Brace yourself: NEW MAXWELL CARDS INCOMING!

Will this second generation of Maxwell GPUs for the Mobile systems?

 

Are 880m's hard to sell used? What would be a price to sell one that is less than 3 months old?

 

Eh, to be fair, my single 880m can do the same and get a steady 60FPS almost everywhere barring the actual options screen (one or two of the DX11 options give me a drop of 15FPS if I toggle them but the impact in game wasn't noticeable in either of the DLC chaoters).

Not saying the 880m isn't great, when it works well, it's just that that's probably the wrong game to benchmark it by! Crysis 3 on the other hand... I get a 50-60FPS range on all high with some settings at Ultra. Now that's one hell of a thing.

 

To be honest I had been using my desktop for all my gaming needs and I went back to the laptop because I feel like it needs some love. Infinite was running well but I was getting some major slowdown in certain areas, then I remembered that when you install new drivers sli is disabled. One forehead smack later and it felt like I had just turned on Lil johns "turn down for what"

Sent from my Nexus 5 using Tapatalk

 

I have no idea, but I would guess that it will be harder to sell them once GTX 980M and R9 M295x arrives.
There is many 880M listed at ebay. Ranging from $780 and above.
gtx 880m | eBay

You could try selling them in the NBR marketplace.

Yep, notebooks will get GM204 as well

 

I am gonna wait for GTX 1100MX before I upgrade.

 

I'm gonna wait till the gtx Titan 9080 Z is released so then I can say "its over 9000!" I could buy 9800gt but a gtx titan 9080 Z sounds cool

 

I'm waiting as well. Trying to get away from all this Haswell, Kepler crippled crap. Broadwell is Haswell, basically. Hopefully Maxwell doesn't have the 880M's handicap.

2016 looks like a good year for upgrading everything.

 

I admire your patience guys, but there is no way in hell I`m waiting til 2016.

Nah-aaah. No way Jose

 

The time I have for gaming has been cut in half, literally, since I am now finishing up school. A couple years ago I had many hours to just sit around and game - I upgraded every year for three years straight. Not anymore. Priorities change.

 

Totally understand.

 

I figure if Maxwell is at least 50%~ over Kepler, and Pascal is rumored to be 100%~ over Maxwell, then 2016 is the ideal time to upgrade for at least 2+ years of "ultra" gaming. Games shouldn't change too much after 2015, as this new stuff (graphics engines, DX 12.X, etc.) will be around for 2-3 years.

 

I just bought mine in March and I am on a 3 - 4 year cycle.

 

Yea I'm going to be keeping my 880m sli setup for at least 2 years if not slightly longer. I won't lie I do get the upgrade itch pretty often.

Its especially ridiculous considering I only game maybe 3 hours a week at most.

Sent from my Nexus 5 using Tapatalk

 

Pascal won't be 2X the power of Maxwell, at least when high end desktops are concerned. In fact it won't even be 50%. This are compute numbers though so anything is possible in High resolution scenarios, but doesn't high bandwidth affect the compute performance the most?

Anyway, here are performance estimates from a nVidia employee himself. NVIDIA Maxwell GM200 and Pascal GP100 confirmed by research paper | VideoCardz.com So you guys will most likely see a Maxwell refresh next year (20 - 16nm most likely) and pascal most likely will come in 2016 and won't be more than 50% faster than Maxwell refresh. (again these are compute numbers so expect less in gaming, with the exception of some crazy and bandwidth hungry display setups).

 

It's way too early to even consider looking at articles like that about Pascal and throwing out any possibilities, lol. I'm not even going to argue your "opinion" on what Pascal will be like over Maxwell. It's pointless. We don't even have confirmed information on Maxwell, and it's coming out next month.

 

The article sites a scientific paper which states that these are rough estimates which I assume that guy in there has really good idea on how to estimate flops , his rough estimates can be taken be somewhat close to what they will be.

 

Yeah, I'm sure it's legitimate. Just like all of the Maxwell articles.

It doesn't matter who writes it - there's no possible way they'd know what Pascal will offer unless NVIDIA released information on it. So, they're basing their "estimates" on speculation and rumors, which are false 99.8% of the time so far in advance. If we were less than 6 months away from Pascal, I'd be more inclined to believe it.

 

You probably should look at the source:
http://arxiv.org/abs/1407.4698

That paper probably is far more concrete evidence on what maxwell and pascal estimated performance will be like than any rumor. it is based on using scientific fact and equations, not some picture nor obscure performance numbers. Science is the basis of what makes GPU design possible.

also we aren't talking about a runabout rumor as the source. This a scientific paper published at a university with a good reputation. Kinda ridiculous to compare the paper to some of the articles we have been receiving about maxwell.

 

I don't think J.Dre is aware of how long the typical GPU development cycle is...

 

Guess not.

In two years, let's pick up this conversation again. I guarantee you they're wrong because they don't have all of the information, and the information they have is most likely innacurate. "Being close" doesn't count.

 

So you're saying Nvidia's own word is wrong?

I think you've been buried in these inane speculation threads so long you've forgotten how to distinguish credible, or at least more reliable, sources from poppycock.

 

Lol they are not gonna hit it on the dime on exactly what the performance will be and they will be off by a tiny bit but they should be very close. So if they are off by +/- 1.5 your gonna say that it didn't count.

What your are expecting is like having the lift and drag on airplane wing to be exactly what was predicted in CFD analysis. Hell the 2D CFD analysis of the cross sectional airfoil at the mid point in the wing (-_-).

 

Numbers on paper don't always translate into numbers in reality. If they did, Haswell wouldn't be a bust. The fact is until a chip goes into production, you can do all the math in the world and still come out ahead or behind the math. Until production starts, its all theoretical.

Sent from my HTC One_M8 using Tapatalk

 

Exactly. By nature, there's no way those estimates will be right on the dot. But they're the best we've got.

 

Those numbers are for compute though. Not very useful for gaming performance.

 

Well, that was my point. So, this entire debate is pointless, lol.

 

So how can they predict if the chip will be more efficient than what what they have now? What basis do they modify their designs by? Im sure most engineers at nvidia or AMD don't get their degrees at Hogwarts. They also don't just put things together like Lego's and see if it comes out better.

They first do a preliminary design to narrow the design choices down to a few chip design. And they do this by running the math. Then they do far more complex mathematical models to get thru Critical design review then start actually building the chip for testing then production models are developed and tested then they move onto the final production model.

EDIT:

FLOPS = floating operations per second , thus they directing relate to gaming. *facepalm* these are not double point floating point operation , these are single.

Also think how does a GPU generate 3d images let alone 2d ones ?

http://en.m.wikipedia.org/wiki/Graphics_processing_unit

Polygon per a second will be approximately proportional to flops

 

Was there ever an expectation that Haswell was going to be the next Sandy Bridge? I didn't think so. All Intel cares about nowadays is mobile this, mobile that; it's been clear from their roadmaps for years. Haswell was designed from the ground up to be a mobile-focused chip. That explains the FIVR and heavy emphasis on iGPU performance and idle power consumption at the expense of higher load power consumption and heat output and less overclocking headroom.

 

So why even participate in these threads in the first place if the most substantial predictions, at least which I've seen so far, are pointless to you? Because that's what everyone is doing here, right? Predicting and theorizing. And exchanging gossip and spreading rumors like a clique of junior high girls.

No, but the progression in performance/watt is the main takeaway from those numbers. The raw data isn't useful, nor does it matter for all intents and purposes of this thread.

 

I can't get enough of you, Octiceps. You... complete me.

 

Considering that Haswell proved impossible to cool when it wasn't even overclocked, Intel underestimated the impact of that change. That becomes especially apparent when it appears they dropped the FIVR for Skylake.

Sent from my HTC One_M8 using Tapatalk

 

It's debatable how accurate FLOPS is as an indicator of gaming performance. After all, it's not taking into account the memory subsystem which is an important part of the graphics card, and games aren't scientific calculations that are mainly bottlenecked by floating-point performance. I'm of the opinion that it's OK to use FLOPS strictly for comparing the relative performance of GPU's within the same generation from the same vendor. Once you start comparing different architectures or vendors, it becomes completely meaningless.

Like I said before, my main takeaways from the paper are the performance/watt numbers.

 

the main reason haswell is impossible is because intel was using the cheapest TIM they could find. the chip itself isn't that bad. if you see temps from devil's canyon you'll notice that it's more on the tim rather than the chip

 

All the reviews I read said Devil's Canyon didn't run much cooler or overclock much better.

 

At least for Ivy Bridge, IDontCare on AnandTech found it was the 0.06mm gap between the IHS and the die caused by the adhesive used to seal the IHS to the chip that was the issue, not the TIM itself between the IHS and the die. I will try to link to the exact post later.

Devil's Canyon ran on average 6C cooler (4790K compared to 4770K) IIRC.

And FWIW, from personal expereience even Nobel Prize winners have published exaggerated claims in their paper, so "high impact factor journal from reputable university with research done by world class scientist" doesn't really mean anything. Still infinitely better than online rumors though.

 

Yeah, the glue. That was it. I think it was the exact same problem on Haswell too.

 

That's a naive way of looking at it. FLOPS will vary depending on the instructions used. Games will not have the same mix of instructions as scientific computation. And the paper is doing calculations that are completely compute bound, which is not true for games.

The paper suggests that Pascal will be much more efficient than Maxwell, but trying to extract an exact % of performance increase from it would be meaningless.

 

Pascal will have much more memory bandwith (stacked vRAM), that for sure is going to increase the performance over Maxwell.

 

Woke up to see 4 new pages to read on this thread. Hoped for Maxwell news, left disappointed. I cri.

 

Source

Source

 

OK that is true but FLOPS should be a good indicator on general performance as long the same instructions was used across the test in the paper.

Also if they set it up so that it stored and recalled data from gpu memory for the computation then it should be a good performance indicator. I will read setup procedure to see what they did.

Edit: I am agreeing on the FLOPS varying.

EDIT2:

pulled that from paper , the interesting part is "As much RAM as possible, to reduce the number of iterations required per second" which he stated as GPU requirement so I would assume he would be storing and recalling from GPU Ram. However the numbers that are predicted are not the primary aspect of the paper therefore , they are just predicted numbers that an Nvidia employee decided to put in his paper.

 

can we please stick with the topic?

 

that 6C is pretty sad going from crap TIM to soldered TIM I was expecting a lot more. shame on intel only extremes with LGA 2011 and onward will get soldered TIM.

 

If you mean, the ever vague topic of Bracing ourselves "NEW MAXWELL CARDS INCOMING" isn't that what we're all doing?

 

I am indeed bracing myself for impact, will remain bracing until something hits me

 

Not fluxless solder. It's still thermal paste like before, just an "improved next-gen polymer TIM" which is so ambiguous it's laughable since it can describe any number of aftermarket pastes commonly being sold. But the TIM wasn't even the cause of the heat issues with Ivy Bridge and Haswell in the first place. (Indeed, the stuff Intel used was just as good if not better than aftermarket pastes.) It was caused by the IHS gap, so Intel probably just fixed the gap and switched to a different color/brand of paste so they could pin it on the "magical new thermal paste." :laugh:

 

The problem of cooling Haswell is because Intel crappy thermal compound inside the chip

 

Did you even bother reading the last couple posts?

Additionally. mobile CPUs which don't have an IHS also run crazy hot, so it's an inherent defect of Haswell.

 

No i didn´t, XD.