R9 290X vs GTX 970

[Rienks]

Never use that website for comparisons it's complete and utter BS.

(Look in my sig)

But 280x over nVidia Titan FTW

[don_svetlio]

You are literally killing me. I have no idea which graphics card to use now :'(. I want a 980 but the 390x looks so good.  

Btw - for compute, the 390X is on par with a 980 Ti/Titan X (Folding counts I believe)

[don_svetlio]

But 280x over nVidia Titan FTW

280X used to be a Titan in 2011 when it was called the 7970 and crushed everything.

[Rienks]

Btw - for compute, the 390X is on par with a 980 Ti/Titan X (Folding counts I believe)

Honestly the 390X looks good and probably performs really well. I just realised that the reason I want a 980 because of all the hype surrounding it and nVidia's bonus' 

[Rienks]

Btw - for compute, the 390X is on par with a 980 Ti/Titan X (Folding counts I believe)

It is done https://pcpartpicker.com/user/Rienks/saved/#view=HR7bt6

[don_svetlio]

It is done https://pcpartpicker.com/user/Rienks/saved/#view=HR7bt6

390X will serve you well. MSI ones are getting very good reviews so unless you're less lucky than me, you should be more than fine

[Kinda Bottlenecked]

Honestly the 390X looks good and probably performs really well. I just realised that the reason I want a 980 because of all the hype surrounding it and nVidia's bonus' 

 

Not all of it is unwarranted. 980 is a beast. The price just isn't good. Same goes for the 390x. They're all priced too high for too low of a gain in performance because of the 970/390.

[Rienks]

390X will serve you well. MSI ones are getting very good reviews so unless you're less lucky than me, you should be more than fine

Wut happened

[don_svetlio]

Wut happened

I bought Asus - the allfather of coil whine

[boboman342]

I changed from a 970 to an r9 290x, and possibly crossfire. Sure the 970 is faster in select games, but weather it be 5fps or 20+fps faster, the average fps being well over 60, does it really matter? Plus 4GB vram vs 3.5GB.

I also sold my 970 for $50 more than my aftermarket r9 290x cost.

And yea, the 390x is slightly slower than the 980 for waaaaay less money.

[vanished]

290X actually has 4GB of VRAM, 970 has 3.5GB.

Show me the ebidance for dis clim

This is actually well known. Well, technically it does have 4 GB but the last 0.5 are basically useless so we call it 3.5.

That said, if that mattered, which you say it does, I would have thought that would show up in benchmarks.

But whatever, sounds like you found the right card in the end, so all is good in the world.

[NuclearPeace]

It's a problem in at least 6 games that use more than 4GB of VRAM at 1080p

Ones I can remember:

Skyrim

Shadow for Mordor

GTA V

 

Thats not true and these rumors need to die.

 

Games and other sorts of monitoring programs can only report the amount of VRAM that is allocated. No software can accurately measure how much VRAM is in active use, let alone is critical for the game to run smoothly. Newer games will generally allocate all of your VRAM because there is very little reason why not to. If you aren't running another graphics intensive task while playing (which is around 99%+ of the use case scenario when people game), there isn't any downsides to allocating all available VRAM to the game.

 

The 970's awkward memory partition won't hurt it at all. Both the 970 and the 290x aren't nearly as powerful for their VRAM pools to be of concern. If the Fury X is performing as well as the Titan X in higher resolutions, then you can deduct that since VRAM isn't a bottleneck for the Fury X, it isn't for the 290x either which is significantly slower.

[don_svetlio]

Thats not true and these rumors need to die.

 

Games and other sorts of monitoring programs can only report the amount of VRAM that is allocated. No software can accurately measure how much VRAM is in active use, let alone is critical for the game to run smoothly. Newer games will generally allocate all of your VRAM because there is very little reason why not to. If you aren't running another graphics intensive task while playing (which is around 99%+ of the use case scenario when people game), there isn't any downsides to allocating all available VRAM to the game.

The thing is, usually when a GPU runs out of VRAM it starts caching in system memory. The 970, however, does not "run out of VRAM" but starts using the slower part so it cannot start caching in sysRAM until all 500Mbs are in use thus causing severe stuttering and degraded performance.

[NuclearPeace]

The 970 is going to run out of compute power far, far, far before you are ever going to need the last 512MB of VRAM. If the Fury X is chugging along well with just 4gb on 4k displays, I hardly see how even VRAM becomes a concern for a 970 user.

[don_svetlio]

The 970 is going to run out of compute power far, far, far before you are ever going to need the last 512MB of VRAM. If the Fury X is chugging along well with just 4gb on 4k displays, I hardly see how even VRAM becomes a concern for a 970 user.

Remind me, what memory is the 970 using and what memory is the Fury X using?

Also - I remember people saying the same about my card yet I have used all 3GB of VRAM without even maxing certain games. I still got 60 fps and 65% GPU usage.

[NuclearPeace]

I dont see how the memory technology has any impact on the capacity. 4GB of HBM is equal to 4GB of GDDR5 in terms of storage.

[don_svetlio]

I dont see how the memory technology has any impact on the capacity. 4GB of HBM is equal to 4GB of GDDR5 in terms of storage.

4GB of HBM is 5 times faster than 4GB of GDDR5. Do you know why that matters? Well, the reason stuttering occurs when the memory runs out is that the speed of the VRAM is the bottleneck so when you increase that by 400%  you suddenly don't feel stutters when you're hitting the barrier. They still happen but they are so tiny and you can't really feel them. You cannot compare different technologies and say "they are the same" as the variables are numerous.

[vanished]

4GB of HBM is 5 times faster than 4GB of GDDR5. Do you know why that matters? Well, the reason stuttering occurs when the memory runs out is that the speed of the VRAM is the bottleneck so when you increase that by 400%  you suddenly don't feel stutters when you're hitting the barrier. They still happen but they are so tiny and you can't really feel them. You cannot compare different technologies and say "they are the same" as the variables are numerous.

 

That doesn't sound right to me... I'm pretty sure that when you run out of VRAM, the bottleneck (and thus reason for the slowness/stuttering) is the speed at which data can be moved in and out of VRAM to/from other places, like system RAM and/or disk.  If that's the case having faster VRAM wouldn't matter with regard to how well is handles running out.

[don_svetlio]

That doesn't sound right to me... I'm pretty sure that when you run out of VRAM, the bottleneck (and thus reason for the slowness/stuttering) is the speed at which data can be moved in and out of VRAM to/from other places, like system RAM and/or disk.  If that's the case having faster VRAM wouldn't matter with regard to how well is handles running out.

It can more quickly swap out information and thus most of the things that usually have to wait their turn spend their time in transit so the actual usage decreases compared to GDDR5

[vanished]

It can more quickly swap out information and thus most of the things that usually have to wait their turn spend their time in transit so the actual usage decreases compared to GDDR5

So you are saying that when VRAM is used up and swapping occurs, the bottleneck is the speed of the VRAM?  That the rest of the process of getting memory from other places read and transferred is sufficiently fast by comparison?  Because if that's true your argument makes perfect sense, but I just can't imagine that is actually the case.  Is there some way to test/show this?

[don_svetlio]

So you are saying that when VRAM is used up and swapping occurs, the bottleneck is the speed of the VRAM?  That the rest of the process of getting memory from other places read and transferred is sufficiently fast by comparison?  Because if that's true your argument makes perfect sense, but I just can't imagine that is actually the case.  Is there some way to test/show this?

I believe it's a similar case to why 1GB of GDDR5 is superior to 2GB of GDDR3

[vanished]

I believe it's a similar case to why 1GB of GDDR5 is superior to 2GB of GDDR3

If the game is using 1 GB or less, then obviously yes, that's a no brainer, but if it was using, say, 1.8 GB would that really be true?  Has someone actually done that test somewhere; I couldn't find it.

[don_svetlio]

If the game is using 1 GB or less, then obviously yes, that's a no brainer, but if it was using, say, 1.8 GB would that really be true?  Has someone actually done that test somewhere; I couldn't find it.

I'm not sure but I believe it's different since the bandwidth is double.

[vanished]

I'm not sure but I believe it's different since the bandwidth is double.

Well I'm not going to be able to believe that one without some sort of proof; it's just too counterintuitive for me.

 

And speaking of that, I was thinking a little more about the logic behind what the bottleneck would be in a situation where VRAM usage exceeds capacity.  If it is true that the VRAM speed is the deciding factor in performance when VRAM capacity is exceeded, that means that it is the bottleneck (like you said).  If that is true, that means you shouldn't notice any change whatsoever in performance whether you are using too much VRAM or not since it will always be the slowest part in the system.