AMD Radeon R9 380X Review and Benchmark Thread

[Castdeath97]

I would consider the R9 290s now that they have dropped in price and they still beat the r9 380x by a large margin. Though most of them come with stock cooler which well all know are just meeehhhh.

[MEC-777]

Yea, that review was junk too. I didn't mention it since I was specifically talking about biases. I knew I should have stopped reading when he compared it to a PS4 in his opening. He actually says that the fact it's a 4 GB card over a 256 bit bus is a drawback. Hmmm, I wonder what other cards have that same configuration.

 

That's the classic Nvidia/AMD double standard right there. 

 

GTX 960 with 128bit bus - no complaints. 

R9 380X with 256bit bus - "it's a drawback".

 

What? Does not compute. 

 

Also noticed that Tom's were unable to get the card to OC without throttling and complained about it, yet Techpowerup OC'd their card just fine, benchmarked it and made no mention of throttling issues... hmmm...  

 

You can tell just by the wording, which reviews/reviewers hold more/less bias. That's usually why I don't bother reading all their comments and just skip straight to the numbers. Digital foundry's comparison videos are, by far, the best source for unbiased, real-world performance analysis, I come across yet. I'm sure they'll have some 380X videos up soon.

[don_svetlio]

That's the classic Nvidia/AMD double standard right there. 

 

GTX 960 with 128bit bus - no complaints. 

R9 380X with 256bit bus - "it's a drawback".

 

What? Does not compute. 

 

Also noticed that Tom's were unable to get the card to OC without throttling and complained about it, yet Techpowerup OC'd their card just fine, benchmarked it and made no mention of throttling issues... hmmm...  

 

You can tell just by the wording, which reviews/reviewers hold more/less bias. That's usually why I don't bother reading all their comments and just skip straight to the numbers. Digital foundry's comparison videos are, by far, the best source for unbiased, real-world performance analysis, I come across yet. I'm sure they'll have some 380X videos up soon.

OC3D is also fine if you ask me - his tone is always "I don't give a shit - have some numbers!"

[MEC-777]

OC3D is also fine if you ask me - his tone is always "I don't give a shit - have some numbers!"

 

Yeah, I like tiny Tom Logan's reviews. He talks about the product in great detail (in his videos anyways) but is also no BS about it. He [generally] tells it like it is, while also throwing in some of his own thoughts/opinions. 

 

That being said, I just read his conclusions of his 380X review and I actually disagree with a number of things he talks about with regards to AMD's supposed performance deficit since the 7000 series. I don't see where he's pulling that from. The 290/X put up a really good fight against the 780/Ti and the 780Ti was Nvidia's response to the 290X. With the Fury X, yes, it fell prey to the 980Ti and still does in most tests, but it is catching up. I don't think we've seen the best from Fiji just yet. But yeah, AMD has been very competitive, performance and price wise, with the 200 and 300 series, so I don't really get where some of his comments about AMD having fallen behind are coming from. 

[Curufinwe_wins]

Yeah, I like tiny Tom Logan's reviews. He talks about the product in great detail (in his videos anyways) but is also no BS about it. He [generally] tells it like it is, while also throwing in some of his own thoughts/opinions. 

 

That being said, I just read his conclusions of his 380X review and I actually disagree with a number of things he talks about with regards to AMD's supposed performance deficit since the 7000 series. I don't see where he's pulling that from. The 290/X put up a really good fight against the 780/Ti and the 780Ti was Nvidia's response to the 290X. With the Fury X, yes, it fell prey to the 980Ti and still does in most tests, but it is catching up. I don't think we've seen the best from Fiji just yet. But yeah, AMD has been very competitive, performance and price wise, with the 200 and 300 series, so I don't really get where some of his comments about AMD having fallen behind are coming from. 

Fiji has never been competitive with GM200, and never will be (when looking OC to OC).

 

The issue with the 290(x) was the reference cooler was so bad it made it seem far less competitive than it actually was, and driver maturity took freaking forever in comparison to the 780/780ti. Likewise, if you could go back and have the improved tessellation back before the 300 "series" launch, I think everything would have come out way differently.

[AlwaysFSX]

God damn that reference card looks good.

[Agosto]

Meanwhile, in Italy...

 

Almost 300€.

yeah, useful

[Curufinwe_wins]

That's the classic Nvidia/AMD double standard right there. 

 

GTX 960 with 128bit bus - no complaints. 

R9 380X with 256bit bus - "it's a drawback".

 

What? Does not compute. 

 

Also noticed that Tom's were unable to get the card to OC without throttling and complained about it, yet Techpowerup OC'd their card just fine, benchmarked it and made no mention of throttling issues... hmmm...  

 

You can tell just by the wording, which reviews/reviewers hold more/less bias. That's usually why I don't bother reading all their comments and just skip straight to the numbers. Digital foundry's comparison videos are, by far, the best source for unbiased, real-world performance analysis, I come across yet. I'm sure they'll have some 380X videos up soon.

 

So a few quick comments.

 

1. That Tom's Hardware review has some seriously bullshit spin on it. Yes it has a narrower bus than Hawaii, but literally everything but Fiji does. For good reason as with the improved color compression of Maxwell AND Tonga, it wasn't needed. At all.

 

BTW yes, literally every Maxwell card is bottlenecked a bit by their bus width, the 960 more than all the rest of the lineup. Hence why memory overclocking makes such a huge difference for Maxwell (even the 980, which is less bus limited than all but GM200, showed memory overclocking returns per Mhz to be nearly 87% of the core clock return per Mhz). But still even GM200 definitely didn't need a 512 bit bus, and it would have been a total waste to give it one.

 

2. So the graphs they show for gpu clock speed (and temps) make sense, but I don't think that was thermal throttling, that looked more like a driver glitch. I can hope that someone mistakenly assumed those dropped clocks were throttling when it clearly wasn't (because OBVIOUSLY Tonga doesn't thermal throttle at 75C) and so just didn't bother trying overclocking or something. IDK but it's bs.

 

3. I like DF, but I really don't like stock to stock reviews because imho it's stupid and no one should run at stock. Plus factory overclocks are never the same, and I wish DF would at least do something to standardize their clocks a little bit (so they don't have the stupid shit on the 2GB vs 4GB videos where the difference in factory overclock makes looking at the difference in performance way less obvious).

[MEC-777]

Fiji has never been competitive with GM200, and never will be (when looking OC to OC).

 

The issue with the 290(x) was the reference cooler was so bad it made it seem far less competitive than it actually was, and driver maturity took freaking forever in comparison to the 780/780ti. Likewise, if you could go back and have the improved tessellation back before the 300 "series" launch, I think everything would have come out way differently.

 

Fiji is certainly competitive at 1440 and 4k in a number of games, stock for stock. OC to OC, no, I agree. 

 

Yep, the issue with the 290/X was indeed the reference cooler. But as of now, we all know the truth of how fast those cards really are, so why still say they are/were "behind" when we now know they weren't/aren't? To me, that is being dishonest because they aren't telling the whole truth. It taints the whole brand when it's not necessary at all.

 

 

So a few quick comments.

 

1. That Tom's Hardware review has some seriously bullshit spin on it. Yes it has a narrower bus than Hawaii, but literally everything but Fiji does. For good reason as with the improved color compression of Maxwell AND Tonga, it wasn't needed. At all.

 

BTW yes, literally every Maxwell card is bottlenecked a bit by their bus width, the 960 more than all the rest of the lineup. Hence why memory overclocking makes such a huge difference for Maxwell (even the 980, which is less bus limited than all but GM200, showed memory overclocking returns per Mhz to be nearly 87% of the core clock return per Mhz). But still even GM200 definitely didn't need a 512 bit bus, and it would have been a total waste to give it one.

 

2. So the graphs they show for gpu clock speed (and temps) make sense, but I don't think that was thermal throttling, that looked more like a driver glitch. I can hope that someone mistakenly assumed those dropped clocks were throttling when it clearly wasn't (because OBVIOUSLY Tonga doesn't thermal throttle at 75C) and so just didn't bother trying overclocking or something. IDK but it's bs.

 

3. I like DF, but I really don't like stock to stock reviews because imho it's stupid and no one should run at stock. Plus factory overclocks are never the same, and I wish DF would at least do something to standardize their clocks a little bit (so they don't have the stupid shit on the 2GB vs 4GB videos where the difference in factory overclock makes looking at the difference in performance way less obvious).

 

1. Yep, agree totally.

 

2. It's not a driver glitch. If Tonga/Antigua is anything like Hawaii and Grenada, there is a power regulation feature called power tune which is the bane of AMD GPU overclockers. It down-clocks the GPU core, even when its running nowhere near the thermal limit. It supposedly keeps temps a little lower so the fans on the GPU don't ramp up as high. It's primarily to keep noise down. My HIS 290 does it. The only way to stop this is to disable over drive in catalyst control center and use a separate OC program (like MSI AB or Sapphire Trixx) and set the power limit to max (+50). Once you do that, you can run any clocks you want and power tune will no longer downclock it. The problem is not all reviewers know about this (Jayztwocents doesn't). Not all AMD GPUs do this, however. My Gigabyte reference 290 doesn't. I'm guessing the sample Tom's tested had power tune interfering, where as Techpowerup's sample did not. They both tested two different brands (Asus and Sapphire), so that could be the case. Regardless, knowing how to properly OC and AMD card and knowing to compensate for power tune is kind of important. 

 

3. Believe it or not, those of us who do OC out GPUs are the minority of people who buy them. Most people buy them and run the stock, so it's better to know how they stack up stock for stock. Then, if you want to know how well the OC, you can do further research. LTT always tests their cards OC'd, so there are sources out there to see that info. But for the general PC gaming audience, I think DF testing at stock for stock is the better way, IMO. 

[Curufinwe_wins]

snip

Well at least Jay always throws power target up to max when testing gpus (and generally uses msi afterburner) so even if he doesn't "know" about the issue, he shouldn't encounter it much. Thanks for the info, I didn't know they rode the power target that hard at stock (BTW if it's anything like nVidia aftermarket, many of the models use significantly increased stock power targets out of the box).

 

I know people don't often OC, but I think that is literally the fault of reviewers and others not parroting out the benefits of overclocking. Everyone should oc, without exception (Even if they don't go balls to the walls).

 

I don't even want to get started on LTT right there. They did a stock to stock only x99 vs z170.... what a crock of crap. Plus I don't actually thing they are testing all of them oc'd anymore (and if so they certainly don't make it a point.)

 

See while other people do benchmarks, DF's visual method really explained to people why the 2GB 380/960 is a bad buy currently, and they would do a great job showing the true benefits of overclocking (which they randomly have in some videos, but I mean really)

[ivan134]

Well at least Jay always throws power target up to max when testing gpus (and generally uses msi afterburner) so even if he doesn't "know" about the issue, he shouldn't encounter it much. Thanks for the info, I didn't know they rode the power target that hard at stock (BTW if it's anything like nVidia aftermarket, many of the models use significantly increased stock power targets out of the box).

 

I know people don't often OC, but I think that is literally the fault of reviewers and others not parroting out the benefits of overclocking. Everyone should oc, without exception (Even if they don't go balls to the walls).

 

I don't even want to get started on LTT right there. They did a stock to stock only x99 vs z170.... what a crock of crap. Plus I don't actually thing they are testing all of them oc'd anymore (and if so they certainly don't make it a point.)

 

See while other people do benchmarks, DF's visual method really explained to people why the 2GB 380/960 is a bad buy currently, and they would do a great job showing the true benefits of overclocking (which they randomly have in some videos, but I mean really)

I'm not sure I agree. Your average lay person doesn't like tinkering with things they don't understand even if you show them how safe it is. I think the fact that outlets like iBuyPower, CyberPower, etc make a lot of money for building for people should be evidence of that. There are numerous tutorials for buildng a PC on YouTube, but a vast majority of people will still opt to pay someone else to do it for them.

 

So I agree with @MEC-777. Besides, you know well that not all coolers are created equally, some AIB partners don't unlock voltage, and even after those, you still have the silicon lottery. Having any one outlet do both seems like an enormous amount of work, so I'm glad that there are some that focus on stock and others that focus on OC.

 

My observation of DigitalFoundry (don't take this as fact) is that their aim is to take what a lot of people see as a complicated thing, and make it as simple as possible for the masses. There's enough room in the world for outlets like them, and outlets that cater to the enthusiasts.

[stconquest]

I am not the "must OC" type for the GPU or CPU... but I want that option.  I only OC the GPU when I come across a game that needs the juice... then I start pumping it up until I get my 60FPS (99% of the time).  Once I finish playing that game, it is back to stock clock for me.

[Curufinwe_wins]

I'm not sure I agree. Your average lay person doesn't like tinkering with things they don't understand even if you show them how safe it is. I think the fact that outlets like iBuyPower, CyberPower, etc make a lot of money for building for people should be evidence of that. There are numerous tutorials for buildng a PC on YouTube, but a vast majority of people will still opt to pay someone else to do it for them.

 

So I agree with @MEC-777. Besides, you know well that not all coolers are created equally, some AIB partners don't unlock voltage, and even after those, you still have the silicon lottery. Having any one outlet do both seems like an enormous amount of work, so I'm glad that there are some that focus on stock and others that focus on OC.

 

My observation of DigitalFoundry (don't take this as fact) is that their aim is to take what a lot of people see as a complicated thing, and make it as simple as possible for the masses. There's enough room in the world for outlets like them, and outlets that cater to the enthusiasts.

"Plus factory overclocks are never the same, and I wish DF would at least do something to standardize their clocks a little bit (so they don't have the stupid shit on the 2GB vs 4GB videos where the difference in factory overclock makes looking at the difference in performance way less obvious)."

 

My annoyance is that there is only one outlet that comes close to focusing on overclocks and that's Jayz. None of the others do more than a little tiny blurb to cover the bases. Or worse we have some that either completely ignore it (recent LTT ehem) and/or do it completely wrong.

 

I never said everyone will overclock. I said everyone SHOULD, and just like over the years building your own pc has become WAY more common due to reviewers/youtuber/word of mouth, likewise the benefits and marginal at best risk of overclocking SHOULD be parroted around.

[Curufinwe_wins]

I am not the "must OC" type for the GPU or CPU... but I want that option.  I only OC the GPU when I come across a game that needs the juice... then I start pumping it up until I get my 60FPS (99% of the time).  Once I finish playing that game, it is back to stock clock for me.

But why? You would rather run the game at a higher utilization (and temperatures) instead of a higher clock speed?

 

I mean you don't have to go crazy with overvolting or shit, but like throw in a modest one in 10 second and never look at it again.

[stconquest]

But why? You would rather run the game at a higher utilization (and temperatures) instead of a higher clock speed?

 

I mean you don't have to go crazy with overvolting or shit, but like throw in a modest one in 10 second and never look at it again.

 

I have never been worried about temps too much.  I constantly monitor and have good airflow in my case.  I don't see a problem with higher utilization of the GPU.

 

When it is needed, the clock goes up, voltage goes up... BSODs happen sometimes. =/  IDK, I guess I don't have the experience with a 24/7 GPU OC, so I figure the manufacturer's setup is good enough for that.

[ivan134]

"Plus factory overclocks are never the same, and I wish DF would at least do something to standardize their clocks a little bit (so they don't have the stupid shit on the 2GB vs 4GB videos where the difference in factory overclock makes looking at the difference in performance way less obvious)."

My annoyance is that there is only one outlet that comes close to focusing on overclocks and that's Jayz. None of the others do more than a little tiny blurb to cover the bases. Or worse we have some that either completely ignore it (recent LTT ehem) and/or do it completely wrong.

I never said everyone will overclock. I said everyone SHOULD, and just like over the years building your own pc has become WAY more common due to reviewers/youtuber/word of mouth, likewise the benefits and marginal at best risk of overclocking SHOULD be parroted around.

one of the best ways of doing that is reviewing individual cards from different AIB partners which a lot of publications like kitguru still do

[MEC-777]

Well at least Jay always throws power target up to max when testing gpus (and generally uses msi afterburner) so even if he doesn't "know" about the issue, he shouldn't encounter it much. Thanks for the info, I didn't know they rode the power target that hard at stock (BTW if it's anything like nVidia aftermarket, many of the models use significantly increased stock power targets out of the box).

 

I know people don't often OC, but I think that is literally the fault of reviewers and others not parroting out the benefits of overclocking. Everyone should oc, without exception (Even if they don't go balls to the walls).

 

I don't even want to get started on LTT right there. They did a stock to stock only x99 vs z170.... what a crock of crap. Plus I don't actually thing they are testing all of them oc'd anymore (and if so they certainly don't make it a point.)

 

See while other people do benchmarks, DF's visual method really explained to people why the 2GB 380/960 is a bad buy currently, and they would do a great job showing the true benefits of overclocking (which they randomly have in some videos, but I mean really)

 

You can crank the power limit in MSI AB for AMD cards, but if you don't disable overdrive in catalyst, it will not take the new power limit setting. Powertune will still override. I spent a lot of time trying to figure out what was going on until I found the official 290/X owners club thread on overclock.net where I learned a lot about OCing AMD cards and the not so well known trick of disabling overdrive. 

 

Jay knows to crank the power limit in MSI AB, but I don't think he knows to disable overdrive. That was pretty evident in his review of the Asus 290X matrix review when he said "I don't know why it's throttling" and the card was nowhere near it's thermal limit. I've tried to contact him about it several times now, but I have yet to get any response. 

 

OCing is more for the enthusiasts and for people who are interested in tinkering. The other thing about OCing is that no two cards are alike and for some, even trying to set a mild OC can create problems. 

 

I have two friends I've know all my life who are also PC gamers and know a fair amount about computers. They both work in networking/IT and one is a licensed electrician. They're running K series Ivy Bridge i5's and i7's on Z series motherboards (high end boards too, not entry level). One has a GTX 680 and the other has a 980Ti. After talking with them, it turns out neither of them are interested in OCing. Not even a little. They don't see the point and they don't want to "risk" it. Doesn't matter how much I try to explain how safe and easy it is, they don't care. Plug and play is all they want. (Why, then do they have $200+ OCing motherboards and K series CPUs? Because they "wanted the option". Smh every time. lol).

 

Anyways, my point here is that even among knowledgeable PC enthusiasts, many still don't OC. I'm not saying everyone is like my friends here, but I would say most PC gamers, most people who buy these cards and who would be looking at DF's comparison videos, won't OC. 

 

OCing is for those who care about and actively want to seek out that little bit more performance. I don't think everyone should OC, just because they can. It's a choice. Not everyone cares about getting an extra 5-10%. "Will it run the games I want to play at the settings I want to use? Yes? Then good enough".

[Dabombinable]

That's the classic Nvidia/AMD double standard right there. 

 

GTX 960 with 128bit bus - no complaints. 

R9 380X with 256bit bus - "it's a drawback".

 

What? Does not compute. 

 

Also noticed that Tom's were unable to get the card to OC without throttling and complained about it, yet Techpowerup OC'd their card just fine, benchmarked it and made no mention of throttling issues... hmmm...  

 

You can tell just by the wording, which reviews/reviewers hold more/less bias. That's usually why I don't bother reading all their comments and just skip straight to the numbers. Digital foundry's comparison videos are, by far, the best source for unbiased, real-world performance analysis, I come across yet. I'm sure they'll have some 380X videos up soon.

Lol, you must have missed all of the posts in multiple threads complaining about the GTX 960 4GB's 128bit bus. 128bit is fine for 2GB vRAM, as for 4GB vRAM-might as well use 4GB DDR3 instead of GDDR5.

[MEC-777]

Lol, you must have missed all of the posts in multiple threads complaining about the GTX 960 4GB's 128bit bus. 128bit is fine for 2GB vRAM, as for 4GB vRAM-might as well use 4GB DDR3 instead of GDDR5.

 

I was referring specifically to reviewers reactions to the 960 vs 380x. 

 

I'm well aware that lots of people are aware of the bus restriction on the 960.  

[Dabombinable]

I was referring specifically to reviewers reactions to the 960 vs 380x. 

 

I'm well aware that lots of people are aware of the bus restriction on the 960.

Oh....yeah I didn't see their reactions.