2016+ CPU/GPU News

[jaxa]

Move Over GTX 1080, There’s A New Titan X In Town

 

 

Nvidia GTX Titan X

 

Transistors 12 Billion

Floating Point Performance 11 TFLOPS 32-bit floating point

Deep Learning Performance 44 TOPS INT8 - Deep learning inference instruction

Cores 3584 CUDA cores

Clock Speed 1.53GHz

Memory 12GB GDDR5X

Memory Bandwidth 480GB/s

 

$1,200, sales starting August 2

Edited July 22, 2016 by jaxa

[Bikerdude]

Hah, only top end Devs or Business users will be able to afford to buy that.

[Anderson]

Like getting cheap internet 10 years ago. Murphy's law!

[jaxa]

Preliminary settlement for U.S. Nvidia GTX 970 owners: $30.

[Bikerdude]

I already got my free game out of Gigabyte ages ago "the Witcher 3"

[jaxa]

I already got my free game out of Gigabyte ages ago "the Witcher 3"

 

Oh, does that invalidate your claim? This is Nvidia dishing out the dough.

 

It doesn't look like limeys qualify though.

Edited July 30, 2016 by jaxa

[Bikerdude]

I think they would be hard pressed to deny any UK users

[jaxa]

AMD Radeon RX 470 4GB Review

[jaxa]

GTX 10-Series for laptops

Tom's

Edited August 16, 2016 by jaxa

[lowenz]

3 GB GeForce 1060 coming in september too.

Really interested in it (~200 euro )

[jaxa]

3 GB GeForce 1060 coming in september too.

Really interested in it (~200 euro )

 

I notice Tom's mentioned "Up to 6 GB" for 1060M. But I see you meant this.

 

We also have some news of the GTX 1050 (aka save your lunch money and buy a better card).

[lowenz]

1060 3 GB: the performance of a 970 (+new Pascal features) @200 Euros/Dollars.

 

It's a good deal.

Edited August 16, 2016 by lowenz

[Bikerdude]

No not really...

 

The 3Gb card is just not worth the effort, remember the shit caused by the 970 not having actual 4GB - a lot of games benefit from 4gb. That and it has less processing power than the 6Gb version. So in summary its slower than the 970 with less ram.

 

And checking online, the only cheap(er) @ £250 6GB card for sale atm are the one made by the cheap-shit manufacturers (palit, Zotac, Gainward - 1-2yr waranties),). Where are the 970 is available at less than £200 from the Likes of Msi, Asus, Gigabyte - 3yr warranties, and better coolers). So the GTX 1060 3Gb would have to be £175 or less to be worth it.

[jaxa]

So with a lot of the new cards coming with 6-8 GB of VRAM (vs 4 GB for GTX 980, and the infamous 3.5 GB for GTX 970), most users of the new cards shouldn't be running into VRAM limits for a few years, right?

[Bikerdude]

Yup, even TDM works better on card with 2GB or more.

[lowenz]

No not really...

 

The 3Gb card is just not worth the effort, remember the shit caused by the 970 not having actual 4GB - a lot of games benefit from 4gb. That and it has less processing power than the 6Gb version. So in summary its slower than the 970 with less ram.

 

And checking online, the only cheap(er) @ £250 6GB card for sale atm are the one made by the cheap-shit manufacturers (palit, Zotac, Gainward - 1-2yr waranties),). Where are the 970 is available at less than £200 from the Likes of Msi, Asus, Gigabyte - 3yr warranties, and better coolers). So the GTX 1060 3Gb would have to be £175 or less to be worth it.

The 970 (false) "problem" was the segmented memory architecture not the VRAM quantity.

 

I just tested yesterday 1 GB vs 2 GB in Shadow of Mordor (Medium -> High):

*2 GB: 0% decrease (Medium -> High)

*1 GB: 5% decrease (Medium -> High)

 

You can live with a 5% loss in performance

Edited August 16, 2016 by lowenz

[MoroseTroll]

Yup, even TDM works better on card with 2GB or more.

Perhaps this is because TDM still uses monstrous 24-bit TGA format for normals and speculars, instead of 8-bit ATI2/BC5. Should TDM cease to use TGA (and all other formats, excluding DXTC/S3TC), it'll drastically reduce its loading time and VRAM requirement.

[Bikerdude]

Perhaps this is because TDM still uses monstrous 24-bit TGA format for normals and speculars, instead of 8-bit ATI2/BC5. Should TDM cease to use TGA (and all other formats, excluding DXTC/S3TC), it'll drastically reduce its loading time and VRAM requirement.

I will take that up with the team, because if its a simple as saving in a better format that wont have any adverse effects then it should be done.

[Bikerdude]

[moved the last few posts to a seperate thread]

[MoroseTroll]

I will take that up with the team, because if its a simple as saving in a better format that wont have any adverse effects then it should be done.

I'm afraid it's not that simple. Not sure who exactly did that, but someone (rebb?) told me many years ago that some TDM shaders should be rewritten in order to support ATI1/BC4 & ATI2/BC5 formats (In OpenGL, those are LATC/RGTC). Also, some people still have extremely old GPUs without the native support of those formats (almost all pre-DX10 ATI's, Intel's, and nVidia's, excluding Radeon X800 and higher), so there should be a workaround for them... Or not?

BTW, I can't find the thread you've mentioned. Where is it? Can you move this message there, too?

[nbohr1more]

http://forums.thedarkmod.com/topic/18280-texture-file-size-project-reduce-loading-time-and-vram-requirement/?do=findComment&comment=393252

 

here ya go...

[jaxa]

Here's something potentially fun: AMD TrueAudio Next Reserves Part Of Your GPU For Physics-Based Acoustic Rendering

[jaxa]

AMD Zen Microarchitecture: Dual Schedulers, Micro-Op Cache and Memory Hierarchy Revealed

 

AMD Unveils Zen Microarchitecture, Demos Summit Ridge Performance

 

Quick summary:

 

• Available in 2017. Mobile APUs later than desktop whatever.
• An 8-core desktop chip with a 3 GHz clock went up against a slightly underclocked Intel Broadwell-E Core i7-6900K (8 cores). Not a fair fight, but could be indicative of these Zen chips smoking Intel's i7 quad cores. Clock rates for Zen subject to change.
• "AMD's original goal was to increase Zen's IPC (Instructions Per Cycle) performance by 40 percent compared to the previous-generation Excavator core. According to the company, it achieved its goals."
• Half a megabyte per core of L2 cache
• 1 or 2 megabyte per core of L3 cache (unconfirmed)
• "Zen will offer up to 5x cache bandwidth to a core over previous designs."
• 14nm FinFET GlobalFoundries
• Simultaneous multithreading, so an 8-core Zen desktop chip has 16 threads. No bullshitdozer.
• The 32-core, 64-thread Naples server chip was also shown.

Edited August 18, 2016 by jaxa

[lowenz]

Oh, nice news!

[nbohr1more]

As much as I gripe about AMD, I'm actually a little sad to see this change.

 

The idea behind Bulldozer was sound. Moving Floating Point operations to the GPU is what should be done.

If they had any competence, they would've done that "seamlessly behind the scenes" on their APU or Motherboard Chipset with

integrated graphics. Ya know, "let existing applications take advantage of this new paradigm" rather than "try to evangelize a bunch of companies\coders

to manually implement the new methods".

 

If you remember the old Nvidia AMD motherboards, (nForce) that was one of the things it did (reduced CPU workload

by intercepting instructions at the chipset and pre-processing them seamlessly... No "evangelize the coders and rewrite" needed. ).

 

Now it seems the dream of someone at AMD waking up and putting 2 plus 2 together is over.

It seems they are just gonna mimic Intel and hope they can compete on price and efficiency. (Good luck.)

 

Fingers crossed for the Zen APU releases. Maybe someone in the engineering department will sneak this idea back into

the architecture unbeknownst to the management at AMD (and do it right this time).

 

Still, it looks like good news for price\performance competition in the PC market for a little while.