|
fozzy fosbourne posted:http://arstechnica.com/gaming/2016/02/directx-12-amd-and-nvidia-gpus-finally-work-together-but-amd-still-has-the-lead/ Drawback of explicit AFR: Frame pacing is garbage unless the cards are exactly the same, has all the same limitations as implicit AFR (temporal effects are death) Drawback of explicit SFR: Nobody has figured out how to use it yet, and none of the big players are going to put much effort into it unless consoles go multi-GPU
|
# ? May 6, 2016 02:53 |
|
|
# ? May 30, 2024 19:04 |
|
THE DOG HOUSE posted:I'd assume a 480 should outperform a 970, I wouldn't worry too much about that. My other worry is will an R9 480 be able to chug along my 450W PSU like my 970 does swimmingly, or if I'll have to upgrade.
|
# ? May 6, 2016 02:57 |
I'm guessing SFR would give you an extremely weird and distracting screen tearing effect.
|
|
# ? May 6, 2016 03:00 |
|
repiv posted:Drawback of explicit AFR: Frame pacing is garbage unless the cards are exactly the same, has all the same limitations as implicit AFR (temporal effects are death) The selling point for SFR is to slice your frame longitudinally so you do things like build your gbuffer on one card while you are rendering your shadow maps on another, then sync the shadow map over to the main card for lighting while the "sidecar" card builds the next shadow map/reflection map/whatever. The problem is that the amount of data you need to transfer is fairly large and the latency going between GPUs will make pipeline flushes today look laughably insignificant by comparison. It's a sound idea, but it's hard to find work that doesn't expose the frame to a lot of potential dependency issues. Also yeah, most applications can't even get AFR right and it's a miracle of engineering and constant effort that SLi/Crossfire work at all to begin with.
|
# ? May 6, 2016 03:20 |
|
xthetenth posted:I'm really interested to see the clock speeds AMD gets.
|
# ? May 6, 2016 03:56 |
PC LOAD LETTER posted:That would be nice but I'm expecting it to have similar clocks and headroom vs current GCN chips. Would love to be wrong about that though. "Free" performance is always welcome especially if the only "cost" is heat and power which I don't care about much. I'd happily run watercooling and deal with the power bill if I could OC a Polaris 30%+ over stock reliably. Ultimately Polaris is a evolution of GCN 1.3/1.2 and I wouldn't expect massive differences because of that. Plus its a GPU, they tend not to overclock all that well, nothing like Sandy Bridge, AthlonXP 1700's or those old Celeron 300A's certainly which really let you blow the doors off stock performance. Maxwell would like a word. Anyway, overclocking all depends on yields, process maturity and where you set stock clocks in relation to your competition's performance so it is somewhat arbitrary. But yeah, I don't expect to see a ton of headroom for OCing on Polaris, both because it's still GCN and because of process immaturity. I just hope AMD's new cards have decent numbers of ROPs, lack of them keeps hurting them compared to Nvidia designs.
|
|
# ? May 6, 2016 04:29 |
|
https://twitter.com/dogrunes/status/728327374564007936?ref_src=twsrc%5Etfw https://twitter.com/comtar/status/728421385559670786
|
# ? May 6, 2016 04:47 |
|
Everyone is all excited for Pascal and I'm just sitting here waiting for news on Polaris so that I can decide if I want to grab it for FreeSync.
|
# ? May 6, 2016 04:51 |
|
Gonkish posted:Everyone is all excited for Pascal and I'm just sitting here waiting for news on Polaris so that I can decide if I want to grab it for FreeSync. I'm waiting on Vega because I'm an idiot, but unless you're top end price inelastic Freesync seems to be the much better overall platform. Having a fat wad of money to go up a bracket if needed makes vendor lock in a lot less painful. I think the Pascal thing is in large part sampling bias for people who get overexcited by giant chips, and partly just that the info is just around the corner.
|
# ? May 6, 2016 04:56 |
|
Yeah, I'm 99% sure that I'm going team red (currently team green) this generation just because my new monitor has Freesync. I am hoping AMD shows Polaris off during DICE's BF5 event.
|
# ? May 6, 2016 04:59 |
|
Nvidia for life!
|
# ? May 6, 2016 05:27 |
|
Vintersorg posted:Nvidia for life! In running out of inconspicuous places for this tattoo, need advice
|
# ? May 6, 2016 05:54 |
|
AVeryLargeRadish posted:Maxwell would like a word.
|
# ? May 6, 2016 06:45 |
|
Well if we're going that route then technically GCN is really great at overclocking, since it's actual stock clocks (according to AMD) are ~800-850mhz and easily overclocks to 1200-1250mhz. They have only has poo poo headroom now because you're pushing older, unrevised hardware to compete against heavily optimized GPUs by eating up that headroom at the OEM level. Also GCN seems pretty good at handling a wide voltage range without odd decreases in performance or wild increases in power consumption, it's just really too bad it can't be translated into higher clocks and performance.
|
# ? May 6, 2016 08:52 |
|
FaustianQ posted:Well if we're going that route then technically GCN is really great at overclocking, FaustianQ posted:Also GCN seems pretty good at handling a wide voltage range without odd decreases in performance or wild increases in power consumption, it's just really too bad it can't be translated into higher clocks and performance. edit:\/\/\/\/\/But we're not fighting over that at all? Don't think we're even really fighting here either, just disagreeing. PC LOAD LETTER fucked around with this message at 11:37 on May 6, 2016 |
# ? May 6, 2016 09:39 |
|
And GCN chugs power because it was an architecture targeted at 20nm that they lopped bits off of to make it fit on 28nm when TSMC shat the bed. I thought we already knew this, people! AMD elected not to-- or more likely didn't have the money to-- make a stopgap architecture (as Nvidia did with Maxwell) and just used what they had to limp along! Why are we still fighting over this in 2016?
|
# ? May 6, 2016 09:56 |
|
SwissArmyDruid posted:And GCN chugs power because it was an architecture targeted at 20nm that they lopped bits off of to make it fit on 28nm when TSMC shat the bed. I thought we already knew this, people! AMD elected not to-- or more likely didn't have the money to-- make a stopgap architecture (as Nvidia did with Maxwell) and just used what they had to limp along! we have always been at war with evga
|
# ? May 6, 2016 10:49 |
|
Gonkish posted:Yeah, I'm 99% sure that I'm going team red (currently team green) this generation just because my new monitor has Freesync. I am hoping AMD shows Polaris off during DICE's BF5 event. Same here. I just need the next card(s) to push 4k resolution + Freesync and I'm happy . Dual 970s are barely cutting it right now.
|
# ? May 6, 2016 11:59 |
|
PC LOAD LETTER posted:Huh? You got a odd definition of overclocking. Overclocking would mean going over 947Mhz for the 290 and over 1000Mhz for the 290X. Most people don't get much over 1100Mhz with GCN. Many can't even get that even with overvolting. 1200Mhz is about the top for most and few get to it even with overvolting and watercooling which is usually necessary. Its been that way for years now, since its introduction. The 390 and 390X didn't really change that unfortunately. Because GCN is just Hawaii? Not Bonaire, Cape Verde, Tahiti, Pitcairn, Oland, or Fiji? Pretty weird of you to reduce it down to just the 390 which as I pointed out, is basically just Hawaii pushed to it's limit already to keep up with Maxwell. All the headroom for GCN now is gone, but they were originally design around much lower core clocks. It'd be like claiming Maxwell is a poor overclocker compared to say Polaris because well, the 980ti just has no headroom when clocked to 1400mhz to keep up.
|
# ? May 6, 2016 12:06 |
|
Yep. Ever since the 7970 GHz GCN cards have generally basically shipped with a big factory OC above their sweet spot.
|
# ? May 6, 2016 13:23 |
|
It isn't just Hawaii of course. Those were just the 2 models I mentioned, mainly because I'm familiar with them and they're still very relevant today. Going over all variations is pointless when 2 will do to illustrate what I'm talking about. And since the 390/X are some of the latest implementations of GCN that are directly comparable to the 290/X (pretty much identical architecturally, just a minor boost in clock speed, more VRAM, and a better HSF) they're perfectly reasonable to bring up as well.FaustianQ posted:It'd be like claiming .... xthetenth posted:GCN cards have generally basically shipped with a big factory OC above their sweet spot. edit: its somewhat arbitrary but can be architecture dependent. It wasn't really a thing for AMD until GCN though and I think it was introduced more as a power save measure really but I can't remember where I saw that. edit2: or do you mean default clockspeed? "base clock" is a specific setting for GCN.\/\/\/\/ PC LOAD LETTER fucked around with this message at 14:09 on May 6, 2016 |
# ? May 6, 2016 13:49 |
|
What determines a "base clock" anyways? It seems like it's mostly synthetic rather than an intrinsic value of the card, right? Maxwell could have been much "worse at overclocking" if they sold it with a higher base clock? I'm just trying to figure out if base clocks have a large marketing factor to them or if there is some fundamental engineering formula that they use to establish them
|
# ? May 6, 2016 13:52 |
|
PC LOAD LETTER posted:I don't see how I'm saying anything remotely like that. I was essentially just pointing out that overclocking means going over stock clocks. That's kind of my point though? Like if you take the 300 series as "GCN", then yea I guess they have like at best 10-15% more headroom, but all the 300 series are is basically 2011-12 parts in prettier packaging, and I'm going off what it's original design parameters are, thus they actually had a lot of room starting at around 800-900mhz. That's also why I think it's relevant to bring up trying to sell Maxwell at near max clocks as OEM factory clocks to compete with newer cards, and then saying Maxwell v2 has poor headroom. Well no poo poo, it can reach those clocks but optimum design was much lower and that's what it should be judged on. fozzy fosbourne posted:What determines a "base clock" anyways? It seems like it's mostly synthetic rather than an intrinsic value of the card, right? Maxwell could have been much "worse at overclocking" if they sold it with a higher base clock? I'm just trying to figure out if base clocks have a large marketing factor to them or if there is some fundamental engineering formula that they use to establish them It's a combination of both, stock clocks and voltages are basically a sweet spot of perf/power for a given design but there is a fair degree of marketing in it, especially if you can afford it (best example, Maxwell).
|
# ? May 6, 2016 14:06 |
|
FaustianQ posted:Well no poo poo, it can reach those clocks but optimum design was much lower and that's what it should be judged on.
|
# ? May 6, 2016 14:18 |
|
https://www.youtube.com/watch?v=cccJhMVFrLs
|
# ? May 6, 2016 14:23 |
|
I'm running a 390 @ 1100 core with a -25mv undervolt but I'll probably lock it down to 1000 with a -100mv undervolt (max allowable in afterburner) when it goes into some other box after I upgrade. It really does just sip power and runs quite cool at that setting which, IIRC, 1000 mhz core is the OEM spec for a 390 anyways.
|
# ? May 6, 2016 14:26 |
|
http://www.hardocp.com/news/2016/05/06/h_in_austin_texas_nvidia#.VyyiKoQrKUk I'm wondering if they are actually getting hands on with them today.
|
# ? May 6, 2016 14:55 |
|
Blackfyre posted:http://www.hardocp.com/news/2016/05/06/h_in_austin_texas_nvidia#.VyyiKoQrKUk "Launched" ? Are they going to be for sale?
|
# ? May 6, 2016 15:37 |
|
It might be a paper launch, although you never know. They might try to pre-empt AMD.
|
# ? May 6, 2016 15:43 |
|
My pascal is ready edit: LOL Don Lapre fucked around with this message at 15:49 on May 6, 2016 |
# ? May 6, 2016 15:46 |
|
Don Lapre posted:
Is it going to be some sort of steampunk alternative history WW1? I...could get on board with that.
|
# ? May 6, 2016 15:49 |
|
Bardeh posted:Is it going to be some sort of steampunk alternative history WW1? I...could get on board with that. Battlefield: Wolfenstein
|
# ? May 6, 2016 15:50 |
|
Back to WW2 full circle
|
# ? May 6, 2016 15:59 |
|
I'd murder for a return of some of 2142's ideas, especially if it were tied into a hosed up steampunk WWII.
|
# ? May 6, 2016 16:01 |
|
Maybe it's set after a Trump presidency and the result of a drunken insult-filled chat with Putin nukes the entire planet back to the 1800s/early 1900s
|
# ? May 6, 2016 16:06 |
|
PC LOAD LETTER posted:But shouldn't products be compared or judged based on how they actually are and how they compare vs a competitor's products and not as they would be in a ideal situation that doesn't exist? I like to bat around "what if" scenarios and all too but ultimately the rubber must meet the road here. But I'm not talking product, I'm talking µarch? Bardeh posted:Is it going to be some sort of steampunk alternative history WW1? I...could get on board with that. Looks like a Sterling though, so definitely WW2 to early Cold War. I could also definitely do a Battlefield: Cold War set in 1950-1970, kind of like an expanded BF:V xthetenth posted:I'd murder for a return of some of 2142's ideas, especially if it were tied into a hosed up steampunk WWII. Literally Battlefield: Steel Battalion
|
# ? May 6, 2016 16:14 |
|
FaustianQ posted:
Looks like a to me
|
# ? May 6, 2016 16:20 |
|
FaustianQ posted:But I'm not talking product, I'm talking µarch?
|
# ? May 6, 2016 16:26 |
|
Don Lapre posted:Looks like a Yeah, that's a C96 in his hand, are you going to ignore the gun on his hip or what?
|
# ? May 6, 2016 16:31 |
|
|
# ? May 30, 2024 19:04 |
|
FaustianQ posted:Yeah, that's a C96 in his hand, are you going to ignore the gun on his hip or what?
|
# ? May 6, 2016 16:35 |