AMD RX 400 series (Polaris) MegaThread! FAQ & Resources

[Martell1977]

On average, the 480 8gb was 6% faster in DX12, it would be interesting to see if nVidia was able to close the gap, but begs the questions as to why, if they did, why did they ignore DX12 from the start?

When recommending GPU's my list currently is:

1050 > 460
470 > 1050ti
480 4gb >1060 3gb
480 8gb > 1060 6gb
1070
1080
1080ti

So, AMD has a pretty strong lineup with Polaris and its not just pricing, Polaris actually outperforms it's competitors on average AND tends to be priced better. Vega could be very compelling, but they need to get it out!

 

[lucas_7_94]

Some leaks shows like "the" VEGA CARD (which ran Doom at E3) is between 1080 and 1080Ti. Let's hope i should not sell my kidney to get one.

 

[renz496]

On average, the 480 8gb was 6% faster in DX12, it would be interesting to see if nVidia was able to close the gap, but begs the questions as to why, if they did, why did they ignore DX12 from the start?

When recommending GPU's my list currently is:

1050 > 460
470 > 1050ti
480 4gb >1060 3gb
480 8gb > 1060 6gb
1070
1080
1080ti

So, AMD has a pretty strong lineup with Polaris and its not just pricing, Polaris actually outperforms it's competitors on average AND tends to be priced better. Vega could be very compelling, but they need to get it out!

not that they ignore it but they were optimizing their driver bit by bit to gain as much they can. remember they don't have ACE like AMD. and their performance vs theoretical performance utilization already very good when compared to AMD. the performance jump that nvidia gain is not from single driver jump. guru3d make comparison with 378.66 vs 378.78 and they only see noticeable gain on hitman. even that is not as big as nvidia claim. but if you look TH titan x pascal review and compared them to the results that they get on 1080ti you can see much bigger performance gain in games like hitman and doom.

and ultimately developer was supposed to do the optimization with this low level API. but in the end developer admit that it is very difficult to do than they still pretty much rely heavily on gpu drivers just like before. one of the main bullet point AMD push for low level API in the first place was to make game developer to be less reliant on drivers from vendor. problem with X gpu? then just release the patch themselves since they have the low level access and did not have to wait for vendor first before they fix the problem. but in reality all this low level thing only increase complexity for game developer.

http://www.dsogaming.com/news/nixxes-on-directx-12-dx12-is-hard-can-be-worth-the-extra-effort-but-it-may-not-be-either/
https://www.pcgamesn.com/microsoft/ubisoft-dx12-performance

nvidia most likely aware about this. so instead of "let developer do the optimization" like many people think of with DX12 and Vulkan now they just double down their effort on their DX12/Vulkan drivers just like they did with their DX11 drivers.

 

[cdrkf]

On average, the 480 8gb was 6% faster in DX12, it would be interesting to see if nVidia was able to close the gap, but begs the questions as to why, if they did, why did they ignore DX12 from the start?

When recommending GPU's my list currently is:

1050 > 460
470 > 1050ti
480 4gb >1060 3gb
480 8gb > 1060 6gb
1070
1080
1080ti

So, AMD has a pretty strong lineup with Polaris and its not just pricing, Polaris actually outperforms it's competitors on average AND tends to be priced better. Vega could be very compelling, but they need to get it out!

not that they ignore it but they were optimizing their driver bit by bit to gain as much they can. remember they don't have ACE like AMD. and their performance vs theoretical performance utilization already very good when compared to AMD. the performance jump that nvidia gain is not from single driver jump. guru3d make comparison with 378.66 vs 378.78 and they only see noticeable gain on hitman. even that is not as big as nvidia claim. but if you look TH titan x pascal review and compared them to the results that they get on 1080ti you can see much bigger performance gain in games like hitman and doom.

and ultimately developer was supposed to do the optimization with this low level API. but in the end developer admit that it is very difficult to do than they still pretty much rely heavily on gpu drivers just like before. one of the main bullet point AMD push for low level API in the first place was to make game developer to be less reliant on drivers from vendor. problem with X gpu? then just release the patch themselves since they have the low level access and did not have to wait for vendor first before they fix the problem. but in reality all this low level thing only increase complexity for game developer.

http://www.dsogaming.com/news/nixxes-on-directx-12-dx12-is-hard-can-be-worth-the-extra-effort-but-it-may-not-be-either/
https://www.pcgamesn.com/microsoft/ubisoft-dx12-performance

nvidia most likely aware about this. so instead of "let developer do the optimization" like many people think of with DX12 and Vulkan now they just double down their effort on their DX12/Vulkan drivers just like they did with their DX11 drivers.

I think it's worth keeping in mind that in this context, the problem is *worse for nVidia* than it is AMD. I think a large driving factor behind AMD wanting low level api support (besides the fact it allowed for better use of their hardware) is that all the large game devs are working closely with GCN through their console projects- an area where historically devs are working 'closer to the silicone' anyhow. In theory that means DX12 / Vulkan should simply allow game developers to tap into that extra optimization they already do for console on PC. I wonder if that is *one* of the reasons AMD typically has picked up more performance from the lower level APIs than nViida has- or at least why it's taken nVidia a while to get to the same point...

 

[renz496]

that is one of AMD intention no doubt. because if they can give more of the optimization responsibility towards game developer then it can lessen their own driver team load. ideally it was supposed to be like that. but due to how triple A games being develop currently it is hard for game developer to do their own optimization for each hardware out there. major concern is time and resource. that's why developer said that maybe you can gain 10% more performance but does the cost/time worth the extra effort spend on it? it might be worthwhile if for 10% more effort will gain you 10% more performance. but it is not when 50% more effort will only cost you 10% performance more. and game developer still need to address their own game issue.

also i think AMD strategy to unify their pc and console optimization might lead the current problem they have right now. since AMD want things that done on the console also benefit them on the PC it seems harder for them to move away from GCN. personally i think AMD can match or even exceed nvidia power efficiency if they willing to make big changes to their architecture (like the one nvidia made from fermi to kepler). but if they do so then they will lose the "console benefit".

ultimately we probably will move away from this "low level" on pc.

 

[Bakkie]

[randomcomment] Why am I even reading this discussion? I've gone with the rx 480 8gb and I'm happy with it. [/randomcomment]

 

[cdrkf]

that is one of AMD intention no doubt. because if they can give more of the optimization responsibility towards game developer then it can lessen their own driver team load. ideally it was supposed to be like that. but due to how triple A games being develop currently it is hard for game developer to do their own optimization for each hardware out there. major concern is time and resource. that's why developer said that maybe you can gain 10% more performance but does the cost/time worth the extra effort spend on it? it might be worthwhile if for 10% more effort will gain you 10% more performance. but it is not when 50% more effort will only cost you 10% performance more. and game developer still need to address their own game issue.

also i think AMD strategy to unify their pc and console optimization might lead the current problem they have right now. since AMD want things that done on the console also benefit them on the PC it seems harder for them to move away from GCN. personally i think AMD can match or even exceed nvidia power efficiency if they willing to make big changes to their architecture (like the one nvidia made from fermi to kepler). but if they do so then they will lose the "console benefit".

ultimately we probably will move away from this "low level" on pc.

Yeah it's interesting. What I have noticed though is that whilst DX12 is proving to be somewhat lacking, Vulkan looks to be *much* stronger. Playing the new DOOM on my now ancient R9 280 yields a 20+ fps frame rate boost at 1080p on high settings. It's glorious as game is now running consistently well over my screen refresh rate, whereas I was having to lower settings under the older render path to keep things smooth.

I guess the caveat to that is that DOOM is written by ID, who are some of the most talented engine developers in the industry. Still on the flip side it looks like Vulkan is likely to gain a strong footing in the mobile games market where developers are typically working with weaker hardware and as a result need to be cleverer about how they use resources.

On the point of '10% more performance vs 50% more effort'- it's still possibly worthwhile depending on where the game performance is. Obviously if the game is already running comfortably above 60 fps on most mainstream hardware then it's probably not worth the effort. Still if you are stuck at 55 fps on a 1080ti at 1080p then maybe that 10% boost is worth a bit more?

Final thought, I don't think AMD's 'power efficiency' issue is really related to them 'sticking' to GCN. There are a number of examples of GCN cards that are actually as efficient / more efficient than nVIdias in each generation- the design isn't as bad as people think. The issue is more that nVidia got better performance than I think AMD was expecting which as forced them to push the clocks up way higher than the design is intended- which puts it firmly into the 'exponential increase in power vs performance increase' territory making the cards rather inefficient. From what I've seen, GCN parts clocked in the 800 mhz range offer good performance whilst sipping power. Push them to 1ghz and they start to get hungry, push them higher- well you get the 390X monster.

When you look at what nVidia did to get their massive perf / w gain- essentially they leveraged their work on mobile graphics socs. They implemented tile based rendering and a few other techniques, whilst also stripping out the compute capability of their cards which coupled together helped them bring power down significantly. I think if anything nVidia responded much better to the delay in silicone node transition- when they realized they were stuck with 28nm they had to make some drastic changes to effectively gain a new process node type jump without it- resulting in the ludicrously efficient 900 series. Definitely a clever move on their part- I think in this situation AMD just didn't adapt quickly enough (effecitly shoehorning Fiji onto 28nm when it was always supposed to be a 20nm part).

 

[cdrkf]

[randomcomment] Why am I even reading this discussion? I've gone with the rx 480 8gb and I'm happy with it. [/randomcomment]

Congratz on a nice card there

 

[renz496]

Yeah it's interesting. What I have noticed though is that whilst DX12 is proving to be somewhat lacking, Vulkan looks to be *much* stronger. Playing the new DOOM on my now ancient R9 280 yields a 20+ fps frame rate boost at 1080p on high settings. It's glorious as game is now running consistently well over my screen refresh rate, whereas I was having to lower settings under the older render path to keep things smooth.

I guess the caveat to that is that DOOM is written by ID, who are some of the most talented engine developers in the industry. Still on the flip side it looks like Vulkan is likely to gain a strong footing in the mobile games market where developers are typically working with weaker hardware and as a result need to be cleverer about how they use resources.

when it comes to this low level business OpenGL and Vulkan definitely can do it better than DirectX can be. extensions. IdTech mentioned because Vulkan did support vendor extensions (just like OGL) they able to optimize Doom depending on the hardware. this is how low level should be done. tweak the software based on architecture specific and strength. Direct X does not have this. but the funny thing is i see some people suggest Khronos group to ditch extensions for Vulkan because it is one of the reason why developer avoiding OpenGL in the past. without extensions we probably will see Vulkan having the same problem as DX12 right now.

vulkan on mobile is interesting topic. i think majority of game developer that develop mobile games will stick with OpenGL ES 2.0 for a very long time. while vulkan can uplift the performance the support probably only come to most recent hardware only not on very old hardware. and developer on mobile have this tendency that they want their game to reach wide audience even people with very dated hardware so they develop their game with that in mind and not pushing the limit. just look at openGL ES 3.0. android add support for OGL ES 3.0 since 2013 but how many games actually use them?

Final thought, I don't think AMD's 'power efficiency' issue is really related to them 'sticking' to GCN. There are a number of examples of GCN cards that are actually as efficient / more efficient than nVIdias in each generation- the design isn't as bad as people think. The issue is more that nVidia got better performance than I think AMD was expecting which as forced them to push the clocks up way higher than the design is intended- which puts it firmly into the 'exponential increase in power vs performance increase' territory making the cards rather inefficient. From what I've seen, GCN parts clocked in the 800 mhz range offer good performance whilst sipping power. Push them to 1ghz and they start to get hungry, push them higher- well you get the 390X monster.

that's why they need to make big changes to them and not just minor changes. they are more power efficient at lower clock but they can't retain that power efficiency when they want faster performance. for the power RX480 use around 1.3ghz nvidia can offer you GTX1070 base performance. try lowering nvidia clock and undervolt them. we might get similar result.

When you look at what nVidia did to get their massive perf / w gain- essentially they leveraged their work on mobile graphics socs. They implemented tile based rendering and a few other techniques, whilst also stripping out the compute capability of their cards which coupled together helped them bring power down significantly. I think if anything nVidia responded much better to the delay in silicone node transition- when they realized they were stuck with 28nm they had to make some drastic changes to effectively gain a new process node type jump without it- resulting in the ludicrously efficient 900 series. Definitely a clever move on their part- I think in this situation AMD just didn't adapt quickly enough (effecitly shoehorning Fiji onto 28nm when it was always supposed to be a 20nm part).

their venture on mobile market really force them to make more power efficient design despite their venture in that segment end up being a failure. nvidia learnt their lesson back in 40nm. because of that they always expect what TSMC do might not always giving them the result that they want.

 

[cdrkf]

Yeah it's interesting. What I have noticed though is that whilst DX12 is proving to be somewhat lacking, Vulkan looks to be *much* stronger. Playing the new DOOM on my now ancient R9 280 yields a 20+ fps frame rate boost at 1080p on high settings. It's glorious as game is now running consistently well over my screen refresh rate, whereas I was having to lower settings under the older render path to keep things smooth.

I guess the caveat to that is that DOOM is written by ID, who are some of the most talented engine developers in the industry. Still on the flip side it looks like Vulkan is likely to gain a strong footing in the mobile games market where developers are typically working with weaker hardware and as a result need to be cleverer about how they use resources.

when it comes to this low level business OpenGL and Vulkan definitely can do it better than DirectX can be. extensions. IdTech mentioned because Vulkan did support vendor extensions (just like OGL) they able to optimize Doom depending on the hardware. this is how low level should be done. tweak the software based on architecture specific and strength. Direct X does not have this. but the funny thing is i see some people suggest Khronos group to ditch extensions for Vulkan because it is one of the reason why developer avoiding OpenGL in the past. without extensions we probably will see Vulkan having the same problem as DX12 right now.

vulkan on mobile is interesting topic. i think majority of game developer that develop mobile games will stick with OpenGL ES 2.0 for a very long time. while vulkan can uplift the performance the support probably only come to most recent hardware only not on very old hardware. and developer on mobile have this tendency that they want their game to reach wide audience even people with very dated hardware so they develop their game with that in mind and not pushing the limit. just look at openGL ES 3.0. android add support for OGL ES 3.0 since 2013 but how many games actually use them?

Final thought, I don't think AMD's 'power efficiency' issue is really related to them 'sticking' to GCN. There are a number of examples of GCN cards that are actually as efficient / more efficient than nVIdias in each generation- the design isn't as bad as people think. The issue is more that nVidia got better performance than I think AMD was expecting which as forced them to push the clocks up way higher than the design is intended- which puts it firmly into the 'exponential increase in power vs performance increase' territory making the cards rather inefficient. From what I've seen, GCN parts clocked in the 800 mhz range offer good performance whilst sipping power. Push them to 1ghz and they start to get hungry, push them higher- well you get the 390X monster.

that's why they need to make big changes to them and not just minor changes. they are more power efficient at lower clock but they can't retain that power efficiency when they want faster performance. for the power RX480 use around 1.3ghz nvidia can offer you GTX1070 base performance. try lowering nvidia clock and undervolt them. we might get similar result.

When you look at what nVidia did to get their massive perf / w gain- essentially they leveraged their work on mobile graphics socs. They implemented tile based rendering and a few other techniques, whilst also stripping out the compute capability of their cards which coupled together helped them bring power down significantly. I think if anything nVidia responded much better to the delay in silicone node transition- when they realized they were stuck with 28nm they had to make some drastic changes to effectively gain a new process node type jump without it- resulting in the ludicrously efficient 900 series. Definitely a clever move on their part- I think in this situation AMD just didn't adapt quickly enough (effecitly shoehorning Fiji onto 28nm when it was always supposed to be a 20nm part).

their venture on mobile market really force them to make more power efficient design despite their venture in that segment end up being a failure. nvidia learnt their lesson back in 40nm. because of that they always expect what TSMC do might not always giving them the result that they want.

It's definately interesting. I'm still not conviced that 'backwards compatibility' is whats holding AMD back from improving power efficeincy (Polaris was a *huge* jump from previous gen, it's much closer to NV than with the 900 series).

I think the issue is that the decisions on these designs is made years in advance- again I think if anything nVidia may well have torn up their original plans and changed tactics when they realized the problems they were facing on nodes resulting in the 900 series- whilst AMD stuck to their original plan. The other thing may just be down to resources- nVidia have a lot less areas they need to spend their R & D budget than AMD. I think in that context, AMD look for their parts to cover more markets- I mean each year nVIdia makes considerably more different GPU's than AMD do allowing them to tailor them more to specific tasks (GP 100 vs GP102 for example- they are almost the same size, just that 100 targets specialist server tasks and GP 102 is the graphics / gaming part). In that context Vega is an 'all rounder' which is going to be good at compute *and* gaming, but arguably not quite as good in either as nVidias dedicated design. It's also why we are only going to see 2 Vega parts this gen- with Polaris making up the bottom of the stack. Meanwhile I fully expect Volta to be a full 4 - 5 gpu product stack.

Whilst it means AMD might not be the outright leader, the parts they do make each gen are good in their target space, and I'ts a very cost efficient way forward. Given things are improving for them we might see AMD start releasing larger ranges of parts again if they can afford to ramp up investment again (last time they did a full product stack was for GCN 1, and even then that was only 3 gpus).

 

[renz496]

GP100 is a lot bigger than GP102. to date GP100 is the biggest GPU that they ever made at 610mm2. in comparison GP102 is 471mm2. that's why even if Vega able to match GTX1080ti performance in theory nvidia can still come up even faster GPU to counter them. also AMD really follow what nvidia did in regards to compute. Tahiti DP is rated at 1/4 vs it's FP32. this is the same regardless the card being Radeon or Fire Pro. with Hawaii AMD start following nvidia step in reducing DP performance for radeon. Hawaii DP in hardware was rated at 1/2 vs it's FP32. but in 290X that was limited to 1/8 only making even 7970/280X are faster in DP vs 290X. with Fiji AMD pretty much follow what nvidia did with maxwell: ditch FP64 in favor of gaming performance. Fiji DP was rated at 1/16 be it in Firepro or Radeon. from the info AMD release so far Vega will pretty much retain 1/16 for DP. AMD true compute card will not coming out until next year in the form of Vega 20.

as for full product stack AMD no longer releasing top to bottom like they did with 7k series. now it is consistent for AMD to release flagship in roughly two year cadence.

7970 - december 2011
290X - october 2013
Fury X - june 2015
Vega - Q2 2017 (a lot of rumor pointing to may release)

following AMD trend they most likely still releasing something new next year but the said card most likely something in the middle. Vega true replacement most likely only come in 2019.

 

[cdrkf]

GP100 is a lot bigger than GP102. to date GP100 is the biggest GPU that they ever made at 610mm2. in comparison GP102 is 471mm2. that's why even if Vega able to match GTX1080ti performance in theory nvidia can still come up even faster GPU to counter them. also AMD really follow what nvidia did in regards to compute. Tahiti DP is rated at 1/4 vs it's FP32. this is the same regardless the card being Radeon or Fire Pro. with Hawaii AMD start following nvidia step in reducing DP performance for radeon. Hawaii DP in hardware was rated at 1/2 vs it's FP32. but in 290X that was limited to 1/8 only making even 7970/280X are faster in DP vs 290X. with Fiji AMD pretty much follow what nvidia did with maxwell: ditch FP64 in favor of gaming performance. Fiji DP was rated at 1/16 be it in Firepro or Radeon. from the info AMD release so far Vega will pretty much retain 1/16 for DP. AMD true compute card will not coming out until next year in the form of Vega 20.

as for full product stack AMD no longer releasing top to bottom like they did with 7k series. now it is consistent for AMD to release flagship in roughly two year cadence.

7970 - december 2011
290X - october 2013
Fury X - june 2015
Vega - Q2 2017 (a lot of rumor pointing to may release)

following AMD trend they most likely still releasing something new next year but the said card most likely something in the middle. Vega true replacement most likely only come in 2019.

Whilst GP100 is *bigger* than GP102- when you look at the specifications I don't think it will be any faster as a gaming part. It doesn't actually have many more shader (3840 for GP100 vs 3584 on GP102). The main difference is in the FP64 rate- GP100 supports 1/2 rate DP, GP102 doesn't. If Vega is faster than GP102 (I don't think it is for the record- at least in games), then it is likey the same or faster than GP100 from a FP32 / games point of view (GP100 is a 12tf FP32 part, GP102 is 11TF and Vega is supposed to be around the 12tf mark as well).

The point is- if Vega performs very well, AMD will have the crown and GP100 can't change that. Yes nVidia *could* build a higher performing gpu- but GP100's design decisions means it isn't going to offer any real benefit as a gaming part.

 

[renz496]

gaming wise GP102 already faster than GP100 (12tflops vs 10tlops). if AMD able to beat nvidia GP102 then obviously GP100 will not be able to beat Vega. what i mean is nvidia making entirely new chip but with similar "trait" to GP102. GP102 is the result when nvidia ditch FP64 from the chip. from 610mm2 they was able to reduce the die size down to 471mm2. now if nvidia use the headroom and make pure FP32 chip for 600mm2 nvidia can still come out on top.

this is one of the frightening thing about nvidia this generation. with 780ti vs 290X and Fiji vs 980ti if AMD able to top nvidia performance significantly back then then nvidia will have no answer to AMD because it is pretty much the the limit they can go with their gpu at the time. but that's not the case this time. although to force nvidia to go this 600mm2 behemoth route AMD need to beat the fastest GP102 config: fully enabled GP102 with base clock performance at 1.8-1.9ghz.

 

[cdrkf]

@rez496- just remember Vega is about the same size as GP102 (estimated around 475mm2 for the actual chip)- so the argument 'nVidia could do a bigger one' works both ways....

 

[renz496]

from the rumor i heard so far Vega die size (for GPU only not including interposer) is estimated around 530mm2-560mm2. haven't heard from anywhere that Vega will have around the same size as nvidia GP102.

 

[cdrkf]

from the rumor i heard so far Vega die size (for GPU only not including interposer) is estimated around 530mm2-560mm2. haven't heard from anywhere that Vega will have around the same size as nvidia GP102.

The estimates are based on scaling from a photo and it's not clear how big it is. I've seen estimates raging from 400 - 600m2 for Vega. It's hard to know for sure, I think it's safe to say it's smaller than GP100. I wouldn't be surprised if it's similar in size to GP102 although it's possible it could be a bit larger? I mean logically it has the same basic specification as Fiji, but on a much smaller process so that'd suggest it might not be all that large. I guess we won't know for sure until AMD release more details. The thing is though *even if* it isn't very size efficient, nVidia have showed their hand this generation imo (they *might* be able to release a marginally faster config of GP102 but I don't think the yields are there). I honestly don't think it'll be quite as fast as the 1080ti though (my guess is top Vega will slot in nicely between the 1080 and the 1080ti- which could be quite a neat position for it to take as the gap is fairly big).

 

[Math Geek]

curious about power usage for Vega. as shown today by Tom's the 1080ti properly cooled can OC to the same speeds as the other pascal chips but gets very close to 300w to do it. is what i expected based on 1080 OC usage. and that's gaming load and not torture load which was over 300w.

since the 480 uses more power than the 1060 for similar performance, i wonder what they will need to power a card between the 1080 and 1080ti?. hopefully it'll stay in similar usage as nvidia so we can avoid all the "space heater" nonsense of the last generation.

 

[cdrkf]

curious about power usage for Vega. as shown today by Tom's the 1080ti properly cooled can OC to the same speeds as the other pascal chips but gets very close to 300w to do it. is what i expected based on 1080 OC usage. and that's gaming load and not torture load which was over 300w.

since the 480 uses more power than the 1060 for similar performance, i wonder what they will need to power a card between the 1080 and 1080ti?. hopefully it'll stay in similar usage as nvidia so we can avoid all the "space heater" nonsense of the last generation.

It's worth keeping in mind that 480 is Polaris whilst Vega is a new design- so 480 power usage can't be used to estimate Vega usage (whereas 1080 and 1080ti are both Pascal). Vega is supposedly a pretty significant departure from previous version of GCN so it's rather an unknown.

 

[Math Geek]

that's true. i'd like to see it stay below what nvidia has at the same performance level. just to check off that box in their favor.

will have to keep waiting to see. not saying the 480 is a power hungry beast, but it does use more than the 1060 at the same performance level by a decent margin. the 1080 ti was actually a touch higher than i expected in power usage but not by a whole lot. power usage has scaled pretty linearly as the cards go up so was not too hard to guess at what that one would do. i do expect to see 325w for the HOF, Classified, Amp Extreme type cards for sure.

 

[Robert Cook]

I really could care less how much power it uses. it could draw 500W but if it is a legendary card, then it will still be great. (like the R9 295X2 and its 600W power draw)

 

[cdrkf]

I really could care less how much power it uses. it could draw 500W but if it is a legendary card, then it will still be great. (like the R9 295X2 and its 600W power draw)

Well- there is a limit to how much power you pump through silicone before it burns out (the 295X2 being a slight exception, being 2 gpu's on one board). The idea is the more efficient an architecture, the higher the performance of the fastest part is- on the assumption that the physical size / transistor count / process node are the same for all parties. I mean Pascal is more efficient than Polaris at the same die size / performance level- so it goes to reason that if you scale up Polaris and Pascal to the maximum (around 600mm2) then Pascal will be faster. That's why Vega power consumption is quite important- sure it might not beat the 1080ti, but if it can match it in perf/w (or even better it) then it bodes very well for the next gen or a larger part based on same gen.

 

[Robert Cook]

I agree, but what I meant to say was screw the fear of name calling over space heaters, AMD will get far more recognition if they trump the titan, even if it means more power used. Computers are essentially large resistors from an electrical standpoint, it is not like people should really give a damn in the first place how efficient it is. (now of course data centers or places with steep electricity prices are entirely different)

But this is becoming a rant. I really just want Vega to be great.

 

[Rogue Leader]

But this is becoming a rant. I really just want Vega to be great.

I want Vega to be great as well, but if to top out the system you need to be dropping $200+ on a good 1000w PSU to run a system on one that isn't gonna fly (And this is from a guy with a $150 titanium PSU lol). Now Ryzen has proven to be efficient (my system at full tilt draws 280 from the wall), the next step is for Vega to continue that efficiency or the stigma will stick to AMD.

My hope is that my 650w will be enough to run 2 Vega in crossfire, that would be a home run. Even though I'm not gonna do crossfire again (it still sucks).

 

[Math Geek]

i want Vega to be great as well. more importantly i want people to acknowledge it when it happens.

so sick of people ignoring the obvious just because it does not say what they hope it will say. so many places in our world have we just decided that facts don't matter and it's keeping us all moving backwards as individuals and as nations.

so maybe in this one small part of the world in one overall insignificant part of the world, we can simply let the facts speak for themselves and admit AMD got it going on lately.

 

[Robert Cook]

i want Vega to be great as well. more importantly i want people to acknowledge it when it happens.

so sick of people ignoring the obvious just because it does not say what they hope it will say. so many places in our world have we just decided that facts don't matter and it's keeping us all moving backwards as individuals and as nations.

so maybe in this one small part of the world in one overall insignificant part of the world, we can simply let the facts speak for themselves and admit AMD got it going on lately.

I fear they will always suffer from over hype until they are clearly better, but they will have difficulty being recognized as better unless they deliver to hype that was not even their own words. (some sort of messed up cycle) This is a polaris thread, and polaris did everything it was supposed to do. yes it sure got flak at release because of rumors of it overclocking to 1500MHz. I feel the same occurring with zen.