According to the specs on Nvidia's page http://www.geforce.com/hardware/notebook-gpus/geforce-gtx-965m/specifications the 965M has only 128-bit memory bus, but it also has 2gb of dedicated Vram.
edit: by sufficient i mean sufficient to not affect framerate if all of the card's memory is in use.
Should i expect the memory bus to be a bottleneck in the GPU's performance or not?? (please refer to the link for further spec details), and why/how??
-
-
thegreatsquare Notebook Deity
http://www.notebookcheck.net/NVIDIA-GeForce-GTX-965M.134120.0.html
The following was taken from the above link.
The 970m is better than the 965m in "Ultra" settings at 1080p by...
...53% in Witcher 3
...29% in Dirt Rally
...36% in Battlefield Hardline
...41% in Far Cry 4
...55% in Call of Duty: Advanced Warfare
...45% in The Crew
The 965m has 80% of the 970m core count, but it loses by 30%-50% because it is stuck with a 128bit memory bus.i_pk_pjers_i likes this. -
Sufficient for what?
You could make a sound enough argument for how the architecture as it is really isn't served by a broader bus - even if you had a desktop board. Before explaining that the internal addressing on the chip basically favors a 128-bit bus towards the type of ram that is actually used (actually, it's not 128 bit either, it's 2x64 - very deliberately designed). While pointing out that most of the driver-optimisation actually relies on commands executing on segments at that size. And that for the current typical uses for graphics cards (read: current and previous apis), it makes very little sense to move towards larger internal addressing space, as well as larger internal units, over simply faster frequency and lower latency for each operation. While you might instead want to create addressing space in graphics card ram that can be written and read via the system bus that has broader bandwidth, and that might benefit from larger transport packets. But that's completely academic, since such a thing would mean an entirely different approach to how graphics cards work.
And that argument would of course rely on the idea that we should keep the current paradigm, or that no practically approachable alternatives exist.
But in practice, what you're going to see is that all the kepler and maxwell cards at the moment essentially don't benefit from a broader internal bus. And in the same way that gddr5 and ddr3 ram really doesn't matter as much as simply higher clocks on the bus-frequency (note: gddr5 runs on double the internal bus-speed to maintain the same throughput as ddr3 ram. The benefit from moving to gddr5 ram is basically that you can put a serial-produced gddr5 ram module on a card, and therefore be able to fit more ram for less money on a graphics card. But performance-wise, the actual boost is negligible, specially if you take into account the fact that it draws more power).
In the same way, the difference in performance from architecture optimisation towards maxwell and possibly the next iteration is much more significant than the hardware customisation of the different variants. For example, the 965m is based on the gk204 chip, but with some smxes disabled. So it ends up in the same range as the gk107 based kepler cards (with the ..*cough* vast 2Mb level2 cache, arguably what makes the most difference in that generation over the last one), with some differences in the feature set, along with better power-efficiency. So you could say that the reason why this board exists is to offer customers an option to take advantage of the better power-efficiency and power-profiles on maxwell, at about the same or slightly higher performance as the gk107 based cards. Which it does offer - you would absolutely pick a cut-down gk204 card over a super-clocked gk107 card - since the former would be easier to maintain without extra cooling, and running into problems when overclocking. If you would at all do that - the standard package is perfectly good without doing anything with it, after all.
So to sum up: if you wanted a card with minimal power-consumption that still has decent enough performance to run previous/current generation games and applications that rely on 3d graphics, then you wouldn't choose a 965, but a smaller maxwell-based card (that also turn up with 2Gb vram, thanks to how it really makes very little sense to differentiate these cards on ram size any more). After all, you're talking about 15-20% better scores over a 30w card (which still put those "throwaway cards" in a performance tier that is over the magic 2600 3dmark11 scores, that gives you comfortable 720p gaming without AA. While to get past that in any significant way, you would need at least triple that score). While if you wanted the best possible performance in a 70-90w envelope (which the 965m will reach on full utilization), then you would want to pick a faster maxwell card with more smxes based on the gk204 chip. And then even underclock it or make use of the dynamic power-profiles when not running the card at peak speed, to get very much the same performance as on a 965m, at about the same power-draw as well..i_pk_pjers_i likes this. -
970m also has 25% more cores than 965m which also contributes to performance gains. 965m isn't a bad card, you just won't be able to run max details in many game, although it should perform great at 1080p at high detail settings if you sacrifice AA.Last edited: Oct 31, 2015i_pk_pjers_i likes this. -
Remniscent of the days of PhysX. Run an AMD card with a nVidia GPU for PhysX. This won't really get you anywhere realistically either. I'd rather the focus be on same card SLI or Crossfire. However if Microsoft can get this to work pretty flawlessly with different OEM video cards, it shows what crap effort is put in at AMD and Nvidia to get SLI and Crossfire working properly when it's their own hardware. Maybe they should ask Microsoft to write their drivers.
-
I have an alienware m17 with the GTX 675m which has gone faulty for the third time, and to cut the story short, Dell is offering me a replacement which has the 965m, which is a take it or leave it offer.
so my concern is, will i be seeing less VRam performance (or any other performance related aspect that is hampered by this 128-bit memory bus) that will bottleneck the card??, or am i going to see performance improvements in all aspects??, and if not in all aspect, then which aspects will i see a drop in? -
thegreatsquare Notebook Deity
...or as the old adage states, "don't look a gift horse in the mouth". -
-
thegreatsquare Notebook Deity
...or in other words...
Last edited: Oct 31, 2015 -
Uh.. well, disregarding how you would get about doubled performance in 3d games overall -- there are theoretical situations with heavy static post-processing (type.. fsaax16, that you won't use anyway) that might, in certain situations, have slightly better performance on the 675m, if you compare it to a similarly clocked 192 or 128 bits internal memory bus.
Seriously, though - no discussion, go for the 965m, and either use it on standard clocks, knowing it will last. Or overclock it till it crackles, and it'll still work until the plastic in the laptop chassis rots. You're being offered a replacement for the heavier Fermi cards, that run hot and need cooling of the kind you don't typically get on a laptop. And it's a card with massively better performance overall as well. Do it now.
edit: oh, and yes, of course - if you compare a gk204 card with 128 bit memory bus to one of the gk204 variants with a 256 bit bus (that typically would be the ones with more smx units), then yes, of course it will benefit those cards. But I assure you - if you don't use post-processing effects on high resolutions (that exacerbate the effect I described earlier), you won't notice any difference whatsoever. There is a good reason why the internal memory bus on the "CUDA" type cards are broader, though. But in practical terms, you won't see the kind of bottlenecks people are talking about.Last edited: Oct 31, 2015 -
-
Old laptop specs as seen on dell site:
New laptop specs as outlined in the e-mail they sent me:
-
Huh? That would be a 17 R3, but the 17 R3 doesn't come with a GTX 965m. You're unable to configure one, and the specifications manual never references the GTX 965m.
No, I smell a potential problem occuring. Either that, or the GTX 965m will be a future option. Either way, this is something to bring up. -
The performance difference between the 675m and 965m isn't enough to reach today/tomorrow's performance demands for games - 2GB of VRAM isn't enough, see if you can get 4GB-8GB.
Think about it this way, you would likely be looking to upgrade about now or soon after, so the extra cost will be a bargain compared to paying for a completely new laptop. It might be even less than paying for a 970m/980m GPU upgrade.
Good luck -
^good idea. Would never have occurred to me to be that greedy, but very good suggestion.
Much better cpu, too.Seems like a good deal to me, unless people know something critical about dell refurbs. Might want to ask if it's been a unit that has been lying around since release, perhaps, something like that. So you know you have to repaste, things like that..
-
Can't you put an MXM 980M inside the M17x R4?
-
-
The 6700HQ is brand new, no dust on that. -
-
-
If you are being inconvenienced due to the vendor failing to provide a promised service, that is costing you valuable time away from your usage of hardware they are supposed to keep running, it is up to the owner to accept offered compensation.
It isn't asking too much for the owner to counter offer, and Dell usually responds positively to those requests. They might not even actually charge for the bump up in GPU, or other component you aren't happy with.
They are asking him to accept performance equal to what he bought years ago, but he should be offered equivalent relative performance today to what he bought then. If he bought top of the line then, he should get top of the line now; not offered a low end build equivalent to years old performance.
There are plenty of circumstances where this isn't appropriate, but in this case, with what he is working with, and a vendor like Dell that is known to accept counter offers, it is appropriate.
It may take escalating to a Manager, or another Department, which they may offer, be honest and hang in there and be polite in your request - and keep asking and explaining nicely, no need to throw a fit, and see what they counter offer.Last edited: Oct 31, 2015 -
-
You don't have to say all that, just ask for replacement to 970m level, and you would be happy to pay for upgrade to 980m.
If nothing else it will be good practice for when your new Dell fails several times and they can't fix it, and you are back in the same compensation / replacement conversation. You'll get good at and comfortable with the give and take after a while. -
-
Please come back and let us know how it works out. -
I guess it just seemed rude to me to decline the first offer, since it was so generous. -
-
Please note: the vRAM isn't the bottleneck, but the actual, slow memory bandwidth. The card's core is too strong for such a pitiful memory bandwidth. nVidrosoft is simply cutting too many corners with the cards. They have exactly three well-balanced cards in their current lineup: the 750Ti, the 980 and the 980Ti. And the 980 is overpriced.
For mobile, the 970M is the best balanced card. -
There is another thing i should mention.
I'm not dealing with dell directly, i'm dealing with a certified third party who is "the middle man" and they handle a lot of the hassle especially if i'm out of warranty (and i was).
So i'll give this third party a call tomorrow and see if it's wise to email the guy at dell and ask him for that upgrade. -
-
-
Just let him know of your honest concerns about the 965m performance, and you aren't comfortable accepting that, and that you would like a 970m as compensation that meets your comfort zone as a replacement.
It's that simple. -
-
2GB vRAM is sufficient for the performance of the card. It's borderline but sufficient. -
The intermediary may make this easier, or more difficult, stay calm and stay connected through the process and be as positive and polite as you can, until you get what you want.
Good luck -
This is the email that i prepared in response to the specs, i will send it in a few hours after i confer with the dell subsidiary.
Email:
The specs are nice but i have some concerns about elements that i believe are a downgrade, and i also have a request.
Concerns:
1. The display does not support 3d, therefore the refresh rate will be only 60hz versus the 120hz that i had and the display will not support 3d functionality.
2. The storage size is 1tb less and it doesn't include an SSD cache drive which i also had.
3. The new spec has a DVD drive instead of a blu ray drive, which is simply a downgrade.
4. The processor clock speed with turbo boost is 0.1 ghz less than what i had.
5. The new graphics card's (the 965m) memory speed is 2500, has only a 128-bit memory bus and 80gb/s memory bandwidth, as opposed to the 1500mhz memory speed, 256-bit memory bus and 96gb/s memory bandwidth that the old card had (the 675m), which means that the new graphics card will have worse VRam (video ram) performance and that will most definitely be a bottleneck.
Request:
I can overlook all the sacrifices mentioned above if only the graphics card was better, if you would kindly be willing to upgrade the swap machine's specs from the Nvidia GTX 965m to the Nvidia GTX 970m that would be fantastic.
The 965m is an option on the alienware m15, while all the new alienwares m17 r3 offer the options of either the 970m or the 980m.
So if you would kindly grant me this request i would deeply appreciate it.
Tell me what you think and what i can add to strengthen my argument to increase chances of success. -
Let's hope it goes over to the right desk at Dell the first time.
That letter is so well done, I think you would have done this yourself even before any of the suggestions here, but I am glad we could help.
Good luck -
Thanks for the support fellas, even if this doesn't work i'm still happy you pushed me to prepare a counter offer as i wouldn't have done so on my own accord
But hey, all around it's still quite a boost in performance so it won't be so bad in any case.hmscott likes this. -
One thing to consider about the bus size is that Maxwell gets more out of any given amount of bandwidth than Kepler and Fermi. The 2MB cache reduces the need to go to memory in the first place, and the new color compression techniques they added mean less bandwidth is used for any given operation.
I'm not saying Maxwell 128bit is better than Kepler 192bit for example, but Maxwell 128bit > Kepler 128bit. GCN did a similar thing for AMD when they moved from VLIW4. -
-
The problem with this card is that the core is stronger than its bandwidth can keep up with in many cases. It is not a problem the 970M has. And it would have been better as a 192-bit memory bus card, but nVidrosoft in their great wisdom, wanted to cut corners because they have no semblance of a decent lineup of cards. It's also not a problem the 960M has, because it has a lot less cores. Even though it does very well when overclocked like crazy, it still is at a basic level weaker. Maybe you could make a 960M MATCH a 965M, but then that's closer to the limits of the 960M and also not how it was intended. I couldn't expect perfect performance out of such a low-end GPU, all things considered. But the 965M is another story.i_pk_pjers_i likes this. -
One thing i noticed, the new m17 r3 has 2 fans but shared heatsinks by the cpu and gpu.
Is that necessarily a problem? -
I know it seems like I'm nitpicking, but they're somewhat important to note.hmscott likes this. -
*** Official Alienware Notebook System Replacement Thread ***
http://forum.notebookreview.com/thr...re-notebook-system-replacement-thread.781015/
You might read that and search around the forums for other examples of getting replacement laptops from Dell for ideas on how to approach them to get what you need to feel you were given a fair deal in trade for your laptop. -
Oh dear, are both the CPU and GPU BGA on the 17 r3??
Will that affect performance?? -
It will not affect performance unless you want to do CPU-intensive tasks. I would still far quicker suggest a P770DM, P775DM or P870DM instead of anything in Alienware's terrible lineup. -
-
BUUUT of course, your budget is going to be an issue. But if you were considering an AW17 R3, you should be able to buy at least a basic machine out of what I listed. -
-
So are Razer Blades.
I don't understand people
Good luck with your exchange though!
Is the GTX 965M Memory bus bandwidth sufficient?
Discussion in 'Gaming (Software and Graphics Cards)' started by abdullah_mag, Oct 31, 2015.