NVIDIA GeForce GTX 760 launches June 25

[Gregster]

The GTX 770 has 1750Mhz rated vRam. see here the new AMD GPU's will get the same 1750Mhz rated vRam, nothing to do with it being 2GB vs 3GB 256Bit vs 384Bit.



Only up to a point, 2Gb vs 3GB is not going to make any difference in the same way it makes no difference in 2GB to 4GB on the System memory.

Your using an analogy and applying it in extremes that simply don't exist.

The proof is in the pudding, the HD 700 overclocks better than the GTX 600 despite the HD 7000 series having more vRam and a wider bus.

Please explain to me why the 4GB 680 (Palit Jetstream) comes away with LESS fps than the 2GB (Palit Jetstream)?

http://www.guru3d.com/articles-pages/palit-geforce-gtx-680-4gb-jetstream-review,24.html

You clearly seem confused and even if it was only 0.0000001% slower, it would still be because the amount of memory modules is putting more heat into the GPU and this in turn lowers clocks.

If the 7970 had 1.5GB of VRAM, it would overclock higher than the 3GB in the majority of cases (GPU quality dependant of course) and if everything being equal, the 1.5GB would win every time.

 

[LtMatt]

Its guru3d Greg they probably had the results mixed up like they did that we all noticed in the thread the other day.

 

[humbug]

Please explain to me why the 4GB 680 (Palit Jetstream) comes away with LESS fps than the 2GB (Palit Jetstream)?

http://www.guru3d.com/articles-pages/palit-geforce-gtx-680-4gb-jetstream-review,24.html

You clearly seem confused and even if it was only 0.0000001% slower, it would still be because the amount of memory modules is putting more heat into the GPU and this in turn lowers clocks.

If the 7970 had 1.5GB of VRAM, it would overclock higher than the 3GB in the majority of cases (GPU quality dependant of course) and if everything being equal, the 1.5GB would win every time.

5%,
This is Nvidia, not AMD, Apples to Oranges.... I don't know? margin of error? different Memory IC's on the 4GB version? higher timings? GPU not able to handle 4GB vs 2GB as well?

Take your pick....

It certainly does not prove Nvidia are better than AMD for using less vRam does it? AMD already overclock higher with 3GB vs Nvidia's 2GB.

Your barking up the wrong tree, you need to look at whats going on with Nvidia and why this happens with them, you can't apply it to AMD because its happening to Nvidia, maybe there is another reason Nvidia GPU's don't come with as much vRam.

 

[Gregster]

At least when I am wrong I put my hands up but even with providing the common sense reasons why more memory modules are harder to overclock than less memory modules is ignored, I give in.

 

[Kaapstad]

5%, I don't know, margin of error, different Memory IC's on the 4GB version, lower clocked? GPU not able to handle 4GB vs 2GB as well?

Take your pick....

It does not matter, whichever way you want to go it points to the fact that larger vram is worse for overclocking than smaller vram.

 

[LtMatt]

There is no proof at all this applies to AMD cards on a 384 bit bus though is there. That could be the key difference. The point is no one knows for sure, its all just peoples opinion being dressed up as fact. Just because one thing may or may not apply to nvidia cards on a smaller bus size may not necessarily be the same for AMD and their cards.

 

[Kaapstad]

There is no proof at all this applies to AMD cards on a 384 bit bus though is there. That could be the key difference. The point is no one knows for sure, its all just peoples opinion being dressed up as fact. Just because one thing may or may not apply to nvidia cards on a smaller bus size may not necessarily be the same for AMD and their cards.

It is basic physics the denser the memory modules the more heat they produce and also the more work for the memory controller. This has got nothing to do with AMD or NVidia, it is basic physics.

 

[humbug]

It does not matter, whichever way you want to go it points to the fact that larger vram is worse for overclocking than smaller vram.

I edited my post.

5%,
This is Nvidia, not AMD, Apples to Oranges.... I don't know? margin of error? different Memory IC's on the 4GB version? higher timings? GPU not able to handle 4GB vs 2GB as well?

Take your pick....

It certainly does not prove Nvidia are better than AMD for using less vRam does it? AMD already overclock higher with 3GB vs Nvidia's 2GB.

Your barking up the wrong tree, you need to look at whats going on with Nvidia and why this happens with them, you can't apply it to AMD because its happening to Nvidia, maybe there is another reason Nvidia GPU's don't come with as much vRam.

 

[LtMatt]

It is basic physics the denser the memory modules the more heat they produce and also the more work for the memory controller. This has got nothing to do with AMD or NVidia, it is basic physics.

The memory on the 7970 is the coolest part of the card, so heat will not play any factor in it at all with any semi decent custom cooling solution.

I'm sorry Kap but just because that's your opinion does not make it fact. I'm not sure that's the case with this Tahiti line of cards.

The anecdotal evidence presented and evidence ive seen myself from looking at a whole host of paid, or user reviews of both cards, (3gb & 6gb) shows that how well the memory will overclock comes down to luck of the draw of the memory IC's you get on the card. Not because one card has 3gb and the other has 6gb.

You cannot present any evidence proving your statement other than saying its basic physics. That's fine, but you will need something with a bit more substance to convince me, Humbug and Cat and god knows who else.

 

[Rusty0611]

He's given you a reason that isn't anecdotal. If you want to disprove the theory then go ahead.

 

[LtMatt]

He's given you a reason that isn't anecdotal. If you want to disprove the theory then go ahead.

I don't need to, I'm not claiming anything as fact. I'm saying there is no proof one version overclocks better than the other and there isn't. Aside from saying its 'simple physics' and using a few benchmark reviews which actually show 6gb 7970's clocking well above their rated speed of 1500 mhz. Having an opinion is fine of course, but you can't just make a blanket statement and then try to apply it to everything without any form of proof.

 

[James J]

What gregster said about the overclocking potential of RAM being linked to the memory controller load is spot on. Sandybridge E cpus are a great example of this, with RAM like mine being rated for 2400mhz "out of the box", yet not every cpu will support this speed due to the amount of pressure that the memory controller is put under, not to mention VRMs. The more memory chips you have on a pcb, the greater the strain on the controller. This is also apparent when you increase the density of the memory also. This has been known in the overclocking market of cpu and RAM for decades now and is proven in computer science.

A Graphics card is a small scale PC, with extremely similar components that are all packed into a very small area with an extremely complex interlink system. More memory available = more memory needs to be allocated by the controller. That means more strain on the entire cards bandwidth, making it necessary to have wider bus lanes in the first place to ease the load.

 

[James J]

I don't need to, I'm not claiming anything as fact. I'm saying there is no proof one version overclocks better than the other and there isn't. Aside from saying its 'simple physics' and using a few benchmark reviews which actually show 6gb 7970's clocking well above their rated speed of 1500 mhz. Having an opinion is fine of course, but you can't just make a blanket statement and then try to apply it to everything without any form of proof.

Benchmark reviews from which sources? If its big companies instead of end users, I wouldn't trust them tbh as they'd have more than likely received cherry picked golden sample cards anyway. Sadly if you are after proof in terms of statistics about this, none of us will find them as there are too many variables with trying to prove this out right. The physics of the way GPUs work is the best argument as it makes sense, but you'd have to dig out chip binning rates for 6gb cards, a memory controller comparison between the 3gb and 6gb variations and then a large pool of end users that you can work out mean values on in both cases.

I'm not having a go so please don't take any of that offensively, I simply want to point out that this discussion cannot be proved effectively by either side, so might as well end it.

 

[LtMatt]

Benchmark reviews from which sources? If its big companies instead of end users, I wouldn't trust them tbh as they'd have more than likely received cherry picked golden sample cards anyway. Sadly if you are after proof in terms of statistics about this, none of us will find them as there are too many variables with trying to prove this out right. The physics of the way GPUs work is the best argument as it makes sense, but you'd have to dig out chip binning rates for 6gb cards, a memory controller comparison between the 3gb and 6gb variations and then a large pool of end users that you can work out mean values on in both cases.

I'm not having a go so please don't take any of that offensively, I simply want to point out that this discussion cannot be proved effectively by either side, so might as well end it.

No I'm in full agreement with you James. Its not something that we can really prove as there are no 7970 reviews on a 384bit bus of 3gb vs 6gb. I only talked about those benchmarks as Kap posted them to trying and prove his point that going by those reviews the 6gb cards overclocked much worse in comparison to 3gb 7970's. Interestingly the worst clocker of the 6 or so reviews he posted was better than what i could achieve with my 3gb card at stock volts.

 

[humbug]

What gregster said about the overclocking potential of RAM being linked to the memory controller load is spot on. Sandybridge E cpus are a great example of this, with RAM like mine being rated for 2400mhz "out of the box", yet not every cpu will support this speed due to the amount of pressure that the memory controller is put under, not to mention VRMs. The more memory chips you have on a pcb, the greater the strain on the controller. This is also apparent when you increase the density of the memory also. This has been known in the overclocking market of cpu and RAM for decades now and is proven in computer science.

A Graphics card is a small scale PC, with extremely similar components that are all packed into a very small area with an extremely complex interlink system. More memory available = more memory needs to be allocated by the controller. That means more strain on the entire cards bandwidth, making it necessary to have wider bus lanes in the first place to ease the load.

What you say is true: at a point, I.E you can't put unlimited amount of vRam on a PCB and expect the IMC to deal with it as effectively.

But Kep is taking that to extremes to argue that Nvidia's solution with less vRam is the better one to AMD's with more vRam simply because Nvidia use less.

Its completely twisted logic, he tried to argue that Nvidia using memory clocked at 1750Mhz is the result of having less vRam, he's put 2 and 2 together and in Black and White maths come up with 4, but as I pointed out to him here there are shades of Gray. I.E Nvidia are using 1750Mhz rated IC's where as AMD are still on the 1500Mhz IC's.

That has nothing to do with with how much vRam there is, its everything to do with the fact that Nvidia are using higher rated IC's.....

But he and Gregster joining in just kept droning on and on and on.... this was actually cleared up with my explaining that to him.

2GB to 3GB does not make that ^^^^ difrance, if any at all.

 

[Kaapstad]

What you say is true: at a point, I.E you can't put unlimited amount of vRam on a PCB and expect the IMC to deal with it as effectively.

But Kep is taking that to extremes to argue that Nvidia's solution with less vRam is the better one to AMD's with more vRam simply because Nvidia use less.

Its completely twisted logic, he tried to argue that Nvidia using memory clocked at 1750Mhz is the result of having less vRam, he's put 2 and 2 together and in Black and White maths come up with 4, but as I pointed out to him here there are shades of Gray. I.E Nvidia are using 1750Mhz rated IC's where as AMD are still on the 1500Mhz IC's.

That has nothing to do with with how much vRam there is, its everything to do with the fact that Nvidia are using higher rated IC's.....

But he and Gregster joining in just kept droning on and on and on.... this was actually cleared up with my explaining that to him.

2GB to 3GB does not make that ^^^^ difrance, if any at all.

Total rubbish

My whole point has been it is easier to overclock less vram than it is to overclock more.

 

[Kaapstad]

More than 2GB on a single card is a mistake, not only do you have to pay for the extra vram, it also hinders performance by making overclocking more difficult.

Extra vram is only a serious option if you are going to use multi GPUs/Multi monitors.

NVidia did a neat trick with the GTX 770, they did not increase the vram or give the card a wider mem bus, no they gave the card faster vram (much more useful).

Those GTX 770s don't half fly.

Above is my original quote everything in it is true

There is no mention of any AMD cards

No single GPU graphics card is powerful enough to use 2gb of vram and get acceptable framerates

Everything else being equal larger memory modules does make overclocking more difficult

By using faster vram it does help the GTX 770 perform better, there are plenty of benchmarks to prove it.

Sounds like an utter BS Nvidia scripted statement.

Trying to turn something that is clearly a cost cutting negative for the end user into a positive preached at the laymen masses, its quite cynical telling people something that is a disadvantage is an advantage.

A faster bus means the GPU is better at handling high texture fill. more vRam means its less likely to run out and as a result push some tasks to the much slower system RAM, it also does not make overclocking even slightly more difficult, I mean tell that to the mass of people running 3 GB 384Bit GPU's running 1750Mhz to 1900Mhz on 12500Mhz to 1500Mhz vRAM

Nvidia added faster 1750Mhz RAM to these GPU's because that is the new standard of Memory IC's, that's all it is and AMD's next GPU's will get the same speed IC's

Above is your reply I think you should read it and have a good think about what you have written, are you really replying to what I posted ?

I don't think I need to comment further.

 

[humbug]

Total rubbish

My whole point has been it is easier to overclock less vram than it is to overclock more.

More than 2GB on a single card is a mistake, not only do you have to pay for the extra vram, it also hinders performance by making overclocking more difficult.

Extra vram is only a serious option if you are going to use multi GPUs/Multi monitors.

NVidia did a neat trick with the GTX 770, they did not increase the vram or give the card a wider mem bus, no they gave the card faster vram (much more useful).

Those GTX 770s don't half fly.

Well if I took that as an assumption to you referring to the 1750Mhz memory speed on the GTX 770 vs the ~1500Mhz memory clock on the 79## series being the indication of the benefits of less vRam, then I apologise.

However, the fact here remains the GTX 770 having 1750Mhz vRam has absolutely nothing to do with having less of it, nor are they overclocked, it is everything to do with the GTX 770 having 1750Mhz rated IC's

AMD's refresh GPU will get the same rated memory IC's and they will also not be overclocked, just designed to run at that speed out of the box.

I think 3GB of 384Bit 1750Mhz rated vRam would have been far more useful on the GTX 770, and yes the GTX 770 would perform better with 3GB of 384Bit 1750Mhz vRam, if it can handle them.

 

[Gregster]

Unbelievable....

 

[Harlequin]

BF3 on max settings can use over 2gb of vram @ 1080p