Images are sometimes not shown due to bandwidth/network limitations. Refreshing the page usually helps.
You are currently reading a thread in /g/ - Technology
You are currently reading a thread in /g/ - Technology
Pascal with HBM2 not available until Aug 2016 - GDDR5X versions releasing April
Glad I didn't fall for the waiting meme
>>52624457
yeah enjoy paying over retail for an incremental upgrade until 2017 gamerbro
>>52624457
>implying they will even have anything to launch until september
nvidia is going to be late to the new node like always
>>52624881
>nvidia
>late
I mean, yes but it's not their fault. Tsmc is notoriously bad at delivering.
Why do you think AMD switched to glofo?
Remember when all the Nvidia shills tried to say they had priority access to HBM?
LOL
I'm sitting on an Ivy Bridge iGPU atm, so I might be tempted by an early AMD/Nvidia offering. How much will HBM2 matter over GDDR, anyway? There doesn't seem to be anything special about the Fury X.
High bandwidth memory is the worst name. In 10 years it won't be high bandwidth anymore. Why don't things get accurately named anymore?
>>52625303
The interface is scalable.
>>52624457
my gtx 980 is still performing fine.
So what's the actual real-world performance difference between the standard GDDR5 and HBM?
>not waiting 10 years so you can buy pascal dirt cheap and not falling for the early adopter meme
fucking faggots
>>52625529
>not waiting n years so you can buy x dirt cheap
made me chuckle tbqh
>>52624457
So are these supposed to be small Pascals with GDDR5X or big Pascal with GDDR5X with the HBM2 versions to follow later?
>>52625263
nvidia was always late even when both were using TSMC
>>52625488
HBM has faster transfer rates. 4 HBM modules deliver 512GB/s and the fastest GDDR5 vRAM delivers around ~350GB/s.
HBM2 is suppose to go up to 1TB/s but will probably only be in top end graphics cards.
>>52624457
is it just me or is that guy pretty jacked for being in the tech industry? i'm assuming he's some nvidia exec?
>>52625709
It's some guy from the crowd. He got to take the GPU home as a present if I remember correctly.
>>52624457
If I remember correctly, NV's Pascal is coming 2Q later than AMD's Polaris.
>>52625672
That's not what real world means.
Real world, it gets you a small percentage more fps
better than buying cards that came out in 2014
>muh 970
>>52625672
Interesting to note is GDDR5X is actually slower (448gb/s max) than even ancient HBM. Pascal is gonna flop HARD because of this. By the time Nvidia puts HBM2 cards on shelves, Polaris would have probably already sold out. Nvidia will be left to eat the crumbs left behind by AMD.
Shit don't look too good.
Isn't GDDR5X as fast as HBM anyway, the only reason why Nvidia wants HBM is for the power savings as they have stated in that conference they did months ago.
>>52626079
lolnope, gddr5x is actually slower than hbm
nvidia fucked up big time.
>>52626078
>Nvidia will be left to eat the crumbs left behind by AMD.
>implying Nvidia won't create new cake that will redefine future cakes (that AMD will be eating crumbs from)
yikes, are you really this delusional?
>>52626079
gddr5x is half the speed of hbm
>>52626107
Don't know how they could have fuck up when a GTX 980Ti with GDDR5 is faster than AMD flaship with HBM.
>>52626079
Almost as fast, but cheaper to implement eg. no interposer, reuses existing GDDR5 tech
>>52626124
It actually isn't. AMD is just bad with drivers. Recent ones BTFO the 980ti. But nvidiacucks don't want you to know that.
>all this confusion from nvidia
Is it all downhill? Also, what will the new Titan be called?
They already used:
- Titan X
- Titan Z
- Titan Black
>>52626147
it isnt it though unless you are talking about 4K which is at an unplayable state even with two cards.
Reminder that AMD believe 20FPS are playable.
>>52626147
not quite btfo, but fury x > 980ti at 4k
what nm is pascal? Can't believe people bought new cards on 28 nm
The gtx 780ti was the last card you should've bought, the GOAT
>>52626113
Bruh, AMD already has their dirty mitts on HBM2 and are probably already working on the successor to Polaris.
>>52626156
Nvidia Colossus, there is no more Titan.
>>52626178
What driver version is the fury X running on in that test?
>>52626179
>The gtx 780ti was the last card you should've bought
Any Ti card is high-value. In fact, they're the only smart buys if you're not a poorfag.
With the continued use of GDDR5 for the first batch of Pascals, it looks like 980 Ti will continue the trend.
Nvidia should just give every fucking card a 512-bit bus, not even the fucking GTX 980Ti has one.
Fucking AMD needs to step up their game or die already so somebody else can buy them.
>they never released a 970 Ti
>>52626219
says Catalyst 15.9.1 Beta.
more here http://www.techpowerup.com/reviews/Sapphire/R9_390_Nitro/5.html
>>52626214
>there is no more Titan.
>no source
I'm sorry, who crowned you king of nvidia?
>>52626246
That's a 980.
If you own a 970/980 and overclock you will notice how memory bandwidth limited those cards are >>52626235
Fucking Nvidia and their loss less color compression shit allows them to jew it up even more on the memory.
>>52624457
>buying 28nm in 2016
Full retard. Scratch that, DOUBLE retard.
>>52626304
The GDDR5X versions of pascal cards will be on 28nm? Holy shit, nvidia has gone insane.
>>52626328
Are you retarded anon?
>>52626304
>>52626328
Isn't Pascal supposed to be 14nm? Just because it's using GDDR5X doesn't mean anything about it being 28nm. Nvidia would be really stupid to release a 28nm product a couple months now before they release something 14nm and tarnish Pascal as a marketing term.
>>52626387
They done stupid shit like that in the past. I'm a little confident nvidia will fuckup pascal in some way. Pascal on 28nm with shitty gddr5x is possible.
>>52626387
I see no reason for Pascal or Polaris to be anything other than a full 1xnm top-to-bottom line-up.
>>52626416
How are they going to release a Pascal in 28nm when it was designed for 16nm.
>>52626156
Primordials.
Nvidia Primordial χ
Nvidia Primordial γ
Nvidia Primordial Ï„
Nvidia Primordial ε
>>52626387
>Nvidia
>14nm parts
No.
>>52626136
Isn't that figure with just a 256bit bus though?
384bit will put it over 650GB/s I don't see the reason why they won't use 384bit or even 512bit unless it's physically impossible for some reason.
>>52626539
NVidia doesn't want to do big memory buses because muh power savings and muh performance per watt.
>>52626634
A wider bus uses less power than jacked up memory, they don't have the die size to spare.
>>52625669
>>52624881
This. When it came to node advances and memory upgrades(gddr5/hbm) Nvidia ALWAYS was second.
>>52626078
Big Pascal won't flop, it'll just be months late to the game.
Expect JewWorks 2.0 shilling to tout new Nvidia Exclusive(tm) bullshit.
>>52626124
>>52626147
you are both idiots.
the three biggests bottlenecks in modern GPUs/games are:
> memory bandwidth
> shader ALU capacity
> fixed function unit (ROP/TMU/tesselator) capacity
For the last few generations, AMD has gone heavy on memory bw and ALUs and skimped on the fixed function units.
For this reason, Nvidia made TWIMTBP/HairWorks/GameWorks middleware that basically floods scenes with tessellated triangles for no reason other than Radeons will choke on those first.
AMD's push is for ALU-heavy stuff, like post-processing effects and async shaders, but they don't seem to bribe as many devs.
>>52626419
>I see no reason for Pascal or Polaris to be anything other than a full 1xnm top-to-bottom line-up.
new lithography nodes are drastically more expensive per wafer and often significantly more expensive per transistor until yields rise and stabilize.
cheapo Polaris/Pascal cards built on 28nm with DP1.3/HDMI2.0/h.265 upgrades and minor tweaks to the rendering pipeline are perfectly reasonable if not exactly guaranteed to happen.
>>52625303
>usb high speed
>usb super super
What's next?
>usb mega speed?
>>52626147
>>52626178
You idiots fucking wish
Go look up the 4k version too, you'll be in for a surprise.
I wouldn't recommend it though, because the Fury X is too weak for 4k and the 4gb vram chokes on some games.
>>52624457
Do people seriously believe that there is gonna be FUCKALL of a performance difference between both memories when they're all probably gonna have like 6gb+ from x70 onwards at least?
I swear VRAM is the new RAM buzzword for tech illiterates. More than ever I'm actually seeing a LOT of people talking about GPUs as if their power was measured by their fucking memory amount. It gets even fucking worse when I witness retards thinking that VRAM is all that matters for 4k.
Reminds me of all those bitrate starved GPUs that have WAY more memory they can handle, such as 760/770 and their 4gbs models or the r9 390 with its retarded 8gb amount of memory.
>>52629914
Just to clarify before some sperg loses his shit. I'm actually talking about real world performance as in FPS gains on games.
Not this MUH SYNTHETIC USELESS BENCHMARKS placebo trash.
>>52626120
>half the speed of hbm
this is bullshit
>>52629914
>2gb gt 610
>4gb r7 240
At least the 4gb is useful in the gtx 680/770, and the 8gb on the r9 390/x.
>>52626079
gddr5x is, at most, twice as fast as gddr5. But that's a "down the line" speed, not what it can currently achieve. Right now it is 1.5x faster or so.
HBM is 4x faster than gddr5, HBM2 is 8x faster. And with a quarter of power usage.
>>52624457
>GTX 770 plays everything I want.
>Most of the games that benefit from 120fps are piss-easy to run.
>By the time I can afford a 4k/superwide monitor and a VR headset, Pascal will already be out.
>probably the GTXdesu80Ti flagship card, too.
I'm totally invested in the waiting meme, OP. You should invest too.
>>52629991
HBM2 at present will scale up to 1TB/s. Future shrinks could double data rate again simply by increase frequency inside of the given power envelope.
GDDR5X is a great low cost stop-gap measure for entry level and mid range GPUs. It is not a competing technology.
>>52630136
>not upgrading every six months
>>52630037
>2gb low end cards are bad
Its not like 4 bad ddr3 4gigabit chips are much more expensive than 4 bad 2gigabit ones.
Also having more memory is always a good thing, i've seen some games refuse to launch if your gpu doesnt have a minimum ammount of memory. 512mb/1gb feels appropriate for those gpus but its not like the price difference matters that much compared to marketing a 512mb card.
Plus if you wanted to see how nice games are with graphics all out you technically could even on those low end models. At 3fps but you could. You're right tho on 4gb versions of those cards.
In the end i am more amazed that the 970 and 980 only hold 4gb of memory while the laptop counterparts 970m and 980m have both 4gb and 8gb versions. Immagine the salt if the 970m had an 8gb version and the 3.5+0.5 meme became 7+1. I bet nobody would care anymore
>>52630237
no i would still care because i got fucking lied to
>>52630237
>2gb low end cards are bad
It is when it's a fucking gt 610.
when are the next AMD cards expected? contemplating whether to buy a 390 now or not.
>>52630536
Hawaii is fucking 3 years old now. You seriously cannot wait TWO MONTHS for a part that you'll use for years?
>>52630560
I literally just asked when the next cards are out you fucking autist.
>>52625549
This made me chuckle.
>>52630536
Q2 is all they've said. There isn't enough info to make a decision about it.
>>52630536
This summer, and so far they only have 2 chips ready - the smaller one was shown being equal to a gf960 but only at 60W power draw. No news about the bigger one.
>>52625303
Ultra bandwidth memory will be next then
i really hope amd doesnt screw up even harder than nvidia
I'll be waiting. I put a 960 in my 4690k build last year to catch up on older games.
>>52629628
Ludicrous speed
>>52625358
>getting one of the worst perf/price cards out there
