| Page 364 of 782 |
Przepraszam
VIP Member
Posts: 14857
Location: Poland. New York.
|
 Posted: Tue, 5th Apr 2016 17:36 Post subject: |
|
 |
| Stige wrote: | | tonizito wrote: | | Aaaaaand most people ignore the postings of fanderps... |
Or they can't just handle the facts? Like the future Pascal GPUs still lacking proper support for Async?  |
Async doesn't matter honestly. While it lacks in that department it will probably make up for it in other departments and performance will be most likely the same at end of the day. |
|
| Back to top |
|
|
tonizito
VIP Member
Posts: 52316
Location: Portugal, the shithole of Europe.
|
| Posted: Tue, 5th Apr 2016 17:44 Post subject: |
|
|
Maybe in your alternate reality that happened, in here I saw quite a few  to those nVidia news. From people who own nv cards.
Guess something else happened back in your dimension. Along with thinking the reason why some had issues with their 390 was because they got one of the broken ones. What's next, someone went to his house with an nDerpia hammer and broke his 390? 
"Now you'll learn that nvidia always wins, pwahahaha!" *clonk* *clonk*
| boundle (thoughts on cracking AITD) wrote: | | i guess thouth if without a legit key the installation was rolling back we are all fucking then |
Last edited by tonizito on Tue, 5th Apr 2016 17:48; edited 2 times in total
|
|
| Back to top |
|
|
Nui
VIP Member
Posts: 5720
Location: in a place with fluffy towels
|
| Posted: Tue, 5th Apr 2016 17:48 Post subject: |
|
|
| Przepraszam wrote: | | Async doesn't matter honestly. |
You know that how?
|
|
| Back to top |
|
|
Morphineus
VIP Member
Posts: 24934
Location: Sweden
|
| Posted: Tue, 5th Apr 2016 18:01 Post subject: |
|
|
| proekaan wrote: | You have to remember that most people only read the link description and not the article itself. And the link...well it mentioned the driver killing cards.  |
Most people in here ignore clickbait titles and sites... and especially the sad fanboy ramblings of Stige.
That doesn't mean we don't acknowledge troubles with drivers on both sides... and it is both sides. The 'benefits' of having both brands in my home running next to each other on a daily basis.  |
|
| Back to top |
|
|
|
|
| Posted: Tue, 5th Apr 2016 18:08 Post subject: |
|
|
|
|
|
| Back to top |
|
|
|
|
|
| Back to top |
|
|
Nui
VIP Member
Posts: 5720
Location: in a place with fluffy towels
|
| Posted: Tue, 5th Apr 2016 19:52 Post subject: |
|
|
So we have one example pro AsynCompute and one against. And as you say developers don't have much experience with DX12 yet... so we just dont know as far as I can tell.
Werelds still holds AsyncCompute in high regards from a theoretical standpoint (?) as far as I understand.
kogel mogel
|
|
| Back to top |
|
|
Przepraszam
VIP Member
Posts: 14857
Location: Poland. New York.
|
| Posted: Tue, 5th Apr 2016 19:55 Post subject: |
|
|
Meh, nothing on Pascal. Computex it is then.
|
|
| Back to top |
|
|
Werelds
Special Little Man
Posts: 15098
Location: 0100111001001100
|
| Posted: Tue, 5th Apr 2016 20:28 Post subject: |
|
|
Why did you guys expect anything on Pascal? Because WCCFTech said so?
Once more, had there been a presentation, it would've been about the architecture. GTC is not used for product launches and announcements. Do not expect the products before or even during the summer. It'll be Q3 if not Q4 I reckon; even if they make Q2, it'll be a smaller chip that none of you will give a fuck about anyway.
Micron announced that they'll ramp up production on GDDR5X mid-summer, WCCF claim Computex launch with GDDR5X. Computex will be held from May 31st - June 4th. Now, I don't know how summer works in WCCF's fantasy world or whatever planet you guys live on, but here in the northern hemisphere on planet Earth summer is from June 21st till September 20th.
Do the fucking math please.
Stop going for the clickbait.
| Przepraszam wrote: | | Async doesn't matter honestly. While it lacks in that department it will probably make up for it in other departments and performance will be most likely the same at end of the day. |
Read my post a few pages ago (I'm too lazy to go find it for you). It's not something they can "make up for" in other departments.
As for those other departments, the one hardware area where AMD have lagged behind a little (geometry, tessellation specifically, although that in particular is mostly by choice to save space) is exactly what Polaris increases massively.
|
|
| Back to top |
|
|
Przepraszam
VIP Member
Posts: 14857
Location: Poland. New York.
|
| Posted: Tue, 5th Apr 2016 20:42 Post subject: |
|
|
Good man, good man. Explaining stuff to us since dawn of the ages.
|
|
| Back to top |
|
|
Stige
Posts: 3579
Location: Finland
|
| Posted: Tue, 5th Apr 2016 23:18 Post subject: |
|
|
| Werelds wrote: | Why did you guys expect anything on Pascal? Because WCCFTech said so?
Once more, had there been a presentation, it would've been about the architecture. GTC is not used for product launches and announcements. Do not expect the products before or even during the summer. It'll be Q3 if not Q4 I reckon; even if they make Q2, it'll be a smaller chip that none of you will give a fuck about anyway.
Micron announced that they'll ramp up production on GDDR5X mid-summer, WCCF claim Computex launch with GDDR5X. Computex will be held from May 31st - June 4th. Now, I don't know how summer works in WCCF's fantasy world or whatever planet you guys live on, but here in the northern hemisphere on planet Earth summer is from June 21st till September 20th.
Do the fucking math please.
Stop going for the clickbait.
| Przepraszam wrote: | | Async doesn't matter honestly. While it lacks in that department it will probably make up for it in other departments and performance will be most likely the same at end of the day. |
Read my post a few pages ago (I'm too lazy to go find it for you). It's not something they can "make up for" in other departments.
As for those other departments, the one hardware area where AMD have lagged behind a little (geometry, tessellation specifically, although that in particular is mostly by choice to save space) is exactly what Polaris increases massively. |
So what is wrong with this WCCFTech site then? I have yet to receive an answer to that.
And I'm pretty sure 31.5. to 4.6. is summer, whole fuckin June is summer really. Unless you live in the arctic maybe. I wonder where you got that summer starts June 21st lol
|
|
| Back to top |
|
|
|
|
| Posted: Tue, 5th Apr 2016 23:29 Post subject: |
|
|
| Quote: | The summer solstice heralds the beginning of summer in the Northern Hemisphere.
In 2016, summer begins with the solstice on June 21 at 12:24 a.m. EDT.
The word solstice is from the Latin solstitium, from sol (sun) and stitium (to stop), reflecting the fact that the Sun appears to stop at this time (and again at the winter solstice). |
|
|
| Back to top |
|
|
|
|
| Posted: Tue, 5th Apr 2016 23:42 Post subject: |
|
|
theres a huge difference between calendarian and meteorological beginn (3 weeks, 1st of june vs 21th of june)
|
|
| Back to top |
|
|
tonizito
VIP Member
Posts: 52316
Location: Portugal, the shithole of Europe.
|
| Posted: Tue, 5th Apr 2016 23:55 Post subject: |
|
|
Doubt that there's any company out there, at least the ones related to computer hardware, that would ever use the "meteorological begin"(?) of whichever season as a date/time for anything.
| boundle (thoughts on cracking AITD) wrote: | | i guess thouth if without a legit key the installation was rolling back we are all fucking then |
|
|
| Back to top |
|
|
|
|
|
| Back to top |
|
|
Neon
VIP Member
Posts: 18941
Location: Poland
|
| Posted: Wed, 6th Apr 2016 10:51 Post subject: |
|
|
The most important bit:
| Quote: | | What all this means is that P100 may never be utilized in a mainstream consumer device. At best, I suspect we might see some new variant of Titan based off P100 in the future, but that could be a long way off. You see, even though Nvidia is spilling the beans on Tesla P100 today—or at least, some of the beans—and the chips are in volume production, Nvidia doesn't plan on full retail availability from OEMs until Q1'2017. |
Man, what a chance for AMD to release a great card this year.
|
|
| Back to top |
|
|
tonizito
VIP Member
Posts: 52316
Location: Portugal, the shithole of Europe.
|
| Posted: Wed, 6th Apr 2016 10:55 Post subject: |
|
|
*great cards
They already have improved their driver support, even if it's just by having more regular updates.
| boundle (thoughts on cracking AITD) wrote: | | i guess thouth if without a legit key the installation was rolling back we are all fucking then |
|
|
| Back to top |
|
|
vaifan1986
Posts: 4645
Location: Birthplace of the necktie.
|
|
| Back to top |
|
|
Werelds
Special Little Man
Posts: 15098
Location: 0100111001001100
|
| Posted: Wed, 6th Apr 2016 11:48 Post subject: |
|
|
| Stige wrote: | | So what is wrong with this WCCFTech site then? I have yet to receive an answer to that. |
The fact that 90% of their posts never pan out and are full of shit because they don't look at what they're posting at all?
That's how they end up with "May launch with GDDR5X!" when Micron won't be producing it in volume until mid-summer. Or how the initial Maxwell rumours had numbers that were just doubled Kepler numbers and they didn't pick up on that at all.
What they do is post every bit of bullshit that's posted on any forum and post it as a "rumour". Later they go in and edit old posts to add shit like "We don't think this is real, but here it is" once it's been disproven.
I know this for a fact, because I've seen people do this to them, not unlike how the hump got journos to pick up on some bullshit a while ago
| Przepraszam wrote: | | Good man, good man. Explaining stuff to us since dawn of the ages. |
http://www.nfohump.com/forum/viewtopic.php?p=2866403#2866403
There's my earlier post about AC btw. It's a wall though
| tonizito wrote: | | Doubt that there's any company out there, at least the ones related to computer hardware, that would ever use the "meteorological begin"(?) of whichever season as a date/time for anything. |
No, generally speaking "summer" actually refers to July-Sept for most of them. June isn't really taken into account.
The conference was never a secret. That's been held every year since 2009.
Here's what I said back in February:
| Werelds wrote: | | We don't know. We've got pretty much no information about Pascal right now and most likely won't until GTC in April, if even then. |
That's just how Nvidia roll, but WCCF expected some weird change in Nvidia's schedule. It's not me picking up on a rumour or making an educated guess. They've been doing it like this since 2009, GTC gives us small bits about an architecture, with a full briefing taking place in June or July, products launching in Q3/Q4. Fermi fucked that up, because while the architecture briefing _did_ take place in July 2009, the actual chip was so fucked that it was delayed by 8 months beyond its initial scheduled date. Aside from that though, Nvidia are quite predictable.
As for the Pascal details: not really, at least not what WCCF claimed. It's not a product launch, much less a consumer product launch. We do get some bits about the architecture now (see https://devblogs.nvidia.com/parallelforall/inside-pascal/ ), but that came as a quiet "aside" on the site and wasn't part of the GTC presentation. but I'll explain why even that doesn't tell us much about consumer chips. Most importantly, this GP100 chip will not make it to us consumers. There's going to be a GP102 or something like that for the desktop market.
This is a pure-bred HPC chip. The only other chip that is comparable is GK210; a chip that was based on GK110 but was engineered specifically for HPC. Bigger registers, more cache. This is the exact same thing, except that instead of it being an evolution of an existing chip, it has been engineered for HPC right from the start and you can tell by the layout.
First of all, there's obviously a shift back to FP64 (DP) performance, something Maxwell sorely lacks. This may very well make its way into the final consumer chip, but I honestly wouldn't count on it. As it stands, they are building a 610mm chip on a brand new node, using brand new memory...yields on this will be relatively bad 
Then there's that whole register story. Register size is not different from Maxwell or Kepler, but because they've made the SMs so much smaller it means that there are more registers overall with each being dedicated to a smaller number of shaders. They've done this by reorganising the layout; where a TPC used to have 1 SM it now has 2, with the register per SM not having changed one bit. For HPC this is great, as it means less time is spent moving data around outside the clusters. It makes it a lot more efficient. For graphics? Meh.
Other than that, they've been incredibly tight lipped about the other bits of the GPU. There is no word about scheduling, geometry or any of the other things that matter for graphics rendering. In fact, as it stands now, GP100 is a massive disappointment in terms of how interesting it is. They've added FP64 ALUs back in, given each TPC 2 SMs...and that's about it. Beyond that it's just a huge Maxwell chip. The block diagram is obviously only a high level overview, but as I said, we still don't know much about the inner bits.
The reorganisation may very well stay though. This puts it a lot closer to the granularity of GCN and should allow Nvidia to apply Hyper-Q much more effectively. At this point though, I'm inclined to say that they haven't changed shit about how any of that works either. Again, fine for HPC (as that's generally fixed latency work and scheduling is almost a non-issue), but for graphics with AC that poses challenges and Nvidia users will remain dependant on driver optimisations.
As for the performance, before everyone starts raging about how amazing it is just by looking at the theoretical FLOPS, don't read into this too much. Yes, it has 5304 FP64 GFLOPS and 10608 FP32 GFLOPS. While that sounds fucking mindblowing, it isn't. If AMD take Hawaii/Grenada and double it they can easily exceed those numbers; a double Grenada would do 5240/10480. Hell, let's say AMD were to take Fiji, reorganise it a bit, put the FP64 performance back in and then double it like Nvidia did here: Fiji now has 13900 FP32 GFLOPS; FP64 is at a 1:16 rate, so that's only 869 GFLOPS (Nvidia moved from 1:3 on GK110 to 1:2 on GP100; GM200 is at 1:32). So let's say they brought that back up to a 1:4 rate, That would put it at 3475 GFLOPS already. But wait, I've not taken clockspeed bumps from the new node into account. Let's say they would "only" get 1350 MHz instead of Nvidia's 1480; that would put FijiV2 at 18765 FP32 GFLOPS...and with a 1:4 rate that's 4691 FP64 GFLOPS.
It's a shame AMD don't have the cash to get into the HPC market (needs a massive cash influx because Nvidia dominate that right now), because they absolutely have the hardware.
And just to throw some oil on the fire; the DGX-1 boards they showed seemed to have real chips on them...but it seems about half the HBM2 stacks were fake, mismatching sizes and matte materials
So much for it being "in production"
| Neon wrote: | The most important bit:
| Quote: | | What all this means is that P100 may never be utilized in a mainstream consumer device. At best, I suspect we might see some new variant of Titan based off P100 in the future, but that could be a long way off. You see, even though Nvidia is spilling the beans on Tesla P100 today—or at least, some of the beans—and the chips are in volume production, Nvidia doesn't plan on full retail availability from OEMs until Q1'2017. |
Man, what a chance for AMD to release a great card this year. |
That's about GP100 being used in HPC products. Not directly related to a Pascal consumer GPU |
|
| Back to top |
|
|
Nui
VIP Member
Posts: 5720
Location: in a place with fluffy towels
|
|
| Back to top |
|
|
JBeckman
VIP Member
Posts: 35846
Location: Sweden
|
|
| Back to top |
|
|
proekaan
VIP Member
Posts: 3651
Location: Finland
|
|
| Back to top |
|
|
Stige
Posts: 3579
Location: Finland
|
| Posted: Thu, 7th Apr 2016 18:35 Post subject: |
|
|
| JBeckman wrote: | http://www.guru3d.com/news_story/rumor_amd_radeon_490_and_490x_polaris_graphics_cards_launch_end_of_june.html
More rumors about the upcoming GPU models, AMD Polaris this time with how they might start to appear in late June and might be showcased at the upcoming Computex event.
(AMD and Nvidia both will probably have stuff to showcase so that's probably not much of a rumor really, availability however seems a bit soon but we'll see.) |
Haven't the new Radeons had a launch date around there for a while now already?
|
|
| Back to top |
|
|
|
|
| Posted: Thu, 7th Apr 2016 19:10 Post subject: |
|
|
yes, already known for a while. though polaris will not be interesting. only vega could be nice. if nv releases pascal around the same date as amd polaris -> amd will lose a lot of money/customers
|
|
| Back to top |
|
|
proekaan
VIP Member
Posts: 3651
Location: Finland
|
| Posted: Thu, 7th Apr 2016 19:13 Post subject: |
|
|
Based on what?
We know nothing about pascal and polaris in terms of performance or effiency. :/
AMD Ryzen 9 7900X 4,7 GHz
Asrock X670E Steel Legend
G.Skill Trident Z5 32 GB DDR5 6400Mhz
Asus TUF RTX 4090 24 GB GDDR6X
NZXT Kraken Z73 RGB
Corsair HX1500i Platinum
NZXT H7 Flow
|
|
| Back to top |
|
|
|
|
|
| Back to top |
|
|
Frant
King's Bounty
Posts: 24973
Location: Your Mom
|
| Posted: Thu, 7th Apr 2016 19:52 Post subject: |
|
|
| Stige wrote: | | So what is wrong with this WCCFTech site then? I have yet to receive an answer to that. |
Because WCCFTech is a rumour-mill. Someone may have heard something by a desk clerk working for some corporation that work with some hardware manufacturer. This is spun, guesses are added and turned into something that seems factual when it's only a factoid. They mix and match facts with factoids and basically everything they publish has to be taken with a huge grain of salt since it's difficult to know what's fact and what's fiction.
FYI:
| Quote: | | The term factoid can in common usage mean either a false or spurious statement presented as a fact, as well as (according to Merriam Webster and the Oxford English Dictionary) a true, if brief or trivial item of news or information. The term was coined originally in 1973 as a neologism by American writer Norman Mailer to mean a "piece of information that becomes accepted as a fact even though it’s not actually true, or an invented fact believed to be true because it appears in print." |
Ph'nglui mglw'nafh Cthulhu R'lyeh wgah'nagl fhtagn!
Last edited by Frant on Thu, 7th Apr 2016 20:02; edited 1 time in total
|
|
| Back to top |
|
|
Frant
King's Bounty
Posts: 24973
Location: Your Mom
|
| Posted: Thu, 7th Apr 2016 19:57 Post subject: |
|
|
As for Polaris, it's the first release of the new GPU-core from AMD. The first cards will be mid-range (and if I remember correctly, for laptops as well). We'll have to wait a while for the performance cards to show up.
Polaris 11 and 10 each have several versions:
| Quote: |
Polaris 11
Greenland
Ellesmere
Baffin
Polaris 10
Weston
Banks |
The above list is somewhat obsolete. Greenland will be the flagship and is internally named "Vega10" (yes, it's kind of confusing) since they've decided to drop the island thingy and will name their GPU's after stars and stuff.
Ph'nglui mglw'nafh Cthulhu R'lyeh wgah'nagl fhtagn!
Last edited by Frant on Thu, 7th Apr 2016 20:00; edited 1 time in total
|
|
| Back to top |
|
|
couleur
[Moderator] Janitor
Posts: 15070
|
|
| Back to top |
|
|
|
|
|
| Back to top |
|
|
| Page 364 of 782 |
All times are GMT + 1 Hour |
