|
Aaaaaaaaaarrrrrrrgh, this is typical AMD. Tech that looks promising, but you'd never actually touch until they work the bugs out a generation down the road!repiv posted:2x 480 isn't better than a 1080? Are you calling #Roy #Taylor a liar? I'm saying that Roy is holding out. There's something they're working on that they're not talking about, and it's probably the interposer-stitched dual-GPU board with a single logical interface that's been hypothesized, but that's so far off AMD will be broken up or bankrupt before it comes out. SwissArmyDruid fucked around with this message at 00:55 on Jul 9, 2016 |
#
?
Jul 9, 2016 00:52
|
|
|
|
| # ? Jun 4, 2024 23:35 |
|
|
Well, it could avoid the multigpu issue with a interposer, correct? Also, for TDP reasons, it'd make more sense to use cut Polaris 10 (two RX470s, not two RX480s). Achieving perfect scaling (99%) would put them consistently ahead of a 1080, for likely ~225W stock.
|
|
#
?
Jul 9, 2016 00:52
|
|
|
where did he get the $200+ cost saving from last I saw 2x 480 8gb is $500 vs $650 for a 1080 assuming you never find one for 600 also the min fps even at 4k is pretty bad for some of those games
|
|
#
?
Jul 9, 2016 00:56
|
|
|
Hahahaha
|
|
#
?
Jul 9, 2016 00:57
|
|
|
EdEddnEddy posted:Also I have been waiting for this Waiting for laptops like this to also include displays that push more pixels than DisplayPort 1.2a can handle, since 1070s/1080s support DP 1.3
|
|
#
?
Jul 9, 2016 00:59
|
|
|
Interposers are, from what I understand, hideously expensive. Now add together the cost of two GPUs and an expensive interposer and you are probably looking at a $700-$800 card. And that still assumes that you can truly stitch together two GPUs to work as one via an interposer, I'm not sure if that is true or not and if it is true I wonder how much R&D you would need to spend to get it all working as well as a single chip. It just seems like such a kludgy solution and makes me think AMD is pretty desperate at this point, I thought the 490 would just end up being the low end Vega product, not some dual GPU Frankenstein card, I'm glad this is just a vague rumor.
|
|
|
#
?
Jul 9, 2016 01:08
|
|
|
MeruFM posted:where did he get the $200+ cost saving from yes the very article roy links isn't exactly saying "if you bought a 1080, return it for crossfired 480's lol" Roy sucks. He gets "away" with so much poo poo. If some dweeb from nvidia acted like roy he'd be hung from a tree
|
|
#
?
Jul 9, 2016 01:29
|
|
|
AMDs made it quite clear they intend to go interposer/MCM with Navi, using Polaris as a testbed for new tech to iron out the inevitable issues makes sense, just as much as releasing Fury X with HBM. If you get it to work, it's not desperate it's a definite advantage. You basically maximize yield per wafer, and all the R&D done to get th Interposer working means less time spent working on various different chips. Nvidia releases a 1536CC part? Whip up an 2x1024 part. Oh, 2560CC? I guess we can just slap together a 3x1024SP part in very short order. You get to play easy counter every time they release a product, and if they're still going monolithic they'll not only hit a wall around ~600mm², but they'll deal with a massively higher defective die rate. Die size limitation wouldn't exist for such designs, beyond attrition rates being more favorable. It might as well be magic until it happens though, but it seems that's what AMD wants to pull off.
|
|
#
?
Jul 9, 2016 01:35
|
|
|
THE DOG HOUSE posted:yes the very article roy links isn't exactly saying "if you bought a 1080, return it for crossfired 480's lol" Why do you think Roy no longer works for NVIDIA? 
Hubis fucked around with this message at 01:42 on Jul 9, 2016 |
|
#
?
Jul 9, 2016 01:38
|
|
|
Hubis posted:Why do you think Roy doesn't work for NVIDIA anymore? Muscling in on JHH's schtick?
|
|
#
?
Jul 9, 2016 01:39
|
|
|
Hubis posted:Why do you think Roy doesn't work for NVIDIA anymore? Why does he even work for AMD? Banish him to the forgotten realms of VIA and Savage.
|
|
#
?
Jul 9, 2016 01:40
|
|
|
FaustianQ posted:Why does he even work for AMD? Banish him to the forgotten realms of VIA and Savage. Send him to Imagination.
|
|
#
?
Jul 9, 2016 01:43
|
|
|
xthetenth posted:Muscling in on JHH's schtick? There can be only one (black motorcycle jacket)
|
|
#
?
Jul 9, 2016 01:43
|
|
|
xthetenth posted:Send him to Imagination. Hey look man, Imagination has enough on their plate right now with Qualcomm products eating up their marketshare and Apple products taking a nosedive  (I am still amused that the most powerful mobile GPUs are Radeon derivatives).
|
|
#
?
Jul 9, 2016 01:47
|
|
|
FaustianQ posted:AMDs made it quite clear they intend to go interposer/MCM with Navi, using Polaris as a testbed for new tech to iron out the inevitable issues makes sense, just as much as releasing Fury X with HBM. If you get it to work, it's not desperate it's a definite advantage. You basically maximize yield per wafer, and all the R&D done to get th Interposer working means less time spent working on various different chips. Nvidia releases a 1536CC part? Whip up an 2x1024 part. Oh, 2560CC? I guess we can just slap together a 3x1024SP part in very short order. You get to play easy counter every time they release a product, and if they're still going monolithic they'll not only hit a wall around ~600mm², but they'll deal with a massively higher defective die rate. Die size limitation wouldn't exist for such designs, beyond attrition rates being more favorable. Especially if direct X 12 irons out the prior issues with multi GPU setups.
|
|
#
?
Jul 9, 2016 01:52
|
|
|
SwissArmyDruid posted:And they're all sporting DVI ports. You could actually put a full complement of DVI + 3xDP + HDMI on a single header slot, I think. You'd just need to go with MiniDP and Mini-HDMI instead of the full-sized versions. They convert fine with a passive adapter/cable. Mini-DP is about the size of the old USB Mini-B connector (not Micro-B), it's perfect and I wish there was more uptake of it.
|
|
#
?
Jul 9, 2016 01:55
|
|
|
Did... did you use the card without removing the giant "REMOVE BEFORE USE" sticker?   I mean, the alternative is that they saved the sticker and put it back on - not sure which is worse Paul MaudDib fucked around with this message at 02:06 on Jul 9, 2016 |
|
#
?
Jul 9, 2016 02:04
|
|
|
I had the 970 version of that edition of MSI, Golden. It was the best air cooled 970 only beaten by water cooling. That full copper block is amazing, but extremely heavy.
|
|
#
?
Jul 9, 2016 02:10
|
|
|
Paul MaudDib posted:Did... did you use the card without removing the giant "REMOVE BEFORE USE" sticker? I have a friend who likes to keep all the protective, adhesive plastic on his stuff because "fingerprints/dust/dirt/damage/it looks better/what if I want to return it?"
|
|
#
?
Jul 9, 2016 02:10
|
|
|
SlayVus posted:I had the 970 version of that edition of MSI, Golden. It was the best air cooled 970 only beaten by water cooling. That full copper block is amazing, but extremely heavy. If I didn't want Step Up I would have gotten the Golden Edition instead of a Classified for my 980 Ti. Amazing card/cooler and like $40 or $50 cheaper than the Classified. From what I remember it's on par with the G1 Gaming, Amp! Extreme, and Kingpin.
|
|
#
?
Jul 9, 2016 02:15
|
|
|
AVeryLargeRadish posted:I hate to say it, but that is a lot of ifs, especially the part about getting the chips to work together without crossfire related problems. That pic of the old bus they were going to try and do between GPU's to improve CF performance always made me wonder why they've never bothered to try and improve on that idea. Even if that implementation would've never worked well enough to bother with the basic concept sounds reasonable. In theory if you get a fast enough bus between the 2 GPU's you could do a pretty good job of faking 1 big GPU with minimal driver BS. It'd have to be a pretty fast and low latency bus though. FaustianQ posted:Well, it could avoid the multigpu issue with a interposer, correct PC LOAD LETTER fucked around with this message at 02:23 on Jul 9, 2016 |
|
#
?
Jul 9, 2016 02:20
|
|
|
I guess there wasn't much of a point in trying to learn how to OC my Gigabyte G1 1070. Valley (Pre-OC) FPS: 89.5 Score: 3745 Valley (Post-OC) FPS: 90.9 Score: 3802 Here are the details of the OC: Clock: 1595 -> 1700 Memory: 2002 -> 2292 Boost: 1785 -> 1890 Afterburner shows the clock at 2075 while running benchmark.
|
|
#
?
Jul 9, 2016 02:25
|
|
|
Paul MaudDib posted:You could actually put a full complement of DVI + 3xDP + HDMI on a single header slot, I think. You'd just need to go with MiniDP and Mini-HDMI instead of the full-sized versions. They convert fine with a passive adapter/cable. Same. I think you could actually use connectors rotated 90 degrees so the mini DP connectors are vertical and take up even LESS space to cram more connections in, but nooooo, everything's gotta have a DVI port. =X
|
|
#
?
Jul 9, 2016 02:26
|
|
|
KingKapalone posted:I guess there wasn't much of a point in trying to learn how to OC my Gigabyte G1 1070. this is fantastic to hear bc i really didn't want to expend the effort
|
|
#
?
Jul 9, 2016 02:28
|
|
|
Trying to sell an 480x2 for $400 would be an awful idea, the 1070 is superior in every possible way.
|
|
#
?
Jul 9, 2016 02:30
|
|
|
FaustianQ posted:I have a friend who likes to keep all the protective, adhesive plastic on his stuff because "fingerprints/dust/dirt/damage/it looks better/what if I want to return it?" Yeah that annoys me to no end but im not even sure why. The problem of course with this sticker is it covers up a whole fan... stickergate lol
|
|
#
?
Jul 9, 2016 02:30
|
|
|
KingKapalone posted:I guess there wasn't much of a point in trying to learn how to OC my Gigabyte G1 1070. My factory boost got up to 1975 as it was, while 100 mhz is never a joke its still ~so close~ now. I'd leave your memory OC though, memory speed is always good.
|
|
#
?
Jul 9, 2016 02:33
|
|
|
Phlegmish posted:I wonder why that is, they're a big company, they should have enough resources for good R&D. AMD grand "strategy" for the past decade was waiting for competitors to screw up except the fact Intel hasn't did since 2006, Nvidia since 2010 and both certainly ain't messing around anytime soon. And when both doesn't, milk the poo poo out of their old inherently inferior uarchs with perf/price alone because AMD has nothing else to compete in any other metric. Palladium fucked around with this message at 02:35 on Jul 9, 2016 |
|
#
?
Jul 9, 2016 02:33
|
|
|
Paul MaudDib posted:Did... did you use the card without removing the giant "REMOVE BEFORE USE" sticker? Well I guess I'm a freak then because I totally saved that sticker from my 1080 and will likely stick it back on when I sell it on SA Mart in 3 years or whatever
|
|
#
?
Jul 9, 2016 02:36
|
|
|
im trying to help someone upgrade their computer. They use autodesk and rhino for architecture stuff. They have a monstrous way too expensive pc at work but this is for home use so they can view the files and do some light work. They currently have a quadro K600 1GB and an i3 4360 which while fast isnt cutting it for multithread workloads. Im going to replace the i3 with an i7 but i dont know which consumer gpu to buy. They are paying for it out of pocket so im not looking for a pro solution. The recommended hardware off the autodesk site has gtx690 and titans. All of which are still expensive. Are there other options? Is there a list of which consumer gpus are also good for work applications? Their power supply is a 300w bronze by seasonic, its good but not very powerful. Not having to buy a new one would be nice, but still okay if needed. Fauxtool fucked around with this message at 02:46 on Jul 9, 2016 |
|
#
?
Jul 9, 2016 02:43
|
|
|
Fauxtool posted:im trying to help someone upgrade their computer. They use autodesk and rhino for architecture stuff. If you need something Kepler for compute, you could look at either the 780 Ti or the GTX Titan. Both are pretty affordable nowadays. The GTX Titan has more VRAM and supports double-precision compute, if he needs that it's the obviously best choice. Bear in mind we're talking about the original Titan, which goes for less than $400 nowadays. If you can get away with single-precision compute you have a lot more options. I would probably look for something like a GTX 970 or a 290/390 (do not get the reference blower cooler). On the high end, 1070. Also if he's doing that kind of productivity work, the HEDT processors like the 5820k may be more appropriate than something like the 4790k/6700k. e: Fauxtool posted:Their power supply is a 300w bronze by seasonic, its good but not very powerful. Not having to buy a new one would be nice, but still okay if needed. You are pretty much going to have to replace that. Even a 970 and a 6700k puts you uncomfortably close to 300W. That's a power supply that's unsuitable for anything but an iGPU, basically. If you don't want to replace it, you will be limited to something in the 750 Ti / 950 / 960 range. Paul MaudDib fucked around with this message at 02:51 on Jul 9, 2016 |
|
#
?
Jul 9, 2016 02:48
|
|
|
Paul MaudDib posted:If you need something Kepler for compute, you could look at either the 780 Ti or the GTX Titan. Both are pretty affordable nowadays. The GTX Titan has more VRAM and supports double-precision compute, if he needs that it's the obviously best choice. Bear in mind we're talking about the original Titan, which goes for less than $400 nowadays. they dont know if its single/double precision compute, how would I know? They gave me an autodesk cd-key and a sample file if that helps The mobo is H97 so i dont think K cpus are needed, but are you saying they have some unique function not related to overclocking? Fauxtool fucked around with this message at 02:55 on Jul 9, 2016 |
|
#
?
Jul 9, 2016 02:53
|
|
|
PC LOAD LETTER posted:Its a 'CF on a stick' card so the CF problems won't go away period. Likely it'll have a minor price premium on it vs 2x RX480's and won't sell well since most don't bother with multi GPU set ups period. They're probably just trying to get something out there for now that they can point to as plugging the gap between the 1070/1080 and the RX480. Actually, interposer-based chips don't necessarily need to be CF-on-a-stick. You could have the memory controller and other shared parts on a single die, with "compute dies" that consist largely/entirely of compute units. Depending on how they designed the RX480 they may be able to disable the control circuitry on one of the units and have the other "master" chip run it as a slave. The chips could run in the same memory space and so on. Also, the interposer is actually a fully-fledged chip on its own, and the chip containing the memory controller and other control electronics can actually be the interposer itself ("active interposer" concept).
|
|
#
?
Jul 9, 2016 02:54
|
|
|
Fauxtool posted:they dont know if its single/double precision compute, how would I know? If Autodesk are suggesting the GTX 690 then it must be a single precision workload - that card has extremely low double precision performance.
|
|
#
?
Jul 9, 2016 02:59
|
|
|
Paul MaudDib posted:Actually, interposer-based chips don't necessarily need to be CF-on-a-stick.
|
|
#
?
Jul 9, 2016 03:02
|
|
|
Paul MaudDib posted:Actually, interposer-based chips don't necessarily need to be CF-on-a-stick. You could have the memory controller and other shared parts on a single die, with "compute dies" that consist largely/entirely of compute units. Depending on how they designed the RX480 they may be able to disable the control circuitry on one of the units and have the other "master" chip run it as a slave. The chips could run in the same memory space and so on. You really get limited by the interface between the chips and the interposer, though.
|
|
#
?
Jul 9, 2016 03:04
|
|
|
Hubis posted:You really get limited by the interface between the chips and the interposer, though. Yeah, this sounds similar to the old way x86 systems were built, where the CPU(s) talked to the northbridge to access the memory. That made for a bottleneck after a certain point, especially with multiple processors, which is why they don't do that anymore and moved the memory controllers on-die.
|
|
#
?
Jul 9, 2016 03:32
|
|
|
Hubis posted:You really get limited by the interface between the chips and the interposer, though. It's a lot better than PCB traces, though. ...come to think of it, I wonder if one of the reasons Polaris is so hot is because it's designed to push outsized amounts of energy to communicate with other chips on the interposer. I have no evidence for that whatsoever but it would be funny if AMD was playing 3-dimensional chess all along. Or at least, 2.5D. 
|
|
#
?
Jul 9, 2016 03:33
|
|
|
It would be pretty nice if we could get a dumb good card out of nowhere. Right now, the manufacture cost of an rx480 is about $100-110 total per reference board for AMD, and likely selling on to AIBs at a $40-50 margin. If they could do two chips without crossfire bullshit and sell it at $300 msrp they'd all but keep the margin per chip selling to AIBs, but have a cheap thing performing up there with 1070.
|
|
#
?
Jul 9, 2016 03:45
|
|
|
|
| # ? Jun 4, 2024 23:35 |
|
|
Paul MaudDib posted:Did... did you use the card without removing the giant "REMOVE BEFORE USE" sticker? I'm going to take a guess that the seller bought the card and took a set of unboxing photos before using it, probably because he intended to resell it relatively quickly. Then he used the older photos in the listing because why not. Or not. Whatever. I don't really care. 
|
|
#
?
Jul 9, 2016 03:46
|
|
