Nvidia GeForce GTX 900 Series MegaThread: FAQ and Resources

Page 15 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
It will have but as the API are directly controlled by Khronos Group they will go with something that every IHV can agree with. Probably why AMD initially want to keep Mantle alive along side OpenGL Next (Now Vulcan). Because they know they will no longer have full control over the API once Khronos group have their hands on it.
 
This is really cool! I can see a lot of applications for something like this far beyond just gaming. It will be a great thing for big data, military, wall street, design, etc. Maybe the targeted ads in my web browser won't be so stupid in the foreseeable future. :)
 
everyone will have parallel compute, difference will be in name and implementation. amd calls it async compute. nvidia will call it something else.
Even though the implementation is very basic right now due to the nature of AFR, it'll have a good future. Hope other game companies will be encouraged to make the best use of dx12.
 

Yes, their own version. I wonder how much of improvement pascal will bring in this field.
 
How big will the improvement's from the maxwell's gddr5 to pascal's hbm 2.0? From what i've seen from amd's hbm play so far their haven't been any major improvements. I don't forsee any massive game changers with hbm 2.0 thoughts?
 
maybe not on performance because our gpu are not bandwidth limited even with GDDR5. but HBM did use less power. right now i just thinking that replacing thermal paste might a bit complicated since those HBM memory also cooled by heatsink directly.
 
Agreed on DDR5 --> HBM/HBM2

Fundamental changes like that are more for preventing bottlenecks to maintain platform longevity. Going from SATA to PCI-e for storage, DDR 4 for RAM, and NVMe for memory interfacing lays the groundwork for the next several years. While those kinds of things do have some performance increase, they provide a much greater increase in potential. CPUs and GPUs can develop around those platform technologies without being limited.

Getting a new VRAM allows information to fly in and out much faster than we currently need. But I think it will eventually prevent the memory system from slowing down graphics delivery for high resolution games with high fidelity, number crunching, and other tasks that deal with large quantities of information. Part of the reason software doesn't currently require those things is because not many programmers write software for hardware that doesn't yet exist, especially for hardware that can't be developed on currently technology. In my view (and I'm definitely open to other views), HBM will allow users to not have to worry about VRAM bandwidth bottlenecks for the foreseeable future - a preventative tech rather than a curative tech.
 
theirs a pro/con for each memory type gddr5 a smaller pipe to throw water faster. or HBM a bigger pipe to throw more water slower. If hbm's a stop gap whats it alluding too? Some hybrid memory? A brand new type of memory all together, is it a fundamental flaw with the current chip design? We're building out instead of down, so whats next? Or is it just hbm's got a lot more to improve upon or is it the techs just not their?
 
I recon GDDR5X will be use for mid range to low end. By mid range i mean GP104 and below. HBM won't be cheap until more memory maker jump onto the market. Just look at AMD pricing. Doubt that they priced nano at 650 just because nano can fit easily in mini itx cases. And Fury X won't be at 650 if not for 980ti pricing.
 


I was saying that I think GDDR5x is the stopgap, not HBM or future versions of it. On-die memory will be the future, but it's too expensive now. GDDR5x will keep things moving along until production on HBM ramps up to the point where it will be affordable. That's my prediction, and it's what I meant when I said that GDDR5x (not HBM) will be a stopgap.
 


It's cheaper than I expected, at any rate. I thought it would be around $750 as did at least one review I read, it's actually $710 msrp. Much better value than the $850 EVGA Kingpin imo. Maybe Asus saves on production costs thanks to Auto Extreme technology?
 
I don't understand why to buy the Asus GTX 980 Ti Matrix when people can find a reference Titan X for the same price if the look hard enough. Clocking the Titan X to the same frequency would be faster, plus you'd have the Titan's better chip with more VRAM.
 


What, a Titan X for $710? Where, used off eBay or a classified on a forum somewhere? At any rate I prefer buying things new and off-the-shelf. Although you pay a bit of a premium for mint-condition stuff.... I've seen enough Scrapyard Wars (LinusTechTips on Youtube) to want to stay away from buying used stuff. Unless I got bored and started doing that sort of thing as a hobby.

The Titan X only comes with a reference board. There's no guarantee that you'll achieve the same clocks because it's Nvidia's reference PCB and VRMs, and the stock blower might not support overclocking as well as an aftermarket cooler could. Whereas the GTX 980 Ti Matrix is picked from the top 10% GPUs that Asus has access to, it has a really beefy heatsink that keeps the factory overclock at or below 60 C according to reviews, and it has plenty of VRMs on a custom PCB. Although the 980 Ti is a cut-down GM200, the GM200 it uses isn't cut down that much, and cards like the Matrix come with aftermarket perks that the reference-only Titan X can't offer.

Now, if you were talking about EVGA's Kingpin, on the other hand... (Side note: I said $850 previously, but the Kingpin is actually $900.) I would be more inclined to agree with you, because $900 is a lot to dish out and I did spot some new Titan X's going for $892 right before I wrote this.
 
Yeah, checking EVGA's B-Stock on the Titan X cards is a good thing to do. I've seen them there for $700. They come with a warranty and is a direct purchase from EVGA, so it's pretty reputable. Plus, they don't have the Titan X cards on B-stock that often. So if you get one and it doesn't work, there's a good chance they'll have to replace it with an A-stock (i.e. regular retail) card. Either way, it's a good way to go for good deals on the best cards if you stroke at the right time.

I generally agree with the used thing, unless you buy from a seller that won't give issues on returns and replacements - hard to find in the used market but possible.

Also, yes, shoutout to scrapyard wars and LinusTechTips generally. Good stuff!
 


CoD is real easy, on that alone you could get away with a 950. Hitman would be the hardest, probably a 970 or even 980 if money is no object. Fifa, no clue but probably easier than Hitman, probably a 960 or even 950. To run all three definitely a 970.
 


Thank you!

I will purchase the latest one as per your suggestion.
 


To give you a better idea, the power of the 900-series cards is roughly like this (click name of card for link to pricing page):

950 . < . 960 . . . . < . . . . 970 . . < . . 980 . . . < . . .980 ti < TItan X
$150 < $200 . . . . < . . . . $320. . < . . $500 . . < . . $670 . < . $1,000

Spacing on the top line is my rough approximation for the performance difference between each card, and the bottom line just shows the price at the time I wrote this (see links for current prices). The point is that going from one card to the next isn't always going to give the same performance boost. As you can see, there is very little difference between the 9880 ti and Titan X, even though the price difference is several hundred dollars. But there is a big difference between the 960 and 970 with a much smaller price difference.
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom