FaustianQ posted:All the work for Polaris 13 was done, they need only schedule production basically. They wasted money, they did not save any, and if the Vega reception is anything to go by they may lose more from that decision. Releasing Polaris 13 late in 2016 would have just been the better financial move, but I think AMD expected more form Vega, that it promised much better performance on paper and that with Vega 11 and Vega 12 (with Vega 11 likely expected to compete with Polaris 13 in this scenario) they'd only end up with an exceptionally short run of an expensive to build GPU. Vega just did not pan out at all and it's barely meeting the potential of Polaris 13 while consuming an insane amount of power to do so. With Vega for 7nm being entirely scrapped I can only assume this means AMD knows what's wrong with Vega and it is utterly unfixable. I am very very interested in an indepth takedown on what is going on inside Vega, I want someone to tell me if the brakes failed and how on this train wreck. I think they did expect much better performance from Vega, it would explain why they did that whole "Poor Volta" thing, they thought they were going to make something significantly faster than whatever Nvidia could do. As far as Polaris, maybe Vega was someone's pet project and Polaris was trashed because of that? Oh well, we'll find out years from now if we find out at all. Man, I wonder what's up with Raja going silent for the past few weeks, you'd expect him to be out there trying to do something, anything PR wise but he's just been silent, internal politics or something?
|
|
# ? Aug 29, 2017 10:18 |
|
|
# ? May 29, 2024 04:59 |
|
Finding out some time ago that 2816 shader Polaris was a thing before it got shelved and and shipped off exclusively to Bone Xs is pretty And wasn't it when yields on Polaris were improving rapidly? Holy poo poo
|
# ? Aug 29, 2017 10:40 |
|
Just as I made a post about where AMD expected actual revenue to come from for Vega - https://globenewswire.com/news-rele...-MI-Series.html If you're wondering where Vega is at all, well it looks like Apple and Baidu took them all. Maybe the earnings report will reveal a hilarious profit in GPUs for AMD regardless of the consumer disaster.
|
# ? Aug 29, 2017 11:28 |
|
The biggest gains in compute datacenters came from AMD APUs though. Sure, you need to buy more chips, but they're cheaper and *far* more power efficient than cpu+dgpu combos for most compute tasks due to some memory bullshit.
|
# ? Aug 29, 2017 11:36 |
|
FaustianQ posted:Just as I made a post about where AMD expected actual revenue to come from for Vega - https://globenewswire.com/news-rele...-MI-Series.html With all their data center stuff with Epyc, AMD is killing it, especially with Baidu: http://www.amd.com/en-us/press-releases/Pages/amd-epyc-datacenter-2017jun20.aspx
|
# ? Aug 29, 2017 11:40 |
|
China is the only place they're being sold atm, they're kind of dragging their feet getting it to general vendors
|
# ? Aug 29, 2017 12:20 |
|
Who was the guy at AMD who said discrete GPUs were going the way of the dodo a few years back? Did AMD actually neuter their GPU division because of what he said? It would explain a lot.
|
# ? Aug 29, 2017 13:06 |
|
yikes http://www.gamersnexus.net/game-bench/3037-destiny-2-gpu-benchmark-frametimes-graphics-optimization
|
# ? Aug 29, 2017 13:25 |
|
Hoooly poo poo, dude.
|
# ? Aug 29, 2017 13:43 |
|
repiv posted:yikes Lol, reminds me of a 660ti graph
|
# ? Aug 29, 2017 13:48 |
|
poor vega, so young and already has parkinson's
|
# ? Aug 29, 2017 13:49 |
|
Eh, that's so bad it's gotta be driver related.
|
# ? Aug 29, 2017 13:50 |
|
Oh man, and Ryzen's pretty drat good with frametimes. SAD!
|
# ? Aug 29, 2017 14:03 |
|
Looking at the data, the frametime results for "High" are what most would consider normal. Some setting in this application's "Highest" preset is causing Vega huge problems right now.
|
# ? Aug 29, 2017 14:03 |
|
Riflen posted:Looking at the data, the frametime results for "High" are what most would consider normal. Some setting in this application's "Highest" preset is causing Vega huge problems right now. Could it be related to the MSAA troubles that Vega has? I think that is one difference between the high and highest setting, SMAA vs MSAA
|
# ? Aug 29, 2017 14:07 |
|
Enos Cabell posted:Could it be related to the MSAA troubles that Vega has? I think that is one difference between the high and highest setting, SMAA vs MSAA Destiny 2s highest preset does enable MSAA, but GN changed it to FXAA for their tests because Bungie says the MSAA path is still a work in progress. So it's not that.
|
# ? Aug 29, 2017 14:10 |
|
It's probably one of the HBAO++++ things. It's a nvidia thing and it makes my 980Ti seem slow as a dead horse and outputs all the loving heat if I turn it on, and there's not even any perceptible difference between it and SSAO or whatever it's called.
|
# ? Aug 29, 2017 14:12 |
|
Truga posted:poor vega, so young and already has parkinson's Nah, that's just the DTs. Drinking to forgets its launch. Or existence.
|
# ? Aug 29, 2017 14:52 |
|
Truga posted:It's probably one of the HBAO++++ things. It's a nvidia thing and it makes my 980Ti seem slow as a dead horse and outputs all the loving heat if I turn it on, and there's not even any perceptible difference between it and SSAO or whatever it's called. Hmm, the highest AO setting is called "3D" which implies it's a world-space AO technique akin to Nvidia's VXAO. Bungie haven't said which algorithm 3DAO is based on, but if it's voxel based like VXAO then Vegas geometry bottleneck might be the problem. Fast camera movement or a large number of moving objects would spike the amount of re-voxelization required per frame, which could push Vegas geometry processor over the edge and cause a stutter. That would explain the lulls where Vegas frametimes become smooth, if those are points where the scene is mostly static. repiv fucked around with this message at 15:26 on Aug 29, 2017 |
# ? Aug 29, 2017 14:54 |
|
1gnoirents posted:Lol, reminds me of a 660ti graph I'm still running one of these. repiv posted:Hmm, the highest AO setting is called "3D" which implies it's a world-space AO technique akin to Nvidia's VXAO. Bungie haven't said which algorithm 3DAO is based on, but if it's voxel based like VXAO then Vegas geometry bottleneck might be the problem. This was my assumption. They did note that 3DAO and DOF settings had the largest impact for an RX 580. kirtar fucked around with this message at 17:17 on Aug 29, 2017 |
# ? Aug 29, 2017 16:49 |
|
AVeryLargeRadish posted:Man, I wonder what's up with Raja going silent for the past few weeks, you'd expect him to be out there trying to do something, anything PR wise but he's just been silent, internal politics or something? Like I said I can't imagine the boardroom is super happy with how this launch went, RTG can't keep stumbling through launch after launch. Their competitiveness is dropping further and further, by the time Vega is actually available in non-paper-launch quantities with actual working drivers it may be going up against Volta. And this time they can't even dump it in the x60 market due to excessive manufacturing costs. The silence from Raja this past month has been notable and at this point I wouldn't be surprised to hear it's because he was attempting to justify his continued tenure as head of RTG. Navi is going to be a long time off, and if he whiffs again where does that leave AMD? I just don't see Raja getting more than one additional shot at the brass ring here, and even that one seems iffy to me. (on the "paper launch" note: on top of the report of "a few hundred units for the North America region", some people who backordered Vega 56 from Amazon yesterday are showing October ship dates) Paul MaudDib fucked around with this message at 18:01 on Aug 29, 2017 |
# ? Aug 29, 2017 17:42 |
|
FaustianQ posted:All the work for Polaris 13 was done, they need only schedule production basically. They wasted money, they did not save any, and if the Vega reception is anything to go by they may lose more from that decision. Releasing Polaris 13 late in 2016 would have just been the better financial move, but I think AMD expected more form Vega, that it promised much better performance on paper and that with Vega 11 and Vega 12 (with Vega 11 likely expected to compete with Polaris 13 in this scenario) they'd only end up with an exceptionally short run of an expensive to build GPU. Vega just did not pan out at all and it's barely meeting the potential of Polaris 13 while consuming an insane amount of power to do so. With Vega for 7nm being entirely scrapped I can only assume this means AMD knows what's wrong with Vega and it is utterly unfixable. I am very very interested in an indepth takedown on what is going on inside Vega, I want someone to tell me if the brakes failed and how on this train wreck. [desire to know more intensifies] Can you source this, or at least make a plausible argument for Polaris 13's existence? Also, can you source 7nm Vega being off the roadmap? I'm a little dubious on the former. XB1X is a semi-custom SOC and the fact that XB1X has a bigger-than-P10 GPU doesn't prove a bigger-than-P10 standalone mask exists. On the other hand I don't think scaling GCN is as difficult as some people think it is, I don't fully understand why 4x 1024-core partitions would be some hard limit. Nothing is ever as easy as "just stamp out another shader engine" in practice but the engineering problems involved seem straightforward. If the latter is true then holy poo poo, I've been musing about whether Vega is actually a solid enough base to build Navi on, and dropping Vega derivatives from the roadmap would be a massive red flag that AMD has absolutely no confidence in Vega. I agree with the general thrust of your argument though. Around the end of Q1 2016 was when AMD dropped Fat Polaris, so I'd say at that stage they must have been confident that Vega would be releasing in Q4 2016, and that it would be competitive. Then it whiffed hardcore on performance, they probably expected it to be 30-50% faster than it actually was. I don't tend to weight the whole "poor volta" thing too seriously, I think it was a throwaway jab from marketing rather than a serious attempt to telegraph that they were going to smoke Volta, but I would say that it probably shows they still had at least some confidence as late as the beginning of Q1 2017. By the start of Q2 they obviously knew it was going to be a dud and went radio silence, then lately of course we've gotten the blind-taste-test and other bullshit. Paul MaudDib fucked around with this message at 18:25 on Aug 29, 2017 |
# ? Aug 29, 2017 17:53 |
|
Paul MaudDib posted:Also, can you source 7nm Vega being off the roadmap? Early leaks mentioned a 7nm Vega20 coming in 2H 2018, but the official roadmap from a few months ago only slates Vega for 14nm and 14nm+. Presumably Vega20 was abandoned.
|
# ? Aug 29, 2017 18:00 |
|
Oh my, dirty Vega
|
# ? Aug 29, 2017 18:06 |
|
Not what you're thinking, they blurred out the connector for the external power supply
|
# ? Aug 29, 2017 18:08 |
|
Also, lol that they were planning to make three 64CU chips in a row. "4096 shaders ought to be enough for anybody" - Raja, probably
|
# ? Aug 29, 2017 18:18 |
|
Destiny 2 card comparison from Youtube: Well with a few lowered settings my 980 ti can play @1440 at 60 fps. At 1080p midrange users in the 970/580/1060 range look alright. Green Gloves fucked around with this message at 18:26 on Aug 29, 2017 |
# ? Aug 29, 2017 18:19 |
|
Who the hell ordered the cards on those graphs? What a shitshow.
|
# ? Aug 29, 2017 18:40 |
|
Why don't they use a horizontal bar chart in desc order
|
# ? Aug 29, 2017 18:46 |
|
CFox posted:Who the hell ordered the cards on those graphs? What a shitshow. Too busy rendering the cosmic reflections
|
# ? Aug 29, 2017 18:47 |
|
Yeah they look super hosed up. Kids in basic excel classes can do a better job with graphics lol. The ones on GamersNexus look much better. But hey at least its good for comparison purposes
|
# ? Aug 29, 2017 19:04 |
|
https://twitter.com/gchip/status/902578993920073728 Seems legit edit: https://twitter.com/cmurray201720/status/902609740693430278 Paul MaudDib fucked around with this message at 20:25 on Aug 29, 2017 |
# ? Aug 29, 2017 20:20 |
|
last seen in a car steered by Nvidia Pascal
eames fucked around with this message at 20:53 on Aug 29, 2017 |
# ? Aug 29, 2017 20:28 |
|
[ XP053D ]
|
# ? Aug 29, 2017 20:32 |
|
Paul MaudDib posted:https://twitter.com/gchip/status/902578993920073728 Oh wow.
|
# ? Aug 29, 2017 20:46 |
|
Paul MaudDib posted:[desire to know more intensifies] I'm using Polaris 13 to refer to fat Polaris considering Polaris 10, 11 and 12 codenames are taken up. Sorry for any confusion. Repiv sources my source of Vega being dropped just below your post. Also the issue with scaling GCN seems to be the 64 ROP barrier, not so much the shader limit. Adding more shaders past 3584 (possibly 3072) is quickly diminishing returns with 64 ROPs, and from what I can tell GCN can't just double the ROPs. Paul MaudDib posted:https://twitter.com/gchip/status/902578993920073728 There is no way Raja isn't in danger of losing his job. No one pulls such a hard duck and cover and have fellow employees risk their reputation on poo poo like this unless it's that serious.
|
# ? Aug 29, 2017 20:48 |
|
Oh no, no no no.
|
# ? Aug 29, 2017 20:52 |
|
Did I miss something, why do they feel the need to explain where the guy is again? Did he miss a meeting or something? Edit: Also I got "Baidu" confused with "Taobao", it's much funnier if all of Vega got snapped up to be resold on Taobao.
|
# ? Aug 29, 2017 21:00 |
|
It wouldn't be RTG without massive, stupid unforced errors.
|
# ? Aug 29, 2017 21:01 |
|
|
# ? May 29, 2024 04:59 |
|
Those charts were rear end, so I fixed them.
|
# ? Aug 29, 2017 21:20 |