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

[-Fran-]

How does the card feel? My XFX RX 460 feels a little cheap, the plastic. But it's a cheap card so what can you expect.

Doesn't feel cheap at all. The matte finish really helps in that regard. I'll do a comparison with a 1060 tomorrow.

EDIT:
GRID2
-> 7970Ghz: fps_total min_fps="73.549255" max_fps="146.199738" av_fps="99.564995" min_fps_ms="13.596331" max_fps_ms="6.839958" av_fps_ms="10.043691"
-> RX480 4GB: fps_total min_fps="114.566299" max_fps="120.410378" av_fps="119.962982" min_fps_ms="8.728570" max_fps_ms="8.304932" av_fps_ms="8.335905"

I have the FRTC at 120FPS as well; no need to go beyond that I think. It's in-line with what I saw in GW2: fluidity doesn't seem much, but in numbers it's a big difference.

Cheers!

 

[renz496]

It is good to hear positive feed back on the RX-480, AMD needs some decent models to get back more market share. Stock on the new AMD models is getting better, catching up with the Nvidia.

As far as AMD partners, HIS always seem like a really cheap low-end brand. I don't know if that is true or not. I've carried them off and on for years and never seen a rep or any marketing info on their products. Compared to say ASUS they come in so often I thing we should give them their own desk. Last time XFX was here they were asking us to help them clear out their Nvidia entry level cards like the 8400, 610... even thought they were switched to AMD they still had lots of stock. I've had very little issues with ASUS, GIGABYTE, MSI, Sapphire. Power Color seemed like a lower-end brand too, but they always had a little extra and they seemed to have improved their quality. But HIS I just don't feel comfortable with that brand as much as the others, and I'm not sure why.

they used to be nvidia first tier partner. so i imagine they end up with a lot of stock back then. recently i saw video about old 480 vs new 480 to investigate if there is performance degradation with old hardware after extensive use. and they get the brand new 480 directly from EVGA.

 

[renz496]

only took how long? I think both AMD and Nvidia will make sure the whole no stock thing does not happen next time around.

the upcoming release will be fine. but you will see similar problem happening once again when both transition to newer node....which might not happen for several years more.

 

[Martell1977]

the upcoming release will be fine. but you will see similar problem happening once again when both transition to newer node....which might not happen for several years more.

From the report about AMD skipping 10nm, I have to assume nVidia will do the same. I think we might see 7nm GPU's and CPU's (AMD anyways) in about 2-3 years. For once Intel might end up a node behind.

 

[Sakkura]

Meh, the naming of processes is inaccurate. It's quite possible that TSMC and GloFo 7nm will be comparable to Intel 10nm.

 

[Rogue Leader]

Meh, the naming of processes is inaccurate. It's quite possible that TSMC and GloFo 7nm will be comparable to Intel 10nm.

Where do you get this idea? In fact TSMC, GloFo, Intel and others all use the same machines. I know because the company I work for makes the machines. 7nm is 7nm is 7nm.

 

[Robert Cook]

Meh, the naming of processes is inaccurate. It's quite possible that TSMC and GloFo 7nm will be comparable to Intel 10nm.

Where do you get this idea? In fact TSMC, GloFo, Intel and others all use the same machines. I know because the company I work for makes the machines. 7nm is 7nm is 7nm.

perhaps it is how the silicon is etched, I have heard that for 14nm Glo flo it is not as small as intel 14nm. although I am not sure how this would work unless they measure something different.

 

[Rogue Leader]

Meh, the naming of processes is inaccurate. It's quite possible that TSMC and GloFo 7nm will be comparable to Intel 10nm.

Where do you get this idea? In fact TSMC, GloFo, Intel and others all use the same machines. I know because the company I work for makes the machines. 7nm is 7nm is 7nm.

perhaps it is how the silicon is etched, I have heard that for 14nm Glo flo it is not as small as intel 14nm. although I am not sure how this would work unless they measure something different.

https://www.semiwiki.com/forum/content/5256-samsung-versus-intel-14nm.html

The processes are very different, but the result is the same.

Sorry statements like this from some people drive me nuts because they believe they can read something on the internet and make a comment like that. The people that design these machines, and the people that run them, are some of the most utterly brilliant multiple PHD folks you will ever meet. A comment like that about naming being inaccurate and 7nm being the same as someone elses 10nm is a complete fabrication and inference from the limited amount of data you can find on the internet. I equivocate it to the anti-vaxxers who somehow think reading online about a bunch of names of chemicals they don't understand means the vaccine is bad, and somehow they are validated by these web sites.

 

[Robert Cook]

savage but warranted.

 

[Rogue Leader]

For those wondering how the RX 480 fares over time, heres mine, XFX RX 480 4GB (bios flash upgraded to 8), reference cooler, newest drivers. Somehow lost about 200 points in Firestrike.

July 8th: http://www.3dmark.com/fs/9199250

September 3rd: http://www.3dmark.com/fs/10039503

Computer config hasn't changed at all other than Windows updates and the latest AMD drivers. Ran it a couple times to be sure. It hits 83 degrees now where in the past it wouldn't go over 81.

 

[TJ Hooker]

Meh, the naming of processes is inaccurate. It's quite possible that TSMC and GloFo 7nm will be comparable to Intel 10nm.

Where do you get this idea? In fact TSMC, GloFo, Intel and others all use the same machines. I know because the company I work for makes the machines. 7nm is 7nm is 7nm.

Is there actually any component of an IC at 'X' nm process node that's actually 'X' nm in size?

 

[Rogue Leader]

Meh, the naming of processes is inaccurate. It's quite possible that TSMC and GloFo 7nm will be comparable to Intel 10nm.

Where do you get this idea? In fact TSMC, GloFo, Intel and others all use the same machines. I know because the company I work for makes the machines. 7nm is 7nm is 7nm.

Is there actually any component of an IC at 'X' nm process node that's actually 'X' nm in size?

The actual material? Nope not a chance (well now anyway), in the simplest terms 14nm (or 7 or whatever) is the size of the transistor etched into the material.

 

[Robert Cook]

Meh, the naming of processes is inaccurate. It's quite possible that TSMC and GloFo 7nm will be comparable to Intel 10nm.

Where do you get this idea? In fact TSMC, GloFo, Intel and others all use the same machines. I know because the company I work for makes the machines. 7nm is 7nm is 7nm.

Is there actually any component of an IC at 'X' nm process node that's actually 'X' nm in size?

the Xnm is the size of a single transistor I am pretty sure. maybe it is the spacing between the transistors that lead to the whole 14nm is not 14nm?

 

[Robert Cook]

Oh, just saw your post rouge leader

 

[TJ Hooker]

The actual material? Nope not a chance (well now anyway), in the simplest terms 14nm (or 7 or whatever) is the size of the transistor etched into the material.

the Xnm is the size of a single transistor I am pretty sure. maybe it is the spacing between the transistors that lead to the whole 14nm is not 14nm?

Is it though? Because if you look at the various pitch lengths, which I thought more or less defined the length of various dimensions found on the chip, none of them are close the the value of the process node. Also, define "size of the transistor". Is it gate length?

 

[Rogue Leader]

The actual material? Nope not a chance (well now anyway), in the simplest terms 14nm (or 7 or whatever) is the size of the transistor etched into the material.

the Xnm is the size of a single transistor I am pretty sure. maybe it is the spacing between the transistors that lead to the whole 14nm is not 14nm?

Is it though? Because if you look at the various pitch lengths, which I thought more or less defined the length of various dimensions found on the chip, none of them are close the the value of the process node. Also, define "size of the transistor". Is it gate length?

It is gate length.

That said (and this is partially my fault) we should take this discussion elsewhere as the thread has gone way off topic.

We should instead get back to what I posted above... my RX 480 has lost 200 points in 3d mark since new.

 

[renz496]

the upcoming release will be fine. but you will see similar problem happening once again when both transition to newer node....which might not happen for several years more.

From the report about AMD skipping 10nm, I have to assume nVidia will do the same. I think we might see 7nm GPU's and CPU's (AMD anyways) in about 2-3 years. For once Intel might end up a node behind.

hard to say actually when each fab have their own definition to their own process. personally i'd expect "true" 7nm might not be ready until 2020 time frame. there already rumors talking about nvidia volta will also going to use the same node as pascal. the thing is GPU is different beast than SoC. 10nm could be well end up the same as 20nm where the jump from 16nm to 10nm did not make the difference needed by GPU.

 

[renz496]

The actual material? Nope not a chance (well now anyway), in the simplest terms 14nm (or 7 or whatever) is the size of the transistor etched into the material.

the Xnm is the size of a single transistor I am pretty sure. maybe it is the spacing between the transistors that lead to the whole 14nm is not 14nm?

Is it though? Because if you look at the various pitch lengths, which I thought more or less defined the length of various dimensions found on the chip, none of them are close the the value of the process node. Also, define "size of the transistor". Is it gate length?

It is gate length.

That said (and this is partially my fault) we should take this discussion elsewhere as the thread has gone way off topic.

We should instead get back to what I posted above... my RX 480 has lost 200 points in 3d mark since new.

when it is known that RX480 end up using more power than 150w despite only having one 6 pin i think AMD introduce new netting in CCC called "compatible" mode or something which will limit the card power consumption to 150w. also it seems they were turned on by default. how this affect real performance?

 

[BurgerandChips66]

the upcoming release will be fine. but you will see similar problem happening once again when both transition to newer node....which might not happen for several years more.

From the report about AMD skipping 10nm, I have to assume nVidia will do the same. I think we might see 7nm GPU's and CPU's (AMD anyways) in about 2-3 years. For once Intel might end up a node behind.

hard to say actually when each fab have their own definition to their own process. personally i'd expect "true" 7nm might not be ready until 2020 time frame. there already rumors talking about nvidia volta will also going to use the same node as pascal. the thing is GPU is different beast than SoC. 10nm could be well end up the same as 20nm where the jump from 16nm to 10nm did not make the difference needed by GPU.

This video about 14nm was a bit of an eye opener. I assumed that when Intel talked about 14nm they were referring to the same fabrication process as AMD (when they talk about 14nm). Apparently not!

https://youtu.be/ktHciPVAZaU

 

[-Fran-]

My RX 480 has lost 200 points in 3d mark since new.

I am willing to put the blame on the cooling, and particularly, the thermal paste used. The reference designs, specially being from XFX, I bet used the cheapest of the cheapest materials available.

It would be worth it re-pasting it and cleaning it (dust bunnies and all in the fins blocking airflow).

Also, I couldn't bench the 1060 6GB of my friends, cause we noticed the CPU didn't come with the HSF, so we had to go hunt one down to town on a Saturday afternoon, haha. We were able to find a 212 EVO in the weirdest, most remote, tech-store in Nottingham. All of that ate 3h of my testing, so we didn't do much other than the fail checks (prime for stability and Rocket League).

Still, same good old nVidia and their sucky TV support. I had a horrible time trying to make the TV work with it.

Cheers!

 

[Rogue Leader]

when it is known that RX480 end up using more power than 150w despite only having one 6 pin i think AMD introduce new netting in CCC called "compatible" mode or something which will limit the card power consumption to 150w. also it seems they were turned on by default. how this affect real performance?

Yes I know about compatible mode but I kept that off, I dont care much about the whole power consumption issue

My RX 480 has lost 200 points in 3d mark since new.

I am willing to put the blame on the cooling, and particularly, the thermal paste used. The reference designs, specially being from XFX, I bet used the cheapest of the cheapest materials available.

It would be worth it re-pasting it and cleaning it (dust bunnies and all in the fins blocking airflow).

No dust bunnies here! My case has 2 200m intake fans and 1 200m and 120m exhaust, card is as clean as the day it was new (shined a light through the heatsink to be sure). It could be the thermal paste to explain the temps but even 83 is well below its limit so that can't explain the 200 point loss... IMO.

 

[Sakkura]

Meh, the naming of processes is inaccurate. It's quite possible that TSMC and GloFo 7nm will be comparable to Intel 10nm.

Where do you get this idea? In fact TSMC, GloFo, Intel and others all use the same machines. I know because the company I work for makes the machines. 7nm is 7nm is 7nm.

http://www.eetimes.com/document.asp?doc_id=1329279

7nm is not just 7nm.

 

[Rogue Leader]

Meh, the naming of processes is inaccurate. It's quite possible that TSMC and GloFo 7nm will be comparable to Intel 10nm.

Where do you get this idea? In fact TSMC, GloFo, Intel and others all use the same machines. I know because the company I work for makes the machines. 7nm is 7nm is 7nm.

http://www.eetimes.com/document.asp?doc_id=1329279

7nm is not just 7nm.

According to the agreement between these companies, ITRS, it is. As well I did not say that Intel doesn't do it better, if you look at the article I linked, it (in not so many words) does say that. The reality is the tech has changed enough that maybe a new metric needs to be used, but one articles interpretation of it does not make it correct. As I said we should take this discussion elsewhere and not take this thread further off topic.

 

[turkey3_scratch]

The 6-pin cable is really 150W just like the 8-pin; the 75W rating should just be ignored.

 

[SteelOwl]

What to choose between RX 480 Nitro+ 8GB or MSI version? Is Nitro+ too loud? Nitro is cheaper and 5% stronger but every review says that is too loud...