Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Post
  • Reply
Happy_Misanthrope
Aug 3, 2007

"I wanted to kill you, go to your funeral, and anyone who showed up to mourn you, I wanted to kill them too."

K8.0 posted:

It's much faster, often 50-100% faster. A 1660 Super is a much better value, though, so long as it will provide you with enough performance to keep you happy.

Yeah, the 1660Ti really doesn't make any sense with the super existing. Maybe pay $20 more for it over the super, tops.

Adbot
ADBOT LOVES YOU

Guigui
Jan 19, 2010
Winner of January '10 Lux Aeterna "Best 2010 Poster" Award

ItBreathes posted:

2nding the don't spend money on a DP-VGA adapter. If you can't find a cheap DP monitor get a second HDMI one, they're cheap, and a passive DP-HDMI adapter.

... True - Unless (as in my case) you still have a CRT monitor that only accepts VGA input.

(comedy option - Get a VGA to composite video adapter, then grab an old crt television with composite input, and hook that up to two 2080's. Take that setup to lan parties and watch heads explode...)

Sininu
Jan 8, 2014

Guigui posted:

... True - Unless (as in my case) you still have a CRT monitor that only accepts VGA input.

(comedy option - Get a VGA to composite video adapter, then grab an old crt television with composite input, and hook that up to two 2080's. Take that setup to lan parties and watch heads explode...)

https://www.youtube.com/watch?v=8ATutWS9FeM

Rosoboronexport
Jun 14, 2006

Get in the bath, baby!
Ramrod XTreme

Rosoboronexport posted:

I'm posting here for a bit of perspective and sanity check: I'm thinking of moving from my current monitor (Dell 2415, 1920x1200@60) to Samsung 27-inch 1440p@144 monitor. However, my current GPU is GTX 1060 6GB, CPU is mildly overclocked i7-3770k.

There are some GTX 1070Ti ex-miner cards being sold for 280 eur so I could buy one and sell the 1060 for around 100-120 eur while waiting for the Ampere cards to pop up. But then again I mostly play CSGO and Battletech, the first one should have no problem running at 1440p and the other I can play at under 20 fps.

Are there other GTX 1060 owners with 1440p displays, how do you manage with this combination, should I just skip updating the GPU? I'm aiming for upgrading the GPU when Cyberpunk is released in September.

Trip report: on the games I currently have installed, the resolution change has caused much less headaches than I expected. Dishonored 2 went from 60 fps to 40-50, which was the most noticeable one. CSGO runs as expected and apparently Battletech is CPU-limited even on OCd i7-3770k as I seem to get exactly the same FPS. Only surprise (that I didn't check) was that VP9 10-bit is not accelerated (Youtube HDR videos) by GTX 1060 so I get frame drops on HDR 60 fps content.

Dunno how badly new Doom will run or I might just postpone getting it until I upgrade the GPU when Cyberpunk is released.

defaultluser
Jan 13, 2007

The person can drink sake for the following five reasons. First of all, for the national holiday. Moreover, it fills with the nectar. Finally, for reasons. Next, to heal the dryness of the place. After that, to refuse the future
Fun Shoe

Rosoboronexport posted:

Trip report: on the games I currently have installed, the resolution change has caused much less headaches than I expected. Dishonored 2 went from 60 fps to 40-50, which was the most noticeable one. CSGO runs as expected and apparently Battletech is CPU-limited even on OCd i7-3770k as I seem to get exactly the same FPS. Only surprise (that I didn't check) was that VP9 10-bit is not accelerated (Youtube HDR videos) by GTX 1060 so I get frame drops on HDR 60 fps content.

Dunno how badly new Doom will run or I might just postpone getting it until I upgrade the GPU when Cyberpunk is released.

My 1060 supports VP9 playback on youtube. Nvidia has supported HDR10 VP9 since the GTX 960.

Are you sure it's not a driver bug?

And yes, I'm hoping I can wait until the end of the year before I upgrade - but I', still driving 1080p.

defaultluser fucked around with this message at 07:52 on Mar 2, 2020

Rosoboronexport
Jun 14, 2006

Get in the bath, baby!
Ramrod XTreme

defaultluser posted:

My 1060 supports VP9 playback on youtube. Nvidia has supported HDR10 VP9 since the GTX 960.

Are you sure it's not a driver bug?

And yes, I'm hoping I can wait until the end of the year before I upgrade - but I', still driving 1080p.

Well that's pretty strange as Nvidia themselves claim that the chip can't do that, only Turing, Volta & GTX1080Ti.

Only registered members can see post attachments!

Kazinsal
Dec 13, 2011

Rosoboronexport posted:

Well that's pretty strange as Nvidia themselves claim that the chip can't do that, only Turing, Volta & GTX1080Ti.



That matrix is showing a no for 10 and 12 bpc VP9. YouTube doesn't encode in anything above 8 bpc.

Rosoboronexport
Jun 14, 2006

Get in the bath, baby!
Ramrod XTreme

Kazinsal posted:

That matrix is showing a no for 10 and 12 bpc VP9. YouTube doesn't encode in anything above 8 bpc.

https://support.google.com/youtube/answer/7126552?hl=en

code:
Resolution 720p, 1080p, 1440p, 2160p
For best results, use UHD rather than DCI widths (for example, 3840x1600 instead of 4096x1716).
Frame rate 	       23.976, 24, 25, 29.97, 30, 48, 50, 59.94, 60 
Color depth 	       10 bits or 12 bits
Color primaries 	Rec. 2020
Color matrix 	Rec. 2020 non-constant luminance

repiv
Aug 13, 2009

https://videocardz.com/newz/nvidia-shifts-gtc-2020-into-online-event-due-to-coronavirus

GTC in-person conference is cancelled but the keynote will go ahead on the planned date as a web-only stream. Do they have something important to announce :thunk:

Paul MaudDib
May 3, 2006

TEAM NVIDIA:
FORUM POLICE

repiv posted:

https://videocardz.com/newz/nvidia-shifts-gtc-2020-into-online-event-due-to-coronavirus

GTC in-person conference is cancelled but the keynote will go ahead on the planned date as a web-only stream. Do they have something important to announce :thunk:

it's way early for consumer ampere (probably end of the year, maybe september, maybe maybe july) but I wouldn't be too surprised if we saw those datacenter cards that have been showing up for the last few weeks

OhFunny
Jun 26, 2013

EXTREMELY PISSED AT THE DNC

repiv posted:

https://videocardz.com/newz/nvidia-shifts-gtc-2020-into-online-event-due-to-coronavirus

GTC in-person conference is cancelled but the keynote will go ahead on the planned date as a web-only stream. Do they have something important to announce :thunk:

https://videocardz.com/newz/mysterious-nvidia-graphics-cards-with-108-and-118-compute-units-spotted

Maybe whatever these are?

Taima
Dec 31, 2006

tfw you're peeing next to someone in the lineup and they don't know

Paul MaudDib posted:

it's way early for consumer ampere (probably end of the year, maybe september, maybe maybe july) but I wouldn't be too surprised if we saw those datacenter cards that have been showing up for the last few weeks

Just out of curiosity, what is the basis for the general argument that it's too early for Ampere? Clearly no one knows, I'm not hating at all, I'm just trying to figure out what people mean by it being too early, because I hear it a lot.

I mean it's definitely the safe answer, but do we have supply chain info or something?

defaultluser
Jan 13, 2007

The person can drink sake for the following five reasons. First of all, for the national holiday. Moreover, it fills with the nectar. Finally, for reasons. Next, to heal the dryness of the place. After that, to refuse the future
Fun Shoe

Rosoboronexport posted:

https://support.google.com/youtube/answer/7126552?hl=en

code:
Resolution 720p, 1080p, 1440p, 2160p
For best results, use UHD rather than DCI widths (for example, 3840x1600 instead of 4096x1716).
Frame rate 	       23.976, 24, 25, 29.97, 30, 48, 50, 59.94, 60 
Color depth 	       10 bits or 12 bits
Color primaries 	Rec. 2020
Color matrix 	Rec. 2020 non-constant luminance

after testing a 4k HDR VP9 youtube stream successfully playing back on my GTX 960, I just assume d they continued that feature-set with Pascal. It works fine with Maxwell 2.0

But I only tested 8-bit playback on the 1060 (because my TV is the only HDR monitor in the house, and Windows has no way for force HDR output.) But it did accelerate 8k playback successfully.

Okay, I think I figured this out. My b Maxwell card doesn't support bt 2020 playback on youtune.

The Pascal cards added VP9 10/12bit support via a driver update.

http://us.download.nvidia.com/Windows/378.66/378.66-win10-win8-win7-desktop-release-notes.pdf

This is why earlier Pascal cards supported nothing, while later ones (GT 1030, GTX 1080 Ti) added support at time of release., and thus have the full set if boxes checked.

Just make sure you're using the latest drivers.

defaultluser fucked around with this message at 06:59 on Mar 3, 2020

DrDork
Dec 29, 2003
commanding officer of the Army of Dorkness

Taima posted:

Just out of curiosity, what is the basis for the general argument that it's too early for Ampere? Clearly no one knows, I'm not hating at all, I'm just trying to figure out what people mean by it being too early, because I hear it a lot.

IMHO not a whole lot, other than next-gen compute cards haven't been announced yet, and people presume that NVidia will favor compute cards for Ampere / 7nm first because they can likely up their profits off compute more than they can off consumers. There's also the assumption that between the next-gen consoles and phones all targeting 7nm already, fab slots for 7nm are likely at a premium right now, which may slow down development and production plans.

But date-wise, it's not insane for consumer cards to be coming soon.

780Ti -> 980Ti was 19 months
980Ti -> 1080Ti was 21 months
1080Ti -> 2080Ti was 18 months
2080Ti released 9/2018 + 18 months = 3/2020

Cygni
Nov 12, 2005

raring to post

Taima posted:

Just out of curiosity, what is the basis for the general argument that it's too early for Ampere? Clearly no one knows, I'm not hating at all, I'm just trying to figure out what people mean by it being too early, because I hear it a lot.

I mean it's definitely the safe answer, but do we have supply chain info or something?

There are a bunch of arguments. First is that AMD isn't pressing them at all and moving to a new arch/process would be more expensive than continuing to run Turing batches on an insanely profitable node.

Second is that Nvidia's datacenter market is becoming very, very profitable, and the primary part for that market is 2 1/2 years old. So it is likely the next target on Nvidia's list, especially with Big Navi supposedly directly targeting the V100.

Third is that Nvidia has consistently gone top down with new processes and architectures. High end parts come out first, then lower end parts later.

And last is that we have leaks of parts that are clearly for the data center from Nvidia, and havent seen much in the way of leaks about Nvidia's next gen consumer part. Board partners LOVE to leak that poo poo, so if they were imminent (as in the next few months), we would have heard something already. We don't even really know the codename!

Lotta arrows pointing one direction and not many pointing the other, but you never know.

Rosoboronexport
Jun 14, 2006

Get in the bath, baby!
Ramrod XTreme

defaultluser posted:

after testing a 4k HDR VP9 youtube stream successfully playing back on my GTX 960, I just assume d they continued that feature-set with Pascal. It works fine with Maxwell 2.0

Did it show the HDR logo instead of HD in lower right and in the "stats for nerds" was the colour output bt.709 or bt.2020? Because YouTube will play the videos in 8bpc if all of the requirements are not met. If everything is ok, the setting cog has HDR letters and colour output is bt.2020.

quote:

But I only tested 8-bit playback on the 1060 (because my TV is the only HDR monitor in the house, and Windows has no way for force HDR output.) But it did accelerate 8k playback successfully.

Again, I know that vp9 8bpc works. 10bpc shouldn't.

defaultluser
Jan 13, 2007

The person can drink sake for the following five reasons. First of all, for the national holiday. Moreover, it fills with the nectar. Finally, for reasons. Next, to heal the dryness of the place. After that, to refuse the future
Fun Shoe

Rosoboronexport posted:

Did it show the HDR logo instead of HD in lower right and in the "stats for nerds" was the colour output bt.709 or bt.2020? Because YouTube will play the videos in 8bpc if all of the requirements are not met. If everything is ok, the setting cog has HDR letters and colour output is bt.2020.


Again, I know that vp9 8bpc works. 10bpc shouldn't.



I just checked, and my GTX 960 is bt 709. definitely only 8-bit!

But Pascal got a driver update to enable VP9 10 and 12-bit playback (they were not working when thee cards shipped). So I have full confidence that if I hooked up the GTX 1060 to my HDR TV , it would playback bt 2020..

https://linustechtips.com/main/topic/738008-1012bit-hw-decoding-vp9hevc-finally-available-for-geforce-geforce-37866/0

I know for a fact that it plays back the bt 709 version, at 8k on my non-hdr monitor

defaultluser fucked around with this message at 08:00 on Mar 3, 2020

gradenko_2000
Oct 5, 2010

HELL SERPENT
Lipstick Apathy
Thanks for the advice earlier upthread: I'm now running on two monitors!

Rosoboronexport
Jun 14, 2006

Get in the bath, baby!
Ramrod XTreme

defaultluser posted:

I just checked, and my GTX 960 is bt 709. definitely only 8-bit!

But Pascal got a driver update to enable VP9 10 and 12-bit playback (they were not working when thee cards shipped). So I have full confidence that if I hooked up the GTX 1060 to my HDR TV , it would playback bt 2020..

Yes it will... but in software decoding mode, because GP106 does not support h/w decoding of VP9 10- or 12-bpc content. As I stated originally, I was just dropping frames because my processor can't handle HDR 1440p 60fps decoding. 50fps was doable barely with all cores pegged.

Edit: to illustrate my point:

60FPS SDR video, no frame drops and note video engine usage at ~20%, CPU at idle clocks. Output color is not reported but it's bt709.


60FPS HDR video, 0% video engine load and CPU pegged, frame drops.

Rosoboronexport fucked around with this message at 10:03 on Mar 3, 2020

Lockback
Sep 3, 2006

All days are nights to see till I see thee; and nights bright days when dreams do show me thee.

DrDork posted:


But date-wise, it's not insane for consumer cards to be coming soon.

780Ti -> 980Ti was 19 months
980Ti -> 1080Ti was 21 months
1080Ti -> 2080Ti was 18 months
2080Ti released 9/2018 + 18 months = 3/2020

I think the Super Refresh line will change that, since the 700,900,and 1000 series didn't have a true across the board +1 helmet upgrade that the 2000 series did. Adding 6 months to the 18 months will bring it more in line with estimates.

At the end of the day, no one knows. It could still be May/June but I am getting less and less optimistic on that. I think late summer/fall is more realistic and December is also probably more likely than Spring at this point.

defaultluser
Jan 13, 2007

The person can drink sake for the following five reasons. First of all, for the national holiday. Moreover, it fills with the nectar. Finally, for reasons. Next, to heal the dryness of the place. After that, to refuse the future
Fun Shoe

Rosoboronexport posted:

Yes it will... but in software decoding mode, because GP106 does not support h/w decoding of VP9 10- or 12-bpc content. As I stated originally, I was just dropping frames because my processor can't handle HDR 1440p 60fps decoding. 50fps was doable barely with all cores pegged.

Edit: to illustrate my point:

60FPS SDR video, no frame drops and note video engine usage at ~20%, CPU at idle clocks. Output color is not reported but it's bt709.


60FPS HDR video, 0% video engine load and CPU pegged, frame drops.



Gotcha.

So, unlike adding h.265 support to the GTX 980, they couldn't come-up with a shader-assist to make performance more realistic (for those of us with less than 16 threads?)

I was already thinking about picking-up a 1660 Super on-clearance for my HTPC system (to replace the 960), but this seals it! And I'll pick-up an entry-level Ampere RTX card to replace my 1060 sometime early next year.

defaultluser fucked around with this message at 16:27 on Mar 3, 2020

DrDork
Dec 29, 2003
commanding officer of the Army of Dorkness

Lockback posted:

I think the Super Refresh line will change that, since the 700,900,and 1000 series didn't have a true across the board +1 helmet upgrade that the 2000 series did. Adding 6 months to the 18 months will bring it more in line with estimates.

At the end of the day, no one knows. It could still be May/June but I am getting less and less optimistic on that. I think late summer/fall is more realistic and December is also probably more likely than Spring at this point.

Yeah, the whole Super thing does potentially throw a wrench in things. I think the biggest evidence so far that the 3xxx series isn't imminent is the lack of leaks: we've seen dribbles of what is clearly a datacenter card (sorry, no, they're not slapping 24GB VRAM on a consumer card), but nothing else. The best I could really see for the upcoming announcement would be the compute cards and maybe a Titan-style semi-stripped down super-expensive "prosumer" card with the actual 3xxx series to follow somewhat later. But NVidia has surprised before, so who knows?

Paul MaudDib
May 3, 2006

TEAM NVIDIA:
FORUM POLICE

Taima posted:

Just out of curiosity, what is the basis for the general argument that it's too early for Ampere? Clearly no one knows, I'm not hating at all, I'm just trying to figure out what people mean by it being too early, because I hear it a lot.

I mean it's definitely the safe answer, but do we have supply chain info or something?

Rumors had consumer Ampere taping out at the very end of the last year (Nov/dec) and it's typically one year-ish from tapeout to launch. Give or take a bit but I'd say we're looking at maybe a September launch. It makes sense that they might try to get it in stores before christmas and they have previously launched products (the 980/Maxwell launch) in that September window. Also, (wishfully), this will be the second batch of chips on the same uarch so maybe it will be a bit quicker than the normal 1 year. Otherwise if not September probably January.

There is an outside chance they could paper launch in the June/July timeframe for back-to-school, so that there is decent stock by Christmas, and to steal Big Navi's thunder. But a 6-7 month turnaround is fairly quick so I think that would be more to the paper launch side of things. Like more of a teaser.

As far as the 1 year rule of thumb - here is datacenter Ampere taping out 11 months ago exactly and now we're starting to see the run-up to launch. It is a reliable rule of thumb across brands and products. Which is why even September may be a bit wishful.
https://twitter.com/aeassa1/status/1112053410860290049
https://twitter.com/aeassa1/status/1112360141469478912
I don't have a citable source for the consumer tapeout date other than "just trust me" but it's a guy I know who is pretty in the loop who says people were telling him consumer taped out in Nov/Dec 2019.

Once you are a month or two out from launch, you will start seeing leaked coolers from partners, a few benchmarks start to appear, etc. So far we don't have those for the consumer cards. So I don't think they will launch with the datacenter cards. That leaves June/July as the very earliest possibility and more probably September or January timeframe. September is a possibility if they hurry a bit but if you are not launched by September then you don't have much product on shelves for christmas/black friday, so in that case the most likely scenario becomes a january launch.

Or, hell, maybe Coronavirus has hosed everything up and they may not be getting samples back until way later than expected and it's now january to march, or even later, who knows.

Lockback posted:

I think the Super Refresh line will change that, since the 700,900,and 1000 series didn't have a true across the board +1 helmet upgrade that the 2000 series did. Adding 6 months to the 18 months will bring it more in line with estimates.

At the end of the day, no one knows. It could still be May/June but I am getting less and less optimistic on that. I think late summer/fall is more realistic and December is also probably more likely than Spring at this point.

The 700 series was the "600 Super" of its day. Same thing but with various chips strategically moved down a segment to make everything a little faster.

Paul MaudDib fucked around with this message at 21:23 on Mar 3, 2020

Malcolm XML
Aug 8, 2009

I always knew it would end like this.

Paul MaudDib posted:

Rumors had consumer Ampere taping out at the very end of the last year (Nov/dec) and it's typically one year-ish from tapeout to launch. Give or take a bit but I'd say we're looking at maybe a September launch. It makes sense that they might try to get it in stores before christmas and they have previously launched products (the 980/Maxwell launch) in that September window. Also, (wishfully), this will be the second batch of chips on the same uarch so maybe it will be a bit quicker than the normal 1 year. Otherwise if not September probably January.

There is an outside chance they could paper launch in the June/July timeframe for back-to-school, so that there is decent stock by Christmas, and to steal Big Navi's thunder. But a 6-7 month turnaround is fairly quick so I think that would be more to the paper launch side of things. Like more of a teaser.

As far as the 1 year rule of thumb - here is datacenter Ampere taping out 11 months ago exactly and now we're starting to see the run-up to launch. It is a reliable rule of thumb across brands and products. Which is why even September may be a bit wishful.
https://twitter.com/aeassa1/status/1112053410860290049
https://twitter.com/aeassa1/status/1112360141469478912
I don't have a citable source for the consumer tapeout date other than "just trust me" but it's a guy I know who is pretty in the loop who says people were telling him consumer taped out in Nov/Dec 2019.

Once you are a month or two out from launch, you will start seeing leaked coolers from partners, a few benchmarks start to appear, etc. So far we don't have those for the consumer cards. So I don't think they will launch with the datacenter cards. That leaves June/July as the very earliest possibility and more probably September or January timeframe. September is a possibility if they hurry a bit but if you are not launched by September then you don't have much product on shelves for christmas/black friday, so in that case the most likely scenario becomes a january launch.

Or, hell, maybe Coronavirus has hosed everything up and they may not be getting samples back until way later than expected and it's now january to march, or even later, who knows.


The 700 series was the "600 Super" of its day. Same thing but with various chips strategically moved down a segment to make everything a little faster.

There is 0 need for nVidia to ship consumer cards, AMD is doing a good job of loving up Navi as-is. They can switch capacity to the $$$$ datacenter cards instead. If you had to sell an A100 (say) die for $10000 as a Tesla or $1200 as an RTX, you'd definitely do the former.

defaultluser
Jan 13, 2007

The person can drink sake for the following five reasons. First of all, for the national holiday. Moreover, it fills with the nectar. Finally, for reasons. Next, to heal the dryness of the place. After that, to refuse the future
Fun Shoe

Malcolm XML posted:

There is 0 need for nVidia to ship consumer cards, AMD is doing a good job of loving up Navi as-is. They can switch capacity to the $$$$ datacenter cards instead. If you had to sell an A100 (say) die for $10000 as a Tesla or $1200 as an RTX, you'd definitely do the former.

Since 40nm, ON A NEW NODE, Nvidia has always waited six-months to a year before launching the big dedicated compute die.

Kepler GK110 combined compute and graphics int one and only took 6 months. Pascal P102 took 4 months after the release of GP104, then 5 months for GP100, then another 3 months for V100(because the released a second intermediate step in Volta)

I would not expect a large die on-release day. So consumers-only! And as we're approaching the two-year mark (the time it took fro NVIDIA to replace their previous two consumer architectures) they really can't coax anything more out of these current chips.

defaultluser fucked around with this message at 08:32 on Mar 4, 2020

Cygni
Nov 12, 2005

raring to post

defaultluser posted:

Since 40nm, ON A NEW NODE, Nvidia has always waited six-months to a year before launching the big dedicated compute die.

Kepler GK110 combined compute and graphics int one and only took 6 months. Pascal P102 took 4 months after the release of GP104, then 5 months for GP100, then another 3 months for V100(because the released a second intermediate step in Volta)

I would not expect a large die on-release day. So consumers-only!

you know you cant will this thing into existence right

Barry Foster
Dec 24, 2007

What is going wrong with that one (face is longer than it should be)
Looking to get into 1440p/higher-than-60hz gaming, and have been sitting on my thumbs for awhile after upgrading my cpu from a i5-6600 to the 3600, waiting for what the end of March brings.

Was willing to wait for awhile to see what the RTX 3000 series looks like, but I'm not really willing to wait 'til this time next year or something, so probably going to go with a 2070 Super now. Any particular models to avoid? I'm looking to get a 3 fan model (shooting for lower noise with this PC build), is it worth it? First hour's research suggests to me the MSI RTX 2070 Super Gaming X Trio or the Gigabyte GeForce RTX 2070 Super Gaming OC 3x 8G. Any issues with these guys? Any better ideas?

Finally, it's probably too much to hope for, but if I do decide to flip this card and get (say) a RTX 3070 in 18 months or something, would it hold any value whatsoever?

TIA gpgoons

repiv
Aug 13, 2009

The Gaming X Trio is great but also loving *huge* so make sure it will actually fit in your case. I almost got it before realising it's too long for my Define C.

Malcolm XML
Aug 8, 2009

I always knew it would end like this.

Cygni posted:

you know you cant will this thing into existence right

He’s right they’ll use a smaller die as a pipe cleaner but it’ll likely be sold as a tesla

Theris
Oct 9, 2007

repiv posted:

The Gaming X Trio is great but also loving *huge* so make sure it will actually fit in your case. I almost got it before realising it's too long for my Define C.

EVGA makes one (XC Ultra, I think) that's two fans but three slots and it's pretty quiet. So that's an option if your case can fit thick but not long.

repiv
Aug 13, 2009

Theris posted:

EVGA makes one (XC Ultra, I think) that's two fans but three slots and it's pretty quiet. So that's an option if your case can fit thick but not long.

Yeah that's the one I ended up getting. It's pretty good besides the fans making a noise whenever they stop or start, which makes the 0rpm mode pointless, but it's so quiet when forced to idle at low rpm instead that it's not a big deal.

Cygni
Nov 12, 2005

raring to post

Malcolm XML posted:

He’s right they’ll use a smaller die as a pipe cleaner but it’ll likely be sold as a tesla

i dont think you need a pipe cleaner for a process thats been run at huge volume for two years, including with huge GPUs like the Radeon VII. but we will see!

TheFluff
Dec 13, 2006

FRIENDS, LISTEN TO ME
I AM A SEAGULL
OF WEALTH AND TASTE

Taima posted:

Just out of curiosity, what is the basis for the general argument that it's too early for Ampere? Clearly no one knows, I'm not hating at all, I'm just trying to figure out what people mean by it being too early, because I hear it a lot.

I mean it's definitely the safe answer, but do we have supply chain info or something?

Nvidia has never taken less than 23 months between new consumer microarch launches in the last 8 years, and it has been trending towards longer and longer between new microarchs:

TheFluff posted:

GTX 680/Kepler, 28nm: March 2012
GTX 780/Kepler, 28nm: May 2013 (14 months)
GTX 750/Maxwell, 28nm: February 2014 (23 months from Kepler)
GTX 980/Maxwell, 28nm: September 2014 (16 months from 780)
GTX 1080/Pascal, 16nm: May 2016 (20 months from 980, 27 months from Maxwell)
RTX 2080/Turing, 12nm: September 2018 (28 months)

680 and 780 were both Kepler, and Maxwell actually launched as early as in February 2014 with the GTX 750, and both Kepler and Maxwell were on the 28nm node.

Q3 or Q4 seems a reasonable assumption for new consumer cards, but we could very well be into 2021 before the 3000 series cards are widely available. The Supers are the usual mid-cycle refreshes. Ampere seems to me like it'll be a compute-only intermediate, like Volta. The consumer microarch might very well be called something else.

e: for once I agree with Paul, Q1 2021 doesn't seem unlikely either

TheFluff fucked around with this message at 19:30 on Mar 4, 2020

Cygni
Nov 12, 2005

raring to post

More evidence that the big part is coming first: its already up and running.. in October no less. 32gb of HBM2e. 7936 cuda cores. Looks to be a PCIe card and not a mezzanine, too.


https://browser.geekbench.com/v5/compute/206992


Cygni fucked around with this message at 22:48 on Mar 4, 2020

lDDQD
Apr 16, 2006
I feel like HBM is a dead giveaway that it's not a consumer part

E: also the amount of it

Paul MaudDib
May 3, 2006

TEAM NVIDIA:
FORUM POLICE

lDDQD posted:

I feel like HBM is a dead giveaway that it's not a consumer part

in case the "32GB" wasn't a tipoff

Gucci Loafers
May 20, 2006

Ask yourself, do you really want to talk to pair of really nice gaudy shoes?


TheFluff posted:

:thunk:

GTX 680/Kepler, 28nm: March 2012
GTX 780/Kepler, 28nm: May 2013 (14 months)
GTX 750/Maxwell, 28nm: February 2014 (23 months from Kepler)
GTX 980/Maxwell, 28nm: September 2014 (16 months from 780)
GTX 1080/Pascal, 16nm: May 2016 (20 months from 980, 27 months from Maxwell)
RTX 2080/Turing, 12nm: September 2018 (28 months)

:words:

This is good timeline but at the moment RDNA2 isn't out yet so why would nVidia bother releasing new cards? I'd just wait until right after it drops and use that as leverage.

K8.0
Feb 26, 2004

Her Majesty's 56th Regiment of Foot
Nvidia is not competing with AMD. They're competing with themselves. Most of their sales are to people who already own Nvidia GPUs. Putting out new, higher performance products makes them more money than trying to flog the same old crap.

Yaoi Gagarin
Feb 20, 2014

AMD, meanwhile, seems to be selling GPUs to the department of energy: https://www.anandtech.com/show/15581/el-capitan-supercomputer-detailed-amd-cpus-gpus-2-exaflops

Adbot
ADBOT LOVES YOU

Fabulousity
Dec 29, 2008

Number One I order you to take a number two.

SwissArmyDruid posted:

Just in time for the weekend. I'll install it on one of my 5700/XT machines and report back.

Any trip report on those new drivers?

I've eaten the black screen thrice so far with a Powercolor Red Dragon 5700: Once on the 4670K system it first inhabited somewhere in early November, twice on the 3700X system it got carried into since mid-December. All of them happened in Firefox just screwing around on the Interwebs with some sort of video element involved on the page. I popped the new drivers in yesterday and everything is fine but I still haven't enabled browser hardware acceleration. Just wondering how your machines are doing so far.

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply