is its own competition at this point. I think they'd love to deliver large year-over-year performance gains at a given price/power point so they can sell more chips in new machines but they really can't. If we still had the kind of gains we were seeing 1993-2008 then I would have replaced my 2500K with a 5.5GHz 8-core by now. Eletriarnation fucked around with this message at 21:38 on Sep 8, 2016 |
# ? Sep 8, 2016 21:34 |
|
Eletriarnation posted:You do accept that there are physical limits that we're approaching though, right? I mean, the atom does have a defined size and we are approaching it, this is not a matter of opinion. The fact that we've always managed to transcend the limits that we saw ahead us before does not mean that the concept of a fundamental limit which cannot be transcended is necessarily invalid. I don't see us getting to atom-sized transistors and going "OK no problem, we'll make them out of unbound quarks next!" but maybe I'm too pessimistic. Ehhh silicon atoms are very small (quick googling tells me .11 nm), but that doesn't really matter. Quantum Tunneling ruins silicon transistors before atom size is ever the issue.
|
# ? Sep 8, 2016 21:50 |
For 450mm wafers, think about how tiling works. If you increase the area, you can fit more dies on it, and there is less wasted edge space since as the size of the die relative to the wafer decreases, a tiled circle of the dies fits closer to a circle.
|
|
# ? Sep 8, 2016 22:42 |
|
Watermelon Daiquiri posted:For 450mm wafers, think about how tiling works. If you increase the area, you can fit more dies on it, and there is less wasted edge space since as the size of the die relative to the wafer decreases, a tiled circle of the dies fits closer to a circle. So more functional dies per wafer, which means cost of dies goes down, which means more can be bought. The hurdle for 450mm wafers is just no one wants to foot the initial bill and cripple themselves, but pooling resources means a single foundry company can't claim the advantage just for themselves?
|
# ? Sep 8, 2016 22:54 |
Yeah, though tbh I'm not very up to date on all the different tool manufacturers and whether or not they have models capable of 450mm wafers. It could be that the reason no one has built a production 450mm fab is that applied materials or whoever hasn't finished the design for some critical tool. At the very least all the load port portions will need to be redesigned. And I'm not too sure how well pooling resources will work considering the abject secrecy these companies operate under. Who would operate the fabs? Would there need to be many different unit part process teams for every single interested party (which would be a giant clusterfuck lol)? How would you segregate everything? LOL case in point: http://seekingalpha.com/article/2329555-450mm-semiconductor-wafer-manufacturing-stalled-and-applied-materials-the-biggest-loser
|
|
# ? Sep 8, 2016 23:34 |
|
|
# ? May 21, 2024 13:19 |
|
Eletriarnation posted:On the competition topic, Intel is its own competition at this point. I think they'd love to deliver large year-over-year performance gains at a given price/power point so they can sell more chips in new machines but they really can't. If we still had the kind of gains we were seeing 1993-2008 then I would have replaced my 2500K with a 5.5GHz 8-core by now. They still do deliver pretty good performance gains for server CPUs, partially because that's a much more lucrative market and partially because server applications can more easily utilize lots of threads. The same could be said for laptops, a Sandy Bridge laptop is pretty outdated now but my 2600k is still going strong even paired with a Pascal Titan X. Trying to provide faster desktop CPUs doesn't make much sense as an economic priority and is also harder from a physics and computer architecture standpoint as both clock frequencies and IPC are becoming harder to improve. Since gaming has increasingly been focused on GPU performance anyways there are a number of factors contributing to Intel's reduced focus on providing more IPC and performance/$ for gaming enthusiasts. If AMD were still competitive they would certainly still be delivering more but since they are not and gaming CPUs are a niche market for them they have little economic incentive to really push hard for fast enthusiast gaming CPUs. MaxxBot fucked around with this message at 00:20 on Sep 9, 2016 |
# ? Sep 9, 2016 00:15 |