|
It is all marketing anyway. Aint nobody in the engineering department using the term "RDNA", I can guarantee you.
|
#
?
Jul 5, 2019 22:41
|
|
|
|
| # ? May 20, 2024 06:08 |
|
|
ConanTheLibrarian posted:That's worse really because it means they tried and still couldn't produce the goods even with a node shrink. Kanter said it in the beginning of the video, AMD is required to provide backwards compatibility for consoles in a way that NVIDIA doesn’t need to. Maxwell optimized games won’t necessarily need to run at perfect efficiency on Turing but GCN titles need to run perfect on Navi. Everything in that video makes sense to me.
|
|
#
?
Jul 5, 2019 22:41
|
|
|
Maxwell was a different architecture from Kepler, but you can describe the specific changes within the SM units that were made. RDNA is a Kepler->Maxwell style improvement. It's not a clean-sheet new architecture in the sense of the change back to VLIW2 that some people were talking about, but it is "a new architecture" in the sense of incremental improvement. The fact that AMD has been calling incremental improvements on the base GCN design a "new architecture" muddies the water here. There really wasn't anything terribly different about Polaris and Tonga, or Tonga and Hawaii, but AMD pretended those were "new architectures". They sold Vega's new pipeline and new compute unit as big improvements too. I think the Kepler->Maxwell change is the most relevant comparison here. Kepler had some real oddities in how you needed to write code for it to achieve good occupancy. Maxwell moved to a new SMX configuration that drastically changed that. Same for RDNA. At the same time, it is obviously not completely new, otherwise it wouldn't be able to run in this "legacy mode" they've created for backwards compatibility with older GCN code (mostly for consoles). It is fundamentally a change in how GCN allocates its work units, and it can flip back to using the old mode if appropriate. (along with secondary improvements in memory compression, probably media core, etc etc) Paul MaudDib fucked around with this message at 23:07 on Jul 5, 2019 |
|
#
?
Jul 5, 2019 22:42
|
|
|
Paul MaudDib posted:RDNA is a Kepler->Maxwell style improvement. It's not a clean-sheet new architecture in the sense of the change back to VLIW2 that some people were talking about, but it is "a new architecture" in the sense of incremental improvement. RDNA halves wavefront width (with legacy 64-wide mode), matches SIMD width to wavefront size and ditches round-robin 4 cycle scheduling. How are you pulling Kepler->Maxwell from this?
|
|
#
?
Jul 5, 2019 23:17
|
|
|
eames posted:Kanter said it in the beginning of the video, AMD is required to provide backwards compatibility for consoles in a way that NVIDIA doesn’t need to. Maxwell optimized games won’t necessarily need to run at perfect efficiency on Turing but GCN titles need to run perfect on Navi. AMD doesn't need perfect efficiency in BC for consoles. They just need to match or exceed previous performance while providing BC. If this is their actual excuse, what they're essentially saying is that this really is the best they can do and anything new they designed would actually be worse. It's loving pathetic. The whole is it a new arch/is it not thing is a proxy discussion for something much more important : Nvidia keeps making meaningful steps forward, while AMD mostly doesn't. This step seems better than most of the ones they've taken, but it's still not doing anything to catch up. They need a clean sheet redesign simply because it's extremely obvious at this point that iterating on GCN will never allow them to compete with Nvidia. That said, with the price cuts and some solid bundles, these cards may actually be a reasonable midrange buy, which is something AMD hasn't had since the 290/390.
|
|
#
?
Jul 5, 2019 23:41
|
|
|
Nvidia has been only iterating the Kepler SMX for just about as long, the difference is they have been extracting meaningful IPC improvements out of it each time. Just look at benchmarks of the GTX 1080 vs the RTX 2070S. Those two cards are REALLY similar in terms of hardware resources for traditional rasterization, they are both 2560:160:64 configurations and even clock speeds aren't that far off other than the memory, but the 2070S holds 130+% the performance of the 1080. Sure we can complain about the prices, and the lack of competition allowing those prices, but you sure as hell can't say Nvidia is slacking on their development when they are consistently pulling out IPC improvements of that scale at similar power efficiencies every generation.
|
|
#
?
Jul 6, 2019 01:06
|
|
|
K8.0 posted:That said, with the price cuts and some solid bundles, these cards may actually be a reasonable midrange buy, which is something AMD hasn't had since the 290/390. The only acceptable bundle IMO would be $50-100 off a Ryzen 3000 CPU with Navi purchase. I'd say something too about the 290 being midrange but then I remembered it launched at $400.
|
|
#
?
Jul 6, 2019 01:16
|
|
|
Indiana_Krom posted:Nvidia has been only iterating the Kepler SMX for just about as long, the difference is they have been extracting meaningful IPC improvements out of it each time. Navi would be great at 200-300. But amd and nvidia have a duopoly so high pricing is both their interest Buying high end Turing is a tad foolish IMO since 7nm is immenent from nvidia is Samsung doesn't gently caress it up and they aren't intel
|
|
#
?
Jul 6, 2019 03:10
|
|
|
Malcolm XML posted:Navi would be great at 200-300. But amd and nvidia have a duopoly so high pricing is both their interest Yeah, I was discussing this with a friend elsewhere: 1. AMD is more content to go after the CPU market than the GPU market at the moment. It's more high profile than the GPU market. 2. AMD and Nvidia are duopoly, as stated. So right now, AMD isn't aiming to swipe Nvidia's market share, but where they do have a significant amount of share is for the console market, where 2/3 currently produced are (XBone, PS4; Switch is Nvidia), and those two will stay AMD for the next generation. So where they're lacking in market share for PCs, they're making it up elsewhere.
|
|
#
?
Jul 6, 2019 03:20
|
|
|
Or the perverse theory: AMD has no problem keeping GPU prices relatively high so poors will gently caress off to consoles where they are literally guaranteed all the money.
|
|
#
?
Jul 6, 2019 03:26
|
|
|
Pc should be a testbed for new amd gpu tech, so when new console gens come around they have mature gpus.
|
|
#
?
Jul 6, 2019 03:26
|
|
|
wargames posted:Pc should be a testbed for new amd gpu tech, so when new console gens come around they have mature gpus. You'd think but it took consoles to give us better threading in games, so...
|
|
#
?
Jul 6, 2019 03:32
|
|
|
Malcolm XML posted:Buying high end Turing is a tad foolish IMO since 7nm is immenent from nvidia is Samsung doesn't gently caress it up and they aren't intel Is Ampere really expected to be that big of an upgrade? Are there any credible rumors on this subject? I feel like this summer is a real golden moment for a new PC builder to take a huge leap forward on the CPU size with Zen2, but maybe it's a good idea to cheap out on the video card side because the golden moment for that industry is 2020? I had been thinking about doing a 2060 Super for my build in the next couple of months, but maybe I should drop it down to a 1660 or 1660 Ti and plan on replacing it with a high-end Ampere GPU next year...
|
|
#
?
Jul 6, 2019 04:04
|
|
|
Also, just because nVidia *could* launch their 7nm parts doesn't mean they will when they've *said* they will.
|
|
#
?
Jul 6, 2019 04:15
|
|
|
surf rock posted:Is Ampere really expected to be that big of an upgrade? Are there any credible rumors on this subject? It's only rumors based on powerpoint slides that they might have something new at some point called Ampere. It could be a year from now before it comes out if it's even a gaming part and not something datacenter-focused like Volta.
|
|
#
?
Jul 6, 2019 05:00
|
|
|
Shipped my card to EVGA earlier this week and already have their shipping notification for the 2070. Cannot wait to put it in and see Prey with the higher graphics settings. This 650 Ti needs to go back into retirement, thankful as I am to have had it around.
|
|
#
?
Jul 6, 2019 05:25
|
|
|
BIG HEADLINE posted:Also, just because nVidia *could* launch their 7nm parts doesn't mean they will when they've *said* they will. Turing is clearly a stopgap until whatever 7nm arch they have is ready. Turing is uncharacteristically hot and large for a mass market GPU, and it hasn't sold particularly well even with a dominant lead over the AMD. If only for cost/mm reductions alone, nVidia will switch to 7nm. I'd wait for a die-shrink, IMO. If you need a GPU, buy one now. If not, then wait.
|
|
#
?
Jul 6, 2019 05:42
|
|
|
K8.0 posted:If this is their actual excuse, what they're essentially saying is that this really is the best they can do and anything new they designed would actually be worse. It's loving pathetic. In the GPU market AMD is in the same situation that it was in back with the Bulldozer variants: struggling to tweak an old and mis-targeted architecture into something passable until they can get a new competitive clean sheet design done. Which takes years. Not as many years as it took for Zen to get finished but years all the same. Supposedly clean sheet GPU's take something like ~2-3 to complete. AMD started work on a new GPU back in early 2018 when they hired new talent and got enough money from Zen to really fund that development. So its waaaay to early to expect them to have something ready by now. Even late-ish 2020 is probably optimistic to expect something (after all, it'll only have been ~2.5yr by that point). Personally I think if they were willing to drop Navi prices to ~$300 max they'd be just fine financially and still would offer great value right now for the performance. They're solidly competitive in the mid-ish range vs what NV's offering right now and they'd sell well at those prices I bet. But with the GPU market being stupid as it is ever since Buttcoins hosed it all up + NV not willing to drop prices either its not too shocking to me that they'd try to keep prices close to what NV's comparative parts are selling for.
|
|
#
?
Jul 6, 2019 08:26
|
|
|
surf rock posted:Is Ampere really expected to be that big of an upgrade? Are there any credible rumors on this subject? Turing is pretty constrained by die area and power, especially since they added RT and tensor cores while staying on what is basically tweaked 16nm. They weren't able to pull a Maxwell and increase performance/watt substantially on the same node either. The move from 16/12nm to 7nm is pretty big both in terms of density and power efficiency, there is opportunity for some big gains compared to what we saw with Turing. The only catch is pricing, especially since AMD might not be offering much for competition at the time when Ampere is released. MaxxBot fucked around with this message at 08:44 on Jul 6, 2019 |
|
#
?
Jul 6, 2019 08:40
|
|
|
I hope that Intels GPU efforts will drive mid range prices down a bit next year but I admit that’s a stretch. Navi 10 seems alright for what it is. They currently can’t compete on design so they’re trying pull advantages out of (currently TSMC’s) process lead. 
|
|
#
?
Jul 6, 2019 08:41
|
|
|
Agreed posted:Cannot wait to put it in and see Prey with the higher graphics settings. You're going to have a blast!
|
|
#
?
Jul 6, 2019 09:22
|
|
|
Agreed posted:Shipped my card to EVGA earlier this week and already have their shipping notification for the 2070. Cannot wait to put it in and see Prey with the higher graphics settings. This 650 Ti needs to go back into retirement, thankful as I am to have had it around. Thinking about doing literally the same (650Ti buyers reprezent lol) and not rely on ampere to be the savior of gaming or whatever. Besides, if the chart from the article you linked here is at least in the ballpark, 7nm could wind up being the next Sony's '599 US dollars' meme. So tell me ITT how your 2070 works, cuz I am curious for myself. To others ITT...I assume SLI double-card stuff works well only with two similar/identical cards and not SLIing a Turing card with whatever Ampere winds up being? CrazyLoon fucked around with this message at 09:43 on Jul 6, 2019 |
|
#
?
Jul 6, 2019 09:31
|
|
|
We can be pretty sure now 7nm GPUs won't be that expensive. It just looked like it might be for a few days when AMD was trying to gouge fanboys for some relatively small GPUs. Nvidia will definitely make big value gains on 7nm, regardless of what the arch is called. SLI never works well, never has, and probably never will. It only works at all with cards that are more or less identical (i.e. 2 2080).
|
|
#
?
Jul 6, 2019 13:52
|
|
|
If the 2070 Super is as powerful as a 2080, I wonder if the 2080 Super is going to make the 2080 Ti redundant? There's no news of a 2080 Ti Super is there? 2070 Super SLI looking real tempting. Also, I keep saying 2070 Supra instead.
|
|
#
?
Jul 6, 2019 14:12
|
|
|
The 2080ti is still miles faster than the 2080super
|
|
#
?
Jul 6, 2019 14:16
|
|
|
Are other 5700 vendors doing their own tweaks or is it reference all the way?
|
|
#
?
Jul 6, 2019 14:29
|
|
|
iastudent posted:Are other 5700 vendors doing their own tweaks or is it reference all the way? Initial release is reference with customs coming soon
|
|
#
?
Jul 6, 2019 14:42
|
|
|
K8.0 posted:We can be pretty sure now 7nm GPUs won't be that expensive. AMD has gone for a much more moderate, almost small-ish by modern mid/high-ish performance GPU standards, die size for Navi which will help yields and prices a whole lot so I'd be careful making any extrapolations.
|
|
#
?
Jul 6, 2019 14:42
|
|
|
Ak Gara posted:If the 2070 Super is as powerful as a 2080, I wonder if the 2080 Super is going to make the 2080 Ti redundant? There's no news of a 2080 Ti Super is there? The super variants are generally high single digit to low single digits improvements, the 2080 super will still be 20ish percent slower than the 2080 ti based on these numbers. The 2080 ti already has a super version, it’s called the RTX Titan and you can snag one for $2500.
|
|
#
?
Jul 6, 2019 14:45
|
|
|
PC LOAD LETTER posted:It'll depend on the die size. If they go anywhere near a huge die on anyone's 7nm it'll be real expensive no matter what. Yeah, that is one of the bigger questions about Ampere. How much of the 7nm space savings are they putting in to increasing the transistor count vs how much they are putting in to smaller die sizes. I'm thinking they will probably want it to be somewhere in the middle ground; smaller than Turing but with a higher transistor budget. If they can wring another 10-15% IPC out of the SMXes while adding 15% more of them it would land them at a nice 25-30% generation over generation performance increase which is probably still doable on a smaller die size. Though room for 15% more SMXes while still aiming for a smaller die might be a problem if they also need more cache to hit the IPC target. An underwhelming option would be just to keep performance relatively flat compared to the previous generation, but just make it smaller, cheaper and more power efficient. Though that is probably unlikely because while AMD is pretty much off the current radar for another year or so, there is still Intel/Xe which Nvidia is probably watching very closely. It wouldn't surprise me to see Nvidia throwing some weight around in the next couple of years to just brute force attack anything Intel has planned. If you think about the possibility of Nvidia and Intel having an expensive slugging match at the high end, AMD just focusing on a relatively safe and affordable console space is probably not a bad idea.
|
|
#
?
Jul 6, 2019 15:34
|
|
|
Who would be the best vendor to get, say, the 2080S from when it's finally out?
|
|
#
?
Jul 6, 2019 17:07
|
|
|
CrazyLoon posted:To others ITT...I assume SLI double-card stuff works well only with two similar/identical cards and not SLIing a Turing card with whatever Ampere winds up being? SLI requires two of the same cards
|
|
#
?
Jul 6, 2019 18:10
|
|
|
I bought a 2070 from Amazon because I need some more oomph for VR but then found out that the super cards are coming out on tuesday, so I'm shipping it back. But now there could be even better cards soon? I currently have a 1070. Should I just try to go with the super, or would it be worth waiting for whatever new things come out?
|
|
#
?
Jul 6, 2019 18:52
|
|
|
Cojawfee posted:I bought a 2070 from Amazon because I need some more oomph for VR but then found out that the super cards are coming out on tuesday, so I'm shipping it back. But now there could be even better cards soon? I currently have a 1070. Should I just try to go with the super, or would it be worth waiting for whatever new things come out? The even better cards are still probably around a year away, buy what you need now.
|
|
#
?
Jul 6, 2019 18:53
|
|
|
There's always something new coming out depending on how long you want to wait. If you need a card now buy a card now and ignore the news until you need a new one.
|
|
#
?
Jul 6, 2019 18:55
|
|
|
Indiana_Krom posted:The even better cards are still probably around a year away, buy what you need now. Ah ok. I'm not trying to constantly wait for the new poo poo, but if it was coming out soon, I'd rather go for that since the prices of cards don't go down anymore. I guess I'll try to get a 2060 super or 2070 super.
|
|
#
?
Jul 6, 2019 19:01
|
|
|
Cojawfee posted:I bought a 2070 from Amazon because I need some more oomph for VR but then found out that the super cards are coming out on tuesday, so I'm shipping it back. But now there could be even better cards soon? I currently have a 1070. Should I just try to go with the super, or would it be worth waiting for whatever new things come out? A 1070 is no slouch at VR. What are you doing that needs more VR oomph currently? Elite? Project Cars? Or..... I guess getting an Index would need more oomph with the increased Res + Refresh Rate.
|
|
#
?
Jul 6, 2019 19:09
|
|
|
Yeah, I have an index coming, so I'm going to need something more. And iracing is already dropping down to half frame rate a bunch with just the rift. Plus I'm gaming at 1440p, so the 1070 struggles a bit with that. It still stays around 60, but I want more.
|
|
#
?
Jul 6, 2019 19:15
|
|
|
Cojawfee posted:Ah ok. I'm not trying to constantly wait for the new poo poo, but if it was coming out soon, I'd rather go for that since the prices of cards don't go down anymore. I guess I'll try to get a 2060 super or 2070 super. Personally the worry for me is that in a year's time shitcoiners could make a return and prices will skyrocket and cards go out of stock again, so yea...I wouldn't blame anyone that buys the supers this year due to this worry. One bird in the hand is worth two in the bush and all that.
|
|
#
?
Jul 6, 2019 19:31
|
|
|
|
| # ? May 20, 2024 06:08 |
|
|
Cojawfee posted:Ah ok. I'm not trying to constantly wait for the new poo poo, but if it was coming out soon, I'd rather go for that since the prices of cards don't go down anymore. I guess I'll try to get a 2060 super or 2070 super. I have a 2070 coming from the EVGA 700/900 thing and all my reading on it has prepared me for a pretty badass card for modern games, and it is basically identical in performance and drat near specs to the 2060S (which, incidentally, is the price it got discounted to under the EVGA promotional thing - I've said before, but if I had waited I would have got a 2060S as the perfect card for my budget).
|
|
#
?
Jul 6, 2019 19:56
|
|