390 or 970 ftw edition

[tribaljet]

-snip-

 

Then they most have patched the drivers...

 

Yes, they have patched the drivers around 7 months or more.

 

And unsurprisingly you're wrong again, the R9 290(X) can easily reach R9 390(X) performance. And you keep on banging your VRAM drum.

[Nena Trinity]

Yes, they have patched the drivers around 7 months or more.

 

And unsurprisingly you're wrong again, the R9 290(X) can easily reach R9 390(X) performance. And you keep on banging your VRAM drum.

If you try that you are most likely gonna brick it be my guest and try!

[tribaljet]

If you try that you are most likely gonna brick it be my guest and try!

 

I don't need to, many have done so successfully. Just because you are unable to do so, don't assume (you know what they say about that word, right?) others can't in a very safe manner.

[Prysin]

@tribaljet

Yes, the 970 would probably performed a few % better with a higher bus. Or rather, the 3.5+0.5 GB issue would be a non issue completely. While as now, some users DO get issues, while others DONT. We cannot say who will or who wont, but we can say that higher base bandwidth would greatly help with these things. It would also make memory OC way more effective in general.

 

Seeing as 970 vs 390. We are talking 256 vs 512 bit bus... so their respective bandwidth (OC memory using ATs 7.8GHz and stock) is:

256 x 7000 / 8 = 224,000 GB/s

256 x 7800 / 8 = 249,600 GB/s

 

compare this to the 390

512 x 6100 / 8 = 390,400 GB/s

I do not have any memory OC numbers for the 390. But even with a pretty decent OC for the 970. It still doesnt come remotely close to the R9 390.

 

Lets look at the GTX 980 Ti

384 * 7010 / 8 = 336,480

 

so even the 980Ti doesnt cut close to the AMD card(s). Matter of fact is, this is probably why AMD wins out at higher resolutions. Nvidias stingyness with the bandwidth simply prevents them from being efficient at higher resolutions without adding a whole lot more horsepower into the mix. (Which is why we can see the 390X keep catching up to the 980, eventually beating it, the higher you increase resolutions).

 

In my opinion. 970 would be better then 390 if it had a 384bit bus. It would simply fare a lot better in 1440p and 4k. Which is where the 390 completely smokes the 970 in most tests.

[Nena Trinity]

I don't need to, many have done so successfully. Just because you are unable to do so, don't assume (you know what they say about that word, right?) others can't in a very safe manner.

Oh proof?

[tribaljet]

*snipped intelligent post for the sake of cleanliness*

 

Agreed that a wider bus would benefit both the GTX 970 and 980. Yet, if looking purely at memory bandwidth, then the GTX 980 Ti does remarkably well against a watercooled Fury X, does it not? After all, memory bandwidth is only a part of the game.

 

But yes, the overall conclusion is that the GTX 970 and 980 should both have a wider bus (as well as the GTX 960 for that matter), and everyone would benefit from such.

 

 

Oh proof?

 

Oh look, a stock OC'd model that outperforms the R9 390X (SAPPHIRE VAPOR-X R9 290X TRI-X OC). What a shock!

 

But please, continue to prove how vastly knowledgeable you are. After all, I still remember you defending AMD-optimized games fiercely without even knowing Far Cry 3 was one lol.

 

Here's a protip for you, learn the difference between stock and aftermarket, and what each card can actually do.

[Nena Trinity]

Oh look, a stock OC'd model that outperforms the R9 390X (SAPPHIRE VAPOR-X R9 290X TRI-X OC). What a shock!

 

But please, continue to prove how vastly knowledgeable you are. After all, I still remember you defending AMD-optimized games fiercely without even knowing Far Cry 3 was one lol.

 

Here's a protip for you, learn the difference between stock and aftermarket, and what each card can actually do.

Sure that is the Vapor-X model it can handle heat very well!

Your average person most likely does not have that model at all or a model that can handle same amount of heat...

And you still need to get a life since you play way to many games!

[Prysin]

Agreed that a wider bus would benefit both the GTX 970 and 980. Yet, if looking purely at memory bandwidth, then the GTX 980 Ti does remarkably well against a watercooled Fury X, does it not? After all, memory bandwidth is only a part of the game.

 

But yes, the overall conclusion is that the GTX 970 and 980 should both have a wider bus (as well as the GTX 960 for that matter), and everyone would benefit from such.

Yes it does, however i think there is more to it then just bandwidth at that point. The furyX is limited by ROPs. While it is massively powerful, it has a stupid setup for its engine. The front end (thats how GCN is made) is split into 4 "chunks". Which each hold 1/4th of the ROPs, shaders and texture units in addition to other things.

its this split into 4 that defines how well GCN does tesselation and other geometry heavy tasks. However, what AMD did with the furyX was to simply "add" more stuff onto the existing "quad channel" system used with Tonga (Tahiti 2.0).

 

The way i see it, is that if AMD had revised the system to use 5 slightly smaller "channels" they may have increased geometry calculation performance by a lot. Which is what differentiates Nvidia and AMDs architectures at the base principle.

If you want to see how AMD holds up to Nvidia, simply take any Gameworks title, force tesselation off in CCC and Nvidia control panel, and suddenly, things look a lot different.

 

In the end, perhaps it isnt "just to split into 5 smaller chunks".... who knows, i certainly do not. As i am far from being as smart as AMDs or Nvidias engineers. Whatever one think of either companies, there is a lot of dedicated, intelligent people working there. They know their stuff, and while money, or the lack there of, may play a role. I do genuinely not think that AMD would jeapordize a "easy win" if all they had to do was "just split the GPU architecture into 5 smaller channels rather then 4 massive"...

[Prysin]

It cost 100USD more smarty pants barely anyone has those!

Only where you and I live....

 

That being said....

 

Proud owner of 2x 7950 OC edt. Vapor-X cards.... in addition to my AMD XFX R9 295x2

[tribaljet]

Yes it does, however i think there is more to it then just bandwidth at that point. The furyX is limited by ROPs. While it is massively powerful, it has a stupid setup for its engine. The front end (thats how GCN is made) is split into 4 "chunks". Which each hold 1/4th of the ROPs, shaders and texture units in addition to other things.

its this split into 4 that defines how well GCN does tesselation and other geometry heavy tasks. However, what AMD did with the furyX was to simply "add" more stuff onto the existing "quad channel" system used with Tonga (Tahiti 2.0).

 

The way i see it, is that if AMD had revised the system to use 5 slightly smaller "channels" they may have increased geometry calculation performance by a lot. Which is what differentiates Nvidia and AMDs architectures at the base principle.

If you want to see how AMD holds up to Nvidia, simply take any Gameworks title, force tesselation off in CCC and Nvidia control panel, and suddenly, things look a lot different.

 

In the end, perhaps it isnt "just to split into 5 smaller chunks".... who knows, i certainly do not. As i am far from being as smart as AMDs or Nvidias engineers. Whatever one think of either companies, there is a lot of dedicated, intelligent people working there. They know their stuff, and while money, or the lack there of, may play a role. I do genuinely not think that AMD would jeapordize a "easy win" if all they had to do was "just split the GPU architecture into 5 smaller channels rather then 4 massive"...

 

It certainly is odd that they went with the same ROP configuration of flagship Hawaii. They kind of seemed to do a 780->780 Ti play in a larger scale, which was interesting for sure.

 

And certainly agreed on how both manufacturers have fundamental differences that can be perceived in multiple ways, with one particularly popular way as of late being tesselation on Witcher 3.

 

What are your thoughts on someone buying AMD or at least the GPU division and seriously infuse cash so R&D could boom?

[Nena Trinity]

Learn not to speak for others. Actually, learn a lot more things in the process, the world will be a better place that way.

You really expect majority to use 100USD more on a card? You need to buy a new brain...

 

 

 

Only where you and I live....

 

That being said....

 

Proud owner of 2x 7950 OC edt. Vapor-X cards.... in addition to my AMD XFX R9 295x2

Cool!

[tribaljet]

You really expect majority to use 100USD more on a card? 

 

-snip-

 

Clearly you are unaware that we are on a computer tech forum where there are users known as enthusiasts and said users know what they go for, and you've already been told there is a smaller price difference that what you're assuming (look, that word again).

[Prysin]

It certainly is odd that they went with the same ROP configuration of flagship Hawaii. They kind of seemed to do a 780->780 Ti play in a larger scale, which was interesting for sure.

 

And certainly agreed on how both manufacturers have fundamental differences that can be perceived in multiple ways, with one particularly popular way as of late being tesselation on Witcher 3.

 

What are your thoughts on someone buying AMD or at least the GPU division and seriously infuse cash so R&D could boom?

To be honest, i wouldnt want AMD to change as a company. Rather, id like to see them out of debt.

 

Reason is APUs... yes yes i know, dead end bla bla bla. But bare with me.

For the low to mid end market, APUs make a ton of sense, and they are the driving force behind the PS4, XBONE and WiiU. While one can hate consoles all day long (i love em equally to my PC :3 ) consoles do ONE thing very well. They bring a sleek, stylish and game centric design into the home, or rather, the HEART of the home, the livingroom.

 

Currently, the best option for APUs are AMD. Sure intel got Broadwell, but that is a overpriced bloated toad that simply serves no other purpose then demonstrate the potential power of intels Iris Pro iGPU prior to Skylake launch. The main reason Intel is "winning" over AMD is because they added a massive 128mb L3 cache specifically for the iGPU. Now over to AMD, and this is the exciting part. With a new manufacturing process (20nm), they may save enough space with their upcoming ZEN architecture to not only put better iGPU cores onto there, but also HBM. Imagine a on-die HBM stack of 256-512MB.... the performance of that iGPU would probably rival that of a 270 or 270X.... That is exciting.

 

AMD also learns from their own departments. Their Bulldozer architecture is a terrible design, but with some help from their GPU designers, they found a way to increase the density of the "Excavator" iteration of Bulldozer by almost 30%. Net result is that they saved so much space they managed to slap a dedicated Hybrid H265 encoder/decoder directly on the APU die.... Thus Carizzo (upcoming laptop APUs, wont come to desktop) can handle 4k streaming and playback better then one would initially anticipate looking at pure iGPU + CPU specs...

 

AMD is a interesting company, they forge ways that  is not clear at first, but in a quirky way, they make sense (from a consumer side, not buisniss). HSA, HBM (AMD was one of the driving forces and big money funders behind it), Freesync (not directly AMDs work, but their making open source stuff, which is nice), Mantle/Vulkan....

 

 

As for ATI/AMD splitting ways again - i do not see any real benefit in that, for AMD/ATI or us consumers. It would simply cause a massive mess in the market for months, maybe even years.

[tribaljet]

To be honest, i wouldnt want AMD to change as a company. Rather, id like to see them out of debt.

 

Reason is APUs... yes yes i know, dead end bla bla bla. But bare with me.

For the low to mid end market, APUs make a ton of sense, and they are the driving force behind the PS4, XBONE and WiiU. While one can hate consoles all day long (i love em equally to my PC :3 ) consoles do ONE thing very well. They bring a sleek, stylish and game centric design into the home, or rather, the HEART of the home, the livingroom.

 

Currently, the best option for APUs are AMD. Sure intel got Broadwell, but that is a overpriced bloated toad that simply serves no other purpose then demonstrate the potential power of intels Iris Pro iGPU prior to Skylake launch. The main reason Intel is "winning" over AMD is because they added a massive 128mb L3 cache specifically for the iGPU. Now over to AMD, and this is the exciting part. With a new manufacturing process (20nm), they may save enough space with their upcoming ZEN architecture to not only put better iGPU cores onto there, but also HBM. Imagine a on-die HBM stack of 256-512MB.... the performance of that iGPU would probably rival that of a 270 or 270X.... That is exciting.

 

AMD also learns from their own departments. Their Bulldozer architecture is a terrible design, but with some help from their GPU designers, they found a way to increase the density of the "Excavator" iteration of Bulldozer by almost 30%. Net result is that they saved so much space they managed to slap a dedicated Hybrid H265 encoder/decoder directly on the APU die.... Thus Carizzo (upcoming laptop APUs, wont come to desktop) can handle 4k streaming and playback better then one would initially anticipate looking at pure iGPU + CPU specs...

 

AMD is a interesting company, they forge ways that  is not clear at first, but in a quirky way, they make sense (from a consumer side, not buisniss). HSA, HBM (AMD was one of the driving forces and big money funders behind it), Freesync (not directly AMDs work, but their making open source stuff, which is nice), Mantle/Vulkan....

 

 

As for ATI/AMD splitting ways again - i do not see any real benefit in that, for AMD/ATI or us consumers. It would simply cause a massive mess in the market for months, maybe even years.

 

Unexpected, considering most people I've talked with did prefer ATI rather than AMD, despite many claiming the HD 2000 and 3000 series being subpar, all agreed that the HD 4000 series was excellent and that there was a clear design path shift on the HD 5000 onwards.

 

APUs do indeed strike a great balance of budget, sufficient CPU grunt and surprisingly acceptable IGP performance, something that Intel hasn't done so far (despite Broadwell's IGP outperforming the best current APUs but at a significant price hike). Regarding consoles, they do the job of simplifying gaming quite well, even if over time they have become more complex as expected. However, do keep in mind I believe consoles peaked at 6th generation.

 

Yes, I very much want to see (let's go wild) something like 1GB HBM onboard, that'll give a proper kick in Intel's pants lol. And looking at HTPC usage, the future could look bright for AMD.

 

And yes, it's like AMD plans several moves well ahead, just that it's a risky method since Intel/Nvidia could pull a surprise move that could end up being disruptive enough to force AMD's hand into changing its original plans.

 

Imagine AMD's CPU and GPU divisions being split, with ARM taking the CPU division and Samsung taking the GPU division, each to improve their respective chips. I believe it would improve the market (further down the line), with the main issue being the matter of x86 licensing availability, which I believe Intel would be forced to concede in order to avoid anti-trust issues.

[JackTheYoungBuilder]

had to recheck AT if they actually had a proper 390 Review, they did not. They do have the introduction post for the 300 series, but it has no performance review or comparison other then MSRP prices.

The notion that 390 is a 290 with higher core and boost clock is long since been debunked. They did more then that to the 300 series. Sure, not a whole lot more, but enough to make a difference.

As for 3D. I cannot speak for HD3D on PC, but i tested 3D on PS4 with Grand Turismo (PS4 is using an AMD APU, so its probably a tweaked version of HD3D in there). Now, to clarify, my best friend is a Nvidia user, and we will return back to his experience with openCL in game creation later, but for now, Nvidia 3D. He has the goggles and the screen. Myself, i got a 32" 1080p Samsung 3D TV and a 34" 3440x1440p Curved VA monitor from Samsung aswell. Now, my TV. It has hardware accelerated 2D to 3D. It is not perfect, but to be honest, my experience with that, contrary to my friends Nvidia 3D screen was well... The TV does it better, but Nvidia 3D gave a better experience then the ingame 3D settings for Grand Turismo. So conclusion is, if you want 3D gaming experience, buy a TV. While you get more input lag, you get better 3D too. Best thing is, you do not need any sort of drivers or specific hardware, the TV does it for you, arguably more user friendly too as you can simply turn the 3D function on and off with the remote while playing, instead of having to Alt Tab then fickle around in Nvidias control panel.

As for OpenCL. It is still not well supported all over, and like i mentioned, my friend is using 670SLI for his main computer ATM (hes been lazy about upgrading). The two of us bought a 295x2 each, because they were on sale. Now he tells me the time 3DSMAX uses to render with the 295x2 is insane in OpenCL, however comparing a 295x2 to a 670SLI is like comparing a slingshot to a revolver. However there was some functions that were disabled due to lack of CUDA, which he didnt find very amusing.

You asked whether or not we should condone shoddy optimization.

The answer is NO. BUT, we should not pretend it doesnt happen, it does, and will continue to do so aslong as there is games being made. However Nvidia should learn that being stingy on their memory bandwidth and VRAM amounts are well, unhealthy, for gamers in general. If it costs them 1% of their margin to give us something wastly better, then they should do that. That is how you get customers. AMD is doing this right now. They are giving customers that little extra, probably with a noticable loss in their margins in comparison to charging higher for more VRAM.

circling back to the driver thing i mentioned.

Several users on these forums, and ATs forums, have complained about Nvidia dropping the ball on their drivers since the release of The Witcher 3. Nvidia has since released 2-3 hotfix WHQL drivers, but theyve either made things worse, or not done a damn shit about the issue. Nvidia finally fixed the issue with their 350.53 driver i think it was. I know Nvidia do have better overhead at DX11 then AMD due to usage of more efficient DX11 liberaries, however taking 2 months and 2-3 patches to fix an issue for their customers, that is nearly unacceptable.

One thing is fixing something and causing a NEW issue, but they didnt. The same issue remained, and in some cases got worse.

On the flip side, ive used AMD Beta drivers for 3 years now, and in those 3 years AMDs non WHQL drivers has had 5 screwups.

As for maxwell OC headroom. While true that GCN doesnt OC very well, i think that comes down to the architecture, more specifically the ROPs and the front end bus handling the geometry. I get the feeling that AMD could OC much better, but their own architectural setup just digs its nose into the ground and becomes unstable. This may also be the very reason why GCN is so compute heavy. It really is a workstation architecture made for the consumer market. For this very reason, all AMD cards up until the 200 series has a equivalent Firepro card, and you can really just swap the bios on AMDs cards to get a firepro card. Some things like double precision is just turned down in the bios, and from what i can gather, the only difference between a firepro W8100 and a 290 is that the W8100 is better binned. What would be interesting though is buying a secondhand W8100, slap a watercooler onto it and swap the bios for a 290 bios and see just how much headroom the GCN architecture really has when you max out the quality of the components.

He probably chose that CPU for the same reason as a lot of others. Because with the 3xxx series i5/i7, piledriver werent such a terrible choice, price to performance wise. Now with the 4xxx series, intel was the clear winner due to the further increase in IPC over the 3xxx series.

When i bought my original setup with a 8320, it was just around when the 4xxx series hit the market, thus they had a not so nice premium to them. So the FX system made the most sense. I however have later on moved onto a i7 4790k.

You know I won't read this right? wayyyyyyyyyyyyyyyyy too long lmao

[Doughnutnator]

 

 

Not really you cannot make a 290 reach the same speed of the 390... Also it still is stuck on 4GB VRAM!

The 290 is unable to remain stable at that high speed and may even fry it if you try it!

 

You're never going to use the 8GB of VRAM.

NEVER. Really, NEVER.

 

If any game (no one exists yet) needs 8GB of VRAM, the 390 will not be able to hand it, at all.

Proof that "The 290 is unable to remain stable at that high speed and may even fry it if you try it!"?

 

 

http://www.anandtech.com/show/7601/sapphire-radeon-r9-290-review-our-first-custom-cooled-290/5

 

 

 

And let's please not bring the JayZ 390, he said that he got a nice chip, and that he NEVER saw 1200Mhz in an AMD GPU.

Anyways, I see these 290's going well above the 1050Mhz 390 clockspeed.

 

So yeah, they can even when they try,

 

It's the same, really. The VRAM makes no sense, it's their selling point, tricking people and making them think they need that much VRAM even though their GPU's can't handle more than the half.

[Doughnutnator]

@JackTheYoungBuilder

 

Also, if you get the 970, don't get the FTW, get the G1 gaming, is the best 970, and you can overclock it balls to the wall without issues, getting great free extra performance out of it

[Dabombinable]

You're never going to use the 8GB of VRAM.

NEVER. Really, NEVER.

 

If any game (no one exists yet) needs 8GB of VRAM, the 390 will not be able to hand it, at all.

Proof that "The 290 is unable to remain stable at that high speed and may even fry it if you try it!"?

 

 

http://www.anandtech.com/show/7601/sapphire-radeon-r9-290-review-our-first-custom-cooled-290/5

 

 

 

And let's please not bring the JayZ 390, he said that he got a nice chip, and that he NEVER saw 1200Mhz in an AMD GPU.

Anyways, I see these 290's going well above the 1050Mhz 390 clockspeed.

 

So yeah, they can even when they try,

 

It's the same, really. The VRAM makes no sense, it's their selling point, tricking people and making them think they need that much VRAM even though their GPU's can't handle more than the half.

TBH, it depends on what your doing as to whether or not the vRAM can be handled fully. For example, at only 1024x768 my GTX 650 ti OC 2GB can use 1.8GB of vRAM while maintaining the 85fps vsync limit.

[Dabombinable]

@JackTheYoungBuilder

 

Also, if you get the 970, don't get the FTW, get the G1 gaming, is the best 970, and you can overclock it balls to the wall without issues, getting great free extra performance out of it

The Gigabyte Windforce (used on all of the G1 Gaming cards) cooler is loud as fuck, just turning my computer on has the damn thing running loud, and when gaming its by far the most noticeable source of noise from my computer.

[desertcomputer]

I'd get the 960 4GB if I were you...

... 960 cant handle 4GB

[tribaljet]

TBH, it depends on what your doing as to whether or not the vRAM can be handled fully. For example, at only 1024x768 my GTX 650 ti OC 2GB can use 1.8GB of vRAM while maintaining the 85fps vsync limit.

 

There's always the whole game modding matter to consider, but pushing crazy amounts of AA can indeed inflate VRAM usage for sure. And of course it's always good to have higher VRAM but whether the card can push acceptable performance while using its entire framebuffer, that's a whole different matter.

 

The Gigabyte Windforce (used on all of the G1 Gaming cards) cooler is loud as fuck, just turning my computer on has the damn thing running loud, and when gaming its by far the most noticeable source of noise from my computer.

 

There seems to be a general consensus of both initial G1 Gaming Maxwell cards to be on the louder side of things, but from a purely performance-centric perspective it is indeed one of the best cards to get. If a strike between performance and noise is desired, something like a MSI 4G might be a better option, especially since it does have its hybrid fan mode where it turns the fans off when idling or at low loads.

[Doughnutnator]

TBH, it depends on what your doing as to whether or not the vRAM can be handled fully. For example, at only 1024x768 my GTX 650 ti OC 2GB can use 1.8GB of vRAM while maintaining the 85fps vsync limit.

Yup, but your 650TI can actually use all that 2GB of VRAM and renders playable framerates in games that use that amount of VRAM

 

AFAIK, there's no game that requires 8GB of VRAM. Gaming at resolutions like 4K requires a lof of VRAM, but not 8GB.

And to be honest, both the 970 and 390 are not meant to play at 4k at all.

Even at 4K it's pretty rare if games use more than 4GB of VRAM.

 

The 390 will never fill the 8GB frame buffer or even use it, you need a powerful GPU to use that much VRAM.

[Dabombinable]

Yup, but your 650TI can actually use all that 2GB of VRAM and renders playable framerates in games that use that amount of VRAM

 

AFAIK, there's no game that requires 8GB of VRAM. Gaming at resolutions like 4K requires a lof of VRAM, but not 8GB.

And to be honest, both the 970 and 390 are not meant to play at 4k at all.

Even at 4K it's pretty rare if games use more than 4GB of VRAM.

 

The 390 will never fill the 8GB frame buffer or even use it, you need a powerful GPU to use that much VRAM.

3 words: four way crossfire

[tribaljet]

3 words: four way crossfire

 

[RyGuy99]

The Gigabyte Windforce (used on all of the G1 Gaming cards) cooler is loud as fuck, just turning my computer on has the damn thing running loud, and when gaming its by far the most noticeable source of noise from my computer.

I would get your card checked out then...I have a G1 gaming 970 and it is silent for me on the default profile. Other reviewers and owners have the same lack of noise.