|
repiv posted:SK Hynix Lists GDDR6 Memory as ‘Available Now’, Publishes Final Specs you mean 3.5 gig gddr non-x ampere 1170.
|
#
?
Feb 6, 2018 16:05
|
|
|
|
| # ? Jun 2, 2024 14:32 |
|
|
Fauxtool posted:Dont buy amd Worst comes to worst you end up flashing the BIOS back to stock on one. If you can flash a mobo BIOS and build your own PC that sort of thing shouldn't scare you at all.
|
|
#
?
Feb 6, 2018 16:25
|
|
|
Navi is apparently 6th generation GCN, new uArch coming 2020-2021 Time to go back to sleep for a few years unless Navi manages to strip a whole bunch of compute stuff out. https://wccftech.com/amd-new-major-gpu-architecture-to-succeed-gcn-by-2020-2021/
|
|
#
?
Feb 6, 2018 19:05
|
|
|
NewFatMike posted:Navi is apparently 6th generation GCN, new uArch coming 2020-2021 wonder if navi is going to try and do the multi-gpu die on a single chip via gpu infinity fabric?
|
|
#
?
Feb 6, 2018 19:10
|
|
|
wargames posted:wonder if navi is going to try and do the multi-gpu die on a single chip via gpu infinity fabric? as far as we know, it's a traditional monolithic design.
|
|
#
?
Feb 6, 2018 19:16
|
|
|
wargames posted:wonder if navi is going to try and do the multi-gpu die on a single chip via gpu infinity fabric? GCN cannot be architecturally scaled to a larger size than Fiji/Vega, and running an interconnect between chiplets will eat away some of the existing performance. So unless there are new tricks in Navi (more powerful units, etc, like Primitive Discard or NGG Fast Path) then you should expect MCM Vega to be no faster than monolithic Vega and probably somewhat slower. Cheaper to manufacture, perhaps. The alternative is a total redesign of the floorplan. Keep the ISA but throw away everything else and redesign it to let you cram in more units... which doesn't sound like something you'd do when you know you have a new uarch coming a year or two later (or the 2020 uarch is actually the redesign). The tagline for Navi on AMD's slides was "scalable architecture" but I think that's been called into serious question if AMD is getting a new macro-architecture after Navi. I just don't see how you scale GCN further as-is, and I don't see why you would do a big redesign and then throw it away. So either the "scalable redesign" is pushed back to 2020/2021 instead of happening with Navi, or the "scalable redesign" is gone entirely and AMD is doing a clean-sheet redesign and kick GCN to the curb. My expectations for Navi are reeeeallly modest at this point. I'd like to see geometry and other uncore running on its own clock, so that you could crank down the shader clocks and get power consumption down while maintaining 1070-ish performance without murdering geometry throughput. If you went with a MCM design on 7nm you could end up with a reasonably efficient 1160 competitor, maybe 1170 at highest. Other than that, I don't know where you really go forward from Vega with the GCN uarch. They've hit the limits, which is why Vega is such a derpy unbalanced uarch in the first place. They certainly aren't going to be competing at the high end, unless Volta is just a straight rebrand with no performance improvement. Paul MaudDib fucked around with this message at 03:47 on Feb 7, 2018 |
|
#
?
Feb 6, 2018 19:18
|
|
|
lDDQD posted:as far as we know, it's a traditional monolithic design. Then the question is how terrible will it be since in theory its the last raja influenced GPU.
|
|
#
?
Feb 6, 2018 19:19
|
|
|
wargames posted:Then the question is how terrible will it be since in theory its the last raja influenced GPU. RX 500 refresh. Maybe with a secondary rebrand of power-binned chips to try to suck out that last few million from cryptospergs. Calling it now.
|
|
#
?
Feb 6, 2018 19:24
|
|
|
GCN going from 28nm to 14nm yielded around a 35% clock boost. Any bets on clocks on 7nm?
|
|
#
?
Feb 6, 2018 19:27
|
|
|
Seamonster posted:GCN going from 28nm to 14nm yielded around a 35% clock boost. Any bets on clocks on 7nm? With the way things have been going lately, it'll be slower, use more power and have worse yields than any time previously.
|
|
#
?
Feb 6, 2018 19:34
|
|
|
Seamonster posted:GCN going from 28nm to 14nm yielded around a 35% clock boost. Any bets on clocks on 7nm? 10% but maybe better power.
|
|
#
?
Feb 6, 2018 19:35
|
|
|
ItBurns posted:This but arranged so that you can have a door open and they all spill out onto the floor and roll around. lmao
|
|
#
?
Feb 6, 2018 21:24
|
|
|
I'm curious how TSMC 10nm will compare to GF 7nm too. Names mean nothing but 7nm is supposedly a legit 10nm-class node and TSMC's 10nm sounds like a 16++. We'll see how yields go on GF 7nm, it sounds like they are deploying it on high-margin products first, which may not be a great sign on yields, then next year Ryzen 3.1 Gen 2 and Epyc will probably eat up a lot of volume. I wouldn't be surprised to see Navi on it, whatever Navi ends up actually being, but I dunno if we will see big monolithic GPU chips in the consumer market before the 2020/2021 chips. If those end up being MCM designs, maybe never. The thought at the time was that Vega and Volta might end up being the last big monolithic chips. But if they can make the yields work, AMD may actually have a bit of a process advantage on NVIDIA, assuming they can get Navi out in early 2019 for real instead of releasing on Raja Time. NVIDIA can probably respond with another shrink at some point but if they release this year they probably will ride it until 2020, which might give AMD a year or so of competitiveness in whatever markets they can manage to squeeze into. Purely talking out my rear end here though, process stuff is not really my area of expertise. Paul MaudDib fucked around with this message at 21:41 on Feb 6, 2018 |
|
#
?
Feb 6, 2018 21:30
|
|
|
Paul MaudDib posted:We'll see how yields go on GF 7nm, it sounds like they are deploying it on high-margin products first, which may not be a great sign on yields, then next year Ryzen 3.1 Gen 2 and Epyc will probably eat up a lot of volume. I wouldn't be surprised to see Navi on it, whatever Navi ends up actually being, but I dunno if we will see big monolithic GPU chips in the consumer market before the 2020/2021 chips. AMD already confirmed that Navi is on 7nm:  Vega refresh is too, but we might not see that used on consumer cards. According to the earlier leaks its Yet Another 64CU GCN Chip just with 1:2 FP64, quad HBM2 stacks and PCI-E 4.0 this time, most of which would be wasted on gaming.
|
|
#
?
Feb 6, 2018 21:52
|
|
|
I just impulse bought a 1080 ti for $800 from newegg (probably out of stock by now) I'm wondering whether to use it or flip it on ebay
|
|
#
?
Feb 6, 2018 22:46
|
|
|
Man I remember leading up to pascal launch, you could find brand new 980tis for $550 or less. I'm almost afraid to see what the 1180 launches at.
|
|
#
?
Feb 6, 2018 22:56
|
|
|
WhiskeyJuvenile posted:I just impulse bought a 1080 ti for $800 from newegg (probably out of stock by now) Yeah, flip on eBay for $1000 and gross $870 (minus shipping), I'm sure that's worth it...
|
|
#
?
Feb 6, 2018 22:57
|
|
|
guess I'll keep it then
|
|
#
?
Feb 6, 2018 23:02
|
|
Enos Cabell posted:Man I remember leading up to pascal launch, you could find brand new 980tis for $550 or less. I'm almost afraid to see what the 1180 launches at. I'm guessing between $600 and $800. If I had to go with a more specific price I would guess around $750. I'm not sure how much the MSRP matters, the cards are going to be in short supply at launch no matter what and the jacked up prices we have seen lately give retailers more latitude to jack up prices on the new cards. 
|
|
|
#
?
Feb 6, 2018 23:08
|
|
|
 Help, my schadenfreude boner is only getting harder!
|
|
#
?
Feb 6, 2018 23:32
|
|
|
SwissArmyDruid posted:
Is that an ad for 100 570s for 680 bucks?
|
|
#
?
Feb 6, 2018 23:47
|
|
|
It had better be, because there is nobody in the world that'll buy them at the alternative. Either way, they must have lost their rear end and need to recoup their losses right away.
|
|
#
?
Feb 6, 2018 23:50
|
|
|
Swords to Plowshares, Miners to Retailers
|
|
#
?
Feb 6, 2018 23:51
|
|
|
wargames posted:Is that an ad for 100 570s for 680 bucks? no, not yet. wait a couple more weeks. 
|
|
#
?
Feb 6, 2018 23:51
|
|
|
SwissArmyDruid posted:It had better be, because there is nobody in the world that'll buy them at the alternative. https://www.newegg.com/Product/Product.aspx?Item=N82E16814202282 Sadly it seems as if I read that add incorrect.
|
|
#
?
Feb 6, 2018 23:54
|
|
|
Next gen one thing is for sure, I'm gonna get a single card to keep but I'm still gonna get the highest end one there is short of Titan, because even mining at very low profit will still cover the electricity and price premium.
|
|
#
?
Feb 7, 2018 00:02
|
|
|
Lol poo poo I would definitely be mining if I had 100 cards but... I guess 570s might actually approach power efficiency issues (?). $675 a pop though you can suckkkkkkk it
|
|
#
?
Feb 7, 2018 00:15
|
|
|
Just imagine trying desperately to unload a closet full of 4GB 570's, one at a time, through craigslist and ebay classifieds 
|
|
#
?
Feb 7, 2018 00:31
|
|
|
BNIB huh  I just happened to have 100 of these lying around still in the boxes, imagine my surprise finding those in the linen closet let me tell you. There's been plenty of scams this round of "box-only" sales, along with selling flashed low-end cards as high-end models, but I wonder if anyone's tried selling 570 and 580 GTX cards to see if they can catch suckers looking for RX's. I guess 470 and 480 GTX cards would work for that too.
|
|
#
?
Feb 7, 2018 07:15
|
|
|
future ghost posted:BNIB huh That would be the long con, sit on a card for 8 years then dump it. You've earned it at that point.
|
|
#
?
Feb 7, 2018 08:40
|
|
|
Purgatory Glory posted:That would be the long con, sit on a card for 8 years then dump it. You've earned it at that point. Speaking of which, did anyone actually pull off that EVGA thing where you give them a specific ancient card and they give you $1000 store credit?
|
|
#
?
Feb 7, 2018 09:36
|
|
|
Zero VGS posted:Speaking of which, did anyone actually pull off that EVGA thing where you give them a specific ancient card and they give you $1000 store credit? I think you were put in a drawing for a chance to win a grand.
|
|
#
?
Feb 7, 2018 11:51
|
|
|
Nah, as I recall it was first lucky hoarder, first paid.
|
|
#
?
Feb 7, 2018 12:16
|
|
|
Is it reasonable to assume that the 1170 would be at about 1080Ti performance level? And launch at around $380-$400? That seems to be historically the case at least. I'm wondering what to do with my 1070 now, I got it about a month ago in Hong Kong which made it almost $200 cheaper than retail here so I could probably still flip it at a profit. Then be stuck for months with a 650Ti, and then maybe get a 1170 at an inflated price?? That would be basically a free upgrade but it's still way more than I'd like to spend...
|
|
#
?
Feb 7, 2018 13:10
|
|
|
The new x70 routinely performs at or a little better than the last generation's Ti, simply so the Ti owners feel compelled to upgrade.
|
|
#
?
Feb 7, 2018 13:13
|
|
|
I kind of have hope for Navi, I know fool's errand, but look at the recent Vega 24/28 MCM. The full chip makes use of 64 ROPs, and apparently has really impressive performance, so it's clear AMD did learn something from the rolling tire fire that is their chip design. Maybe they can focus on improving cache size and latency, fix the uncore, and just design around the weird ROP limitation? I don't think there is anything inherent to GCN that prevents them from beefing up the uncore and shaders themselves to produce an effectively more balanced design. Also, apparently first gen Navi is consumer, another good sign IMHO, with 7nm Vega being thier ML/data center card until 2020 apparently.
|
|
#
?
Feb 7, 2018 14:59
|
|
|
FaustianQ posted:I kind of have hope for Navi, I know fool's errand, but look at the recent Vega 24/28 MCM. The full chip makes use of 64 ROPs, and apparently has really impressive performance, so it's clear AMD did learn something from the rolling tire fire that is their chip design. Maybe they can focus on improving cache size and latency, fix the uncore, and just design around the weird ROP limitation? I don't think there is anything inherent to GCN that prevents them from beefing up the uncore and shaders themselves to produce an effectively more balanced design. 
|
|
#
?
Feb 7, 2018 15:32
|
|
|
Everyone has to have something to believe in, hopeless or not.
|
|
#
?
Feb 7, 2018 15:43
|
|
|
FaustianQ posted:I kind of have hope for Navi, I know fool's errand, but look at the recent Vega 24/28 MCM. The full chip makes use of 64 ROPs, and apparently has really impressive performance, so it's clear AMD did learn something from the rolling tire fire that is their chip design. Maybe they can focus on improving cache size and latency, fix the uncore, and just design around the weird ROP limitation? I don't think there is anything inherent to GCN that prevents them from beefing up the uncore and shaders themselves to produce an effectively more balanced design. (a) Intel probably specified the chip, since it's a semi-custom and all (b) Smaller designs have usually worked well for AMD (Hawaii was competitive, 480 was competitive), but this doesn't seem to extend to the ability to produce a decent larger chip. To be blunt about it, Vega M GH is Tonga, but with the Vega tweaks to the GCN uarch, and a few extra resources added on. It performs well but it's nothing particularly new, Scorpio Engine was larger and single-chip, and AMD has had something in this performance class since Tahiti. Fiji and Vega are maxed out on ROPs and geometry processors, AMD outlined those limits back in the Hawaii days and has never done the redesign it would take to break them. eg Anandtech from the Fiji launch. The fact that AMD relied on making the units more powerful instead of simply adding more units suggests that those limits still apply today. quote:These 64 CUs in turn are laid out in a manner consistent with past GCN designs, with AMD retaining their overall Shader Engine organization. Sub-dividing the GPU into four parts, each shader engine possesses 1 geometry unit, 1 rasterizer unit, 4 render backends (for a total of 16 ROPs), and finally, one-quarter of the CUs, or 16 CUs per shader engine. The CUs in turn continue to be organized in groups of 4, with each group sharing a 16KB L1 scalar cache and 32KB L1 instruction cache. Meanwhile since Fiji’s CU count is once again a multiple of 16, this also does away with Hawaii’s oddball group of 3 CUs at the tail-end of each shader engine. I'm sure AMD would do fine with something like Vega 32, after all it would be their third time producing something with that same basic layout, and like Vega M GH they could squeeze some extra performance by giving it the same geometry/ROP configuration as full Vega 10. It's the rest of the lineup that is iffy - GCN really isn't designed to handle more resources than Fiji/Vega in an architectural sense, so Vega 64 is the upper bound of performance you could get from it. MCM scaling (chiplets) doesn't change the underlying architectural issues that prevent GCN from being scaled up, and will actually cost some performance. Without a redesign to a more scalable layout, the options for "fixing" Vega are pretty limited. You could put the uncore (geometry processors, ROPs) on a separate clock so they can run at full speed while clocking down the shaders to save power. Or you can make the execution units more powerful somehow. And 7nm will be a little faster and cooler just by itself. But you're not going to get something drastically faster than Vega 10, probably just more efficient. Which is fine, but it'll be launching a year after Ampere again, so it'll likely fall into the same "it's an 1170 but a year late and burning too much power" trap that Vega did. I'm kinda doubting they are doing a whole new layout if they are replacing GCN after Navi. The tagline on the Navi slides was "scalable", at this point I think that's out the window, it wouldn't make sense to do a big redesign and then throw it away. It sounds like the 2020/2021 chip is now the "scalable" uarch and Navi is going to be whatever utility AMD can squeeze out of GCN until that happens. (but hey, it's WCCFtech, they'll report on a Reddit post if it generates clicks) Paul MaudDib fucked around with this message at 17:20 on Feb 7, 2018 |
|
#
?
Feb 7, 2018 16:18
|
|
|
|
| # ? Jun 2, 2024 14:32 |
|
|
Is there usually this little news about a new GPU generation within months of launch
|
|
#
?
Feb 7, 2018 17:11
|
|