NVIDIA's GeForce GTX 1080 Sounds Great, Can't Wait To Try It On Linux

[johnc]

Good news it will be supported in BSD and Linux. But I'm more worried about the product itself, especially with:

• lack of details (specs) for a product that is done and will start selling before this month ends. Look around, specs are missing.
• benchmarks in presentation were done on an overclocked 2GHz+ part, while chip will sell with a lot lower clock. That means lower numbers.
• can't recall any mention of driver or hardware internals. Only mention new multiviewport render and the multimonitor angle fix.
• Do not consume less power than a 980, in fact it consumes a little more. Probably will not affect any build anyway.
• Benchmarks are completely MIA. Usually when a video card is announced, benchmarks follows. Not this time.... Hummm what are they hiding.

So if you look at those points, then the 3D Mark test that is on the Internet that claims part is only as fast as an overclock 980 Ti may be true after all.

And that's not saying is bad. Not at all, but the high curve they were claiming it may not be.

Few things...

Specs and benchmarks will be released by the reviewers, who got their cards and had their own press event on the same day as the announcement. The reviews' NDA is rumored to lift on May 17, so that's plenty of time for a buyer to do one's homework before the cards go on sale. Most of the important specs for the 1080 are known though.

The 2 GHz overclock at the presentation occurred during one of the UE demos. I don't think we can conclude that their other charts (such as those that showed a 20% improvement over a Titan X) were drawing a comparison with a 2 GHz 1080. We really don't know, but they're probably showing stock-to-stock.

A stock 1080 may only be as fast as an overclocked 980 Ti, but that's not exactly a letdown. An overclocked 1080 will probably maintain a 20% performance gap over a 980 Ti with a much lower TDP. The die size is half that of a 980 Ti, and it has about 10% fewer transistors.

The GP100 will probably be at least 50-70% faster than a 980 Ti / Titan X.

[nils_]

Good news it will be supported in BSD and Linux. But I'm more worried about the product itself, especially with:

• lack of details (specs) for a product that is done and will start selling before this month ends. Look around, specs are missing.
• benchmarks in presentation were done on an overclocked 2GHz+ part, while chip will sell with a lot lower clock. That means lower numbers.
• can't recall any mention of driver or hardware internals. Only mention new multiviewport render and the multimonitor angle fix.
• Do not consume less power than a 980, in fact it consumes a little more. Probably will not affect any build anyway.
• Benchmarks are completely MIA. Usually when a video card is announced, benchmarks follows. Not this time.... Hummm what are they hiding.

So if you look at those points, then the 3D Mark test that is on the Internet that claims part is only as fast as an overclock 980 Ti may be true after all.

Have to agree with you there. I'm usually buying those cards for Windows Gaming (using Intel graphics) in Linux but I might as well skip this generation or get the 1080Ti. I think a lot of the performance gain is achieved through Moving to a smaller process node + higher clock + higher TDP. I think with Pascal most of the Focus went into the HPC market with gaming as an afterthought. I was also surprised that they don't use HBM2.

[juno]

Respect cpus amd not have any chance* if zen have haswell ipc per core then intel shows cannonlake (cannonlake could have around 20 to 30% about more ipc compared with haswell)

*Is more realistic compete with nvidia than intel: intel is huge monster (around 50 more money than amd), them have practically unlimited resources

All of this depend of ipc per core of zen, if zen have ipc of haswell then cannonlake shows big improvement but if zen have lower ipc than haswell and then intel only shows minimum improve

Zen maybe shows in lastest part of 2016 for have more information and cannonlake appears around 1st quarter or 2017 with 10nm

For now this depend how much performance have zen if is low, intel delay cannonlake

But if zen have same ipc of haswell then cannonlake appears

Resume if amd dont put competitive product in market, intel dont have any need of put better cpu in market

I think you underestimate the effort and investment put into a microarchitecture from design to release. There are no more tons of low hanging fruits around for easy ongoing improvements like there have been some years ago. Intel puts a huge amount of money in R&D and continuously improves their architectures. Saying they are not evolving and blaming AMD is quite ignorant and unfair for both companies.
Also, there are not screws where Intel can easily tweak their IPC in a short amount of time. What they could tweak in a short-term reaction are clocks and prices, but don't expect them to. Developing a chip is a process for years, late changes in the design require much work to be done again afterwards and are not likely to be worthwhile.
BTW., it doesn't seem that Cannonlake is planned for release in Q1 anymore, more like Q3.

they're probably showing stock-to-stock.

A stock 1080 may only be as fast as an overclocked 980 Ti, but that's not exactly a letdown. An overclocked 1080 will probably maintain a 20% performance gap over a 980 Ti with a much lower TDP. The die size is half that of a 980 Ti, and it has about 10% fewer transistors.

The GP100 will probably be at least 50-70% faster than a 980 Ti / Titan X.

Fully agree on that. I don't know what miracles some have expected again.

I think a lot of the performance gain is achieved through Moving to a smaller process node + higher clock + higher TDP. I think with Pascal most of the Focus went into the HPC market with gaming as an afterthought. I was also surprised that they don't use HBM2.

Well, that should basically be it. Sure, there might be few improvements in the architecture, but the bigger ones come with Volta. This was quite clear when they first announced Pascal as a interjectional generation quite late on their roadmap.
HBM is still expensive and the performance levels and power budget of this cards don't exact HBM. "4" chips (GK104, GM204, GP104) by Nvidia always have been cheap to produce and sold with huge margin, priced in the high-end sector, when they are actually the new performance/mid-range products. This has been Nvidia's strategy for years now. The new performance chip placed just above the previous-gen high-end, praised as "the new king" -> profit. If the demoed overclocked card (>2.1 GHz) was not an extraordinary case, Nvidia could as well have given us 20% performance on top without adding more execution units and increasing die size, still with reasonable power draw, using 2 HBM stacks for higher bandwidth. But that would not bring that kind of margins.

[pinguinpc]

Intel puts a huge amount of money in R&D and continuously improves their architectures.

Saying they are not evolving and blaming AMD is quite ignorant and unfair for both companies.

Also, there are not screws where Intel can easily tweak their IPC in a short amount of time.

What they could tweak in a short-term reaction are clocks and prices, but don't expect them to.

Developing a chip is a process for years, late changes in the design require much work to be done again afterwards and are not likely to be worthwhile.

BTW., it doesn't seem that Cannonlake is planned for release in Q1 anymore, more like Q3.

For intel amd dont represent any competitor because them have around 85% of market plus most contracts with OEMs

This is correct develop chip is years process but with unlimited resources of intel are very simple* compared with amd, for amd need much time because them not have resources

*Thanks to lower amd performance, intel dont see need put cannonlake in market (this chip must be ready) because amd dont offer something require cannonlake, for intel with actual haswell - broadwell and skylake is enough for now

Without forget time given for amd during lastest years, intel only put delays in new cpus until appears competition

However as commented before if zen shows expected performance then intel see need for something more case cannonlake

[microcode]

As of today, nvidia doesn't give a shit if you don't buy their cards. You are one out of thousands and thousands. You count Nothing.

they won't go driver open source. They deliver professional drivers in all (commercial) scenarios and as of today like it or not they are the only ones having a driver that works 99% of times in situations that "count".

The rest is gossip for people like you on phoronix.

Well duh, but you don't need to be such a dick about it. Nobody asked you if NVIDIA would stop the presses for me; and frankly you leave me very disappointed in humanity.

[GreekGeek]

Hi yall,

nVidia have been "obscure" regarding the difference between the "stock" $599.00 version vs the $699.00 "Founders Edition." You could almost say, misleading....

Here is another link to Guru3D ( https://www.guru3d.com/news-story/ge...explained.html ), that explains things.

Gotta say I was mislead, as were many, hence the article... Quote from the link, "Nvidia miscommunicated the message somewhat as everybody expected a 'reference design' and then another SKU called the 'Founders edition'. What feels rather wrong is that they are asking a 100 USD price premium for what they call the 'Founders Edition'. The product is in fact your standard reference card." The rest is at the above link.

This is a poor way to release a new card, even if it looks to be a "Bat Outta Hell" edition.

#sourtaste

GreekGeek :-)

[Guest]

There are people paying "extra" just to get reference cards (with "reference" cooler) into their machines because they like how they look and fact that you can't get to them easily. That is they pay extra to get a crappy cooler. Once these things ship officially I'm sure there will be niche users willing to pay 100$ extra for "worse than stock" card.

Or perhaps they will deliver on higher quality fan and bulkier heat sink, but non-reference cards will likely still bring way more to the table, especially with how much better watercooling blocks / dual-triple fan cards are in terms of performance in single card setups.

[artivision]

Few things...

Specs and benchmarks will be released by the reviewers, who got their cards and had their own press event on the same day as the announcement. The reviews' NDA is rumored to lift on May 17, so that's plenty of time for a buyer to do one's homework before the cards go on sale. Most of the important specs for the 1080 are known though.

The 2 GHz overclock at the presentation occurred during one of the UE demos. I don't think we can conclude that their other charts (such as those that showed a 20% improvement over a Titan X) were drawing a comparison with a 2 GHz 1080. We really don't know, but they're probably showing stock-to-stock.

A stock 1080 may only be as fast as an overclocked 980 Ti, but that's not exactly a letdown. An overclocked 1080 will probably maintain a 20% performance gap over a 980 Ti with a much lower TDP. The die size is half that of a 980 Ti, and it has about 10% fewer transistors.

The GP100 will probably be at least 50-70% faster than a 980 Ti / Titan X.

They said 9Tflops, TitanX is 6, that is a +50%. Plus Async_compute =2x.