MSI R9 390X GAMING vs ASUS STRIX R9 Fury Review @ [H]

Theres really no strange magic on why the 390x can match the R9 Fury. What we are seeing are games that aren't shader limited at that resolution. AMD made a bigger core to pack in a shit ton more texture power and SP power but didn't expand on their ROP engine at all and guess where we are mostly limited? ROP's have been gaming's biggest bottle neck since forever, it really isn't astounding to see that the R9 Fury can be matched by hardware with a similar engine... also the obvious bottle neck by having 4gb of memory only.

Nvidia took a chance (ok not really...) by lowering their SP's to fit more ROP's. Having a smaller number of SP's also happens to be easier to optimize for.
 
AMD's really gotta correct their prices. They could've had a 4870-like knockout to nVidia's complete lineup if they priced accordingly.

The Fury-X should have been $550, it's performance is close enough to the GTX980-Ti to sacrifice a little performance for a $100 saving. That was my philosophy when I picked a Radeon 4870 over a GTX260, and when I picked my R9-290 over a GTX970. Then you have the "value" proposition of two R9-Fuy-X's costing a little more than a single Titan-X while destroying it in performance.

The Fury would look a lot more attractive against the GTX980 were it $450. Driver improvements, DX12 and newer games may help it distances itself from the R9-390x.

With performance noticeably better than the GTX970, but also coming very close to to the Fury, the 390x should be a bit cheaper than the R9-Fury at $400. The price/performance would be a bit better than the GTX970, and with current performance being so close between it and the Fury in current games the $50 for the Fury will buy you a bit more future-proofing (if games use more tesselation) which may be worth it for some.

The R9-390 generally performs right on par with the GTX970 and R9-290x, and seems to be the only card in the line-up that is priced well. Some of the Custom cards are very nice, and seem to keep temps in check. This card is a fantastic alternative to the GTX970.

So, I think AMD has a giant gaping whole in their line-up. A hole that can be patched and probably give them the crown for price/performance. The Tonga-XT R9-380x. A 3GB($240) or 6GB($280), 384-bit full Tonga-GPU AMD's had on the back-burner for a while. With faster-clocks, more memory bandwidth, it would perform about on par with an R9-290.

The R9-380 would then contend perfectly with the GTX960 with 2GB version for $200 and 4GB versions going for $230.

An R9 3xx/Fury launch with those prices would have knocked nVidia in the knees and garnered much more positive critical praise. I just think back to the Radeon 4870 launch, had AMD priced that card at the same $450 the GTX260 was going for, I probably would have gotten a GTX260 or just stay with my 8800GT until the next round. I was glad that AMD priced it at $300, and it was indeed a substantial upgrade from the 8800GT.

Right now, I'm looking to upgrade from my R9-290, where the Fury $450, it'd have my attention, at $550, it's only $100 less than GTX980-Ti, which is a much faster card. The GTX980-Vanilla isn't really that much of an upgrade. I know I'm not in the only one in this situation, people with GTX780/780Ti/R9-290x/R9-290/Titan looking for a tangible performance upgrade really only have the GTX980-Ti to look at. The Fury-X can't really justify it's current price, despite it being a substantial performance upgrade from those older cards because of it's performance deficit to the GTX980-Ti.

BTW, I know their's a chart floating out regarding AMD's marketshare through the years. It's no coincidence that AMD saw it's biggest profits during the Radeon 4870 and Radeon 5870 era's, which mimicked price/performance ratio's vs nVidia very similar to what I proposed for the current line up. Makes you think.
 
I still wonder about 390X really improve tessellation ? (from article MSI R9 390X review in Witcher 3).
It will be very good if add XFX DD 290X (reference clock with no throttling) with Catalyst 15.7 into comparison.
Anyway nice review.
 
hmms, makes you wonder what the actual problem is with the Fury's Performance. I mean Fury pro has 3584 shaders and the 390x has 2816. Seems as if there is nearly zero scaling for adding more shaders.

People were talking about the ROP's, and how the 390x and the fury pro/Fury X all have 64 Rops. Perhaps this is the bottleneck.

I see the future looking rather grim for AMD.
 
AceCR42 said:
I guess one advantage of the Fury is it seems to draw about 120W less power than the 390X.
Click to expand...

yeah was going to say the same thing, but is it worth basically a dollar per watt compared to the 390x? i guess if you live in an area that has really high prices for electricity it would be worth it, but then why not buy the 980 or TI which is slightly lower power usage?
 
Almost every review out there shows the R9 Fury as 10-12% faster than R9 390X. To say the R9 Fury and R9 390X is on par is flat out lying. You guys have already decided to bash Fiji based Fury GPUs. So your reviews are pretty much cooked as the conclusion has already been decided. Anyway in your own testing R9 Fury shows its advantages with improved tesselation in Farcry 4 with enhanced godrays and whats more its playable.

http://www.techpowerup.com/reviews/ASUS/R9_Fury_Strix/31.html

http://www.hardwarecanucks.com/foru.../69792-amd-r9-fury-performance-review-19.html

http://www.computerbase.de/2015-07/sapphire-radeon-r9-fury-tri-x-oc-test/4/

http://hexus.net/tech/reviews/graphics/84512-sapphire-radeon-r9-fury-tri-x-oc/?page=12

http://www.guru3d.com/articles_pages/asus_radeon_r9_fury_strix_review,14.html

Anyway guys keep bashing the Fiji GPU to your hearts fill. :D
 
durquavian said:
well considering the Nano has just one 8 pin to the 390X's 2 8 pins, the performance aspect would be astounding.
Click to expand...

I fail to see what is "astounding." An extra 8 pin can mean anything from 1 watt to 150 watt difference. This tells us NOTHING. Overclock a cheap 290x and you can pay for years of extra power costs.
 
ebduncan said:
hmms, makes you wonder what the actual problem is with the Fury's Performance. I mean Fury pro has 3584 shaders and the 390x has 2816. Seems as if there is nearly zero scaling for adding more shaders.

People were talking about the ROP's, and how the 390x and the fury pro/Fury X all have 64 Rops. Perhaps this is the bottleneck.

I see the future looking rather grim for AMD.
Click to expand...

dont think its an issue with performance.
the dx11 cpu overhead does a bad thing for amd GCN tech as they havent poured the same resources to fix games. recently they build driver for win 10 that has helped them.
its doing good.

4gb real HBM works as good as 6GB gddr5 980ti.
AMD are a star trek miracle worker.
ROP isnt the bottleneck its drawcall cpu overhead.
 
Nice review. Should have thrown in the 290X as well, so people don't have to go jumping around the place to compare them. 2 years and only a few frames show for it. Makes me wonder what going on at AMD. I'm so glad I jumped on the NV GTX970 bandwagon.
 
Fiji isn't a top-to-bottom new design. It's more of a derivative of Hawaii and Tonga. All these designs have four Shader Engines with a maximum possible 16 ROPs each per engine. (Tonga has 8 ROPs per engine.) I think AMD went as big as they could while staying within the existing GCN "framework." So, 64 ROPs it is. This is also the reason for the HDMI 1.4a on Fiji; it's derived from existing GCN designs that offer that same version.

Can't prove it, but I'd bet a dollar that the Fiji design was finalized quite a long time ago. I suppose AMD had to do as much as possible with existing tech because I doubt they really wanted to break any new ground at 28nm, especially given their financial situation. And that same financial situation, or so it seems to me, is probably why both Fury cards are priced as they are. These things are just different enough, mostly due to the HBM, that there were probably some pretty significant costs incurred putting Fury into production. Simple truth could be that AMD might not be able to lower the retail price on either model all that much as a result.
 
Rauelius said:
AMD's really gotta correct their prices. They could've had a 4870-like knockout to nVidia's complete lineup if they priced accordingly.

The Fury-X should have been $550, it's performance is close enough to the GTX980-Ti to sacrifice a little performance for a $100 saving. That was my philosophy when I picked a Radeon 4870 over a GTX260, and when I picked my R9-290 over a GTX970. Then you have the "value" proposition of two R9-Fuy-X's costing a little more than a single Titan-X while destroying it in performance.

The Fury would look a lot more attractive against the GTX980 were it $450. Driver improvements, DX12 and newer games may help it distances itself from the R9-390x.

With performance noticeably better than the GTX970, but also coming very close to to the Fury, the 390x should be a bit cheaper than the R9-Fury at $400. The price/performance would be a bit better than the GTX970, and with current performance being so close between it and the Fury in current games the $50 for the Fury will buy you a bit more future-proofing (if games use more tesselation) which may be worth it for some.

The R9-390 generally performs right on par with the GTX970 and R9-290x, and seems to be the only card in the line-up that is priced well. Some of the Custom cards are very nice, and seem to keep temps in check. This card is a fantastic alternative to the GTX970.

So, I think AMD has a giant gaping whole in their line-up. A hole that can be patched and probably give them the crown for price/performance. The Tonga-XT R9-380x. A 3GB($240) or 6GB($280), 384-bit full Tonga-GPU AMD's had on the back-burner for a while. With faster-clocks, more memory bandwidth, it would perform about on par with an R9-290.

The R9-380 would then contend perfectly with the GTX960 with 2GB version for $200 and 4GB versions going for $230.

An R9 3xx/Fury launch with those prices would have knocked nVidia in the knees and garnered much more positive critical praise. I just think back to the Radeon 4870 launch, had AMD priced that card at the same $450 the GTX260 was going for, I probably would have gotten a GTX260 or just stay with my 8800GT until the next round. I was glad that AMD priced it at $300, and it was indeed a substantial upgrade from the 8800GT.

Right now, I'm looking to upgrade from my R9-290, where the Fury $450, it'd have my attention, at $550, it's only $100 less than GTX980-Ti, which is a much faster card. The GTX980-Vanilla isn't really that much of an upgrade. I know I'm not in the only one in this situation, people with GTX780/780Ti/R9-290x/R9-290/Titan looking for a tangible performance upgrade really only have the GTX980-Ti to look at. The Fury-X can't really justify it's current price, despite it being a substantial performance upgrade from those older cards because of it's performance deficit to the GTX980-Ti.

BTW, I know their's a chart floating out regarding AMD's marketshare through the years. It's no coincidence that AMD saw it's biggest profits during the Radeon 4870 and Radeon 5870 era's, which mimicked price/performance ratio's vs nVidia very similar to what I proposed for the current line up. Makes you think.
Click to expand...

overclock the 980 to make it a true upgrade... That's what i did lol. 1570 mhz makes it extremely competitive with even an overclocked fury.
 
Last edited:
Stoly said:
Considering that the Fury performs relatively worse at lower res, I wonder if the 390X beats it at 1080p
Click to expand...

I agree! Looking at best 1080P 144Hz gaming... and then of course... rather curious if Win 10/DX12 brings other 'improvements' that can't be seen with these new cards (and Nvidias) that really starts to show the difference in the tech? if any?

Nice informative review!
 
Flopper said:
dont think its an issue with performance.
the dx11 cpu overhead does a bad thing for amd GCN tech as they havent poured the same resources to fix games. recently they build driver for win 10 that has helped them.
its doing good.

4gb real HBM works as good as 6GB gddr5 980ti.
AMD are a star trek miracle worker.
ROP isnt the bottleneck its drawcall cpu overhead.
Click to expand...


Cpu overhead will not prevent scaling from additional shaders. If it did then a faster cpu will show positive scaling which is not the case. Pretty much the same fps between a 3770k and a 4770k.

4gb HBM is exactly 2 gb less than the 6gb gddr5 on the 980ti. HBM may provide more bandwidth but it still has less capacity. So if you run out of vram it will become a slide show.

From Nvidia they increased the ROP to 128 and well there are clear gains over the 980(64) and 970(56) compared to the Titan x and 980ti.

As much as I'd like a new driver to magically fix all their issues, I do not see it happening. Rest assured if that does happen you can quote this post and make me eat some crow.
 
Modred189 said:
The million dollar question, Kyle/Brent is: WHY are the numbers so close. The vram can't be making the difference, I'd think, at these resolutions, and if the Fury is the next gen flagship GPU (As evidenced by its tessellation performance), why is it (relatively) losing out? Is it the OC'd clocks of the 390X? Immature drivers?
Click to expand...

Perhaps the games are ROP-limited? Every other aspect of Fury was improved, but the ROPs are untouched.

It's an unbalanced architecture, which is why it's been panned. Performance is too inconsistent from one game to another at 1440p.

Only at 4k does the memory bandwidth advantage assert itself.
 
Trimlock said:
Theres really no strange magic on why the 390x can match the R9 Fury. What we are seeing are games that aren't shader limited at that resolution. AMD made a bigger core to pack in a shit ton more texture power and SP power but didn't expand on their ROP engine at all and guess where we are mostly limited? ROP's have been gaming's biggest bottle neck since forever, it really isn't astounding to see that the R9 Fury can be matched by hardware with a similar engine... also the obvious bottle neck by having 4gb of memory only.

Nvidia took a chance (ok not really...) by lowering their SP's to fit more ROP's. Having a smaller number of SP's also happens to be easier to optimize for.
Click to expand...

both NV and AMD are guilty of this and i have said for a wile now ROP performance / number WILL be the limiting factor for 4k and VR


tl;dr MOAR ROPS
 
ROShay said:
Fiji isn't a top-to-bottom new design. It's more of a derivative of Hawaii and Tonga. All these designs have four Shader Engines with a maximum possible 16 ROPs each per engine. (Tonga has 8 ROPs per engine.) I think AMD went as big as they could while staying within the existing GCN "framework." So, 64 ROPs it is. This is also the reason for the HDMI 1.4a on Fiji; it's derived from existing GCN designs that offer that same version.

Can't prove it, but I'd bet a dollar that the Fiji design was finalized quite a long time ago. I suppose AMD had to do as much as possible with existing tech because I doubt they really wanted to break any new ground at 28nm, especially given their financial situation. And that same financial situation, or so it seems to me, is probably why both Fury cards are priced as they are. These things are just different enough, mostly due to the HBM, that there were probably some pretty significant costs incurred putting Fury into production. Simple truth could be that AMD might not be able to lower the retail price on either model all that much as a result.
Click to expand...

called this one to
blame Mantle
AMD bet the farm on Mantle and lost
 
I love everyone telling me how the Fury will get so much better with overclocking. I know a couple people that are positive it will hit 1200mhz easily at the minimum "as soon as the voltage is unlocked". And then I read this:

while overclocking on the R9 Fury still remains to be somewhat retarded.
Click to expand...

:eek:

I'll stick with my 290X.
 
enterprise24 said:
I still wonder about 390X really improve tessellation ? (from article MSI R9 390X review in Witcher 3).
It will be very good if add XFX DD 290X (reference clock with no throttling) with Catalyst 15.7 into comparison.
Anyway nice review.
Click to expand...

I'd love to see this too, especially in Crossfire. As a 290X owner, I'm looking at the 980Ti, 390X, Fury X etc, and wondering if my best choice for an upgrade isn't just buying a 2nd 290X. The Cat15.7's seem to have given the 390X a nice boost, I'd be curious to see where the 290X stands now.
 
LigTasm said:
I love everyone telling me how the Fury will get so much better with overclocking. I know a couple people that are positive it will hit 1200mhz easily at the minimum "as soon as the voltage is unlocked". And then I read this:



:eek:

I'll stick with my 290X.
Click to expand...

In the main Fury review topic Kyle mentioned that he heard not to expect much for overclocking with the Fury right now. I wonder if voltage unlocking will really do a lot for either Fury card. I hope so, but I'm not expecting miracles.
 
dewbak75 said:
I'd love to see this too, especially in Crossfire. As a 290X owner, I'm looking at the 980Ti, 390X, Fury X etc, and wondering if my best choice for an upgrade isn't just buying a 2nd 290X. The Cat15.7's seem to have given the 390X a nice boost, I'd be curious to see where the 290X stands now.
Click to expand...

290x = 390x. It's the same damn chip clocked slightly faster, with faster-clocked ram.

Why is there so much confusion?

If you saw improvements in the 390x with a new driver update, you'll see similar improvements with your 290x and the same driver.
 
Derangel said:
In the main Fury review topic Kyle mentioned that he heard not to expect much for overclocking with the Fury right now. I wonder if voltage unlocking will really do a lot for either Fury card. I hope so, but I'm not expecting miracles.
Click to expand...

AMD said the fury/fury x cards will be the best overclockable cards on the market... so when they get unlocked we will see.. I would be happy to see my fury x OC to +25% Frame performance
 
Can anyone actually buy a Fury yet? When are these things supposed to actually be around?
 
Thanks for another great review, guys.

Bought a Gigabyte GV-R929XOC-4GD R9 290X used on eBay for $219 shipped for my secondary rig...could not be happier with it especially after the 15.7 driver release (big tessellation improvements).

AMD's R9 290 series overall is in the toilet in the used market, guys - take advantage! Total buyers market. As you can see, the R9 390X a.k.a. tweaked 290X performs very close to the latest stuff. Although I'm sure AMD will improve things over time with drivers...I just can't see it changing that drastically.
 
Nihilus1 said:
Sorry, but that is completely false. There are countless reviews showing that the 4 GB HBM is NOT the bottleneck for 4k. These guys had a problem with 1440p with Dying light, but everything else has shown the Fury X to match the 980ti at 4k. Even with crossfire playing 4k, the 4 GB memory was not an issue. Please stop with the B.S.

Awesome review BTW.
Click to expand...

Thats true, 4k is limited by the horsepower of the video card...(Fury) put two Fury's in and it does 4k no problem. The Fury has a huge ass memory bandwidth, like a huge pipe instead of sucking through a straw with GDDR5...DirectX12 may change this more, although current games are just written for current video cards and not HBM type video cards..
 
durquavian said:
well considering the Nano has just one 8 pin to the 390X's 2 8 pins, the performance aspect would be astounding.
Click to expand...

It was the 290x, not the 390x...with better performance than the 290x and a TDP of 175w thats quite amazing.. well for AMD it is..
 
ravensteel said:
Thats true, 4k is limited by the horsepower of the video card...(Fury) put two Fury's in and it does 4k no problem.
Click to expand...

From what I've read, Fury X in crossfire is extremely fast, better than highly overclocked 980ti's in SLI. The scaling is terrific and there are rumors the Fury X utilizes memory stacking in crossfire.
Do a Google search and you can't find a single review anywhere that shows the 980ti or Titan X in SLI beating crossfire Fury X's.
 
Yeah it seems these guys are being quiet on the Fury Crossfire review, asked multiple times and havent heard a response on when the review might be coming, might be hard to get a hold of two cards, but lesser sites seem to have done it. Just want to see the [H]ard perspective on Fury crossfire performance. May make them eat crow though, as the scaling is much stronger than Nvidia's and paint a different picture than what they have currently established in the handful of games they test.
 
psyclist said:
Yeah it seems these guys are being quiet on the Fury Crossfire review, asked multiple times and havent heard a response on when the review might be coming, might be hard to get a hold of two cards, but lesser sites seem to have done it. Just want to see the [H]ard perspective on Fury crossfire performance. May make them eat crow though, as the scaling is much stronger than Nvidia's and paint a different picture than what they have currently established in the handful of games they test.
Click to expand...

It wouldn't really be "eating crow."
1- The reviews to date are what they are based on the numbers. Different performance at different resolutions wouldn't change that.
2- The [H] crew has always been willing to be pleasantly surprised.
3- Good scaling != good performance.

That said, I've seen the other reviews out there and would also like to see the [H] treatment. I read articles these days, and as soon as they mention synthetic benchmarks I have a tendency to just stop reading.
 
sk3tch said:
Thanks for another great review, guys.

Bought a Gigabyte GV-R929XOC-4GD R9 290X used on eBay for $219 shipped for my secondary rig...could not be happier with it especially after the 15.7 driver release (big tessellation improvements).

AMD's R9 290 series overall is in the toilet in the used market, guys - take advantage! Total buyers market. As you can see, the R9 390X a.k.a. tweaked 290X performs very close to the latest stuff. Although I'm sure AMD will improve things over time with drivers...I just can't see it changing that drastically.
Click to expand...

No doubt on this. It's basically a free-for-all out there if you want a used 290x for the next couple weeks while people "upgrade" - a card that is essentially a 390x with 4gb of VRAM.

And yea, 15.7 came with a great amount of features and fixes. Freesync compatibility and the improved profiles pretty much made Crossfire viable again for me.
 
Have you guys had a chance to test Fury or Fury X under Eyefinity yet?
 
ravensteel said:
Thats true, 4k is limited by the horsepower of the video card...(Fury) put two Fury's in and it does 4k no problem. The Fury has a huge ass memory bandwidth, like a huge pipe instead of sucking through a straw with GDDR5...DirectX12 may change this more, although current games are just written for current video cards and not HBM type video cards..
Click to expand...

so does gddr5 but both are stuck on the pci-e bus
 
Some comments on testing:

Why use Witcher 3 to compare AMD/Nvidia hardware when it has one vendors code that hinders the other due to a particular design strength?
  • You can use Hairworks to improve quality, it is an option with AMD hardware, just control Tesselation to something less than 64x. At 16x most if not all will be hard pressed to find any difference in quality over the default setting.
  • How is HardOCP prevented from using hairworks on this review? Maybe stance on not touching driver settings (at least explore and tell the readers some of the options in both camps) needs to reviewed or expanded upon after the base testing. For example having better IQ hair in Witcher 3 and yet performance to go along with it is possible even over Nvidia solutions due to a driver option.
  • I am both an Ethusiast as well as IQ seeker - I do use driver settings to give myself a better experience - If one uses other programs to overclock and that is acceptable for reviews then why not enhance and use features already incorporated in the drivers?
  • If Nvidia or AMD has a very cool feature in the drivers that the other does not - please expose it.

Maybe HardOCP and others just had no go luck or it caused more problems then it is worth using driver features and exposing them. AMD drivers now have frame rate control, it works great on evening out the frame rate to prevent great changes in frame rates which would go along with a CFX review if ever done.
 
psyclist said:
Yeah it seems these guys are being quiet on the Fury Crossfire review, asked multiple times and havent heard a response on when the review might be coming, might be hard to get a hold of two cards, but lesser sites seem to have done it. Just want to see the [H]ard perspective on Fury crossfire performance. May make them eat crow though, as the scaling is much stronger than Nvidia's and paint a different picture than what they have currently established in the handful of games they test.
Click to expand...

What exactly would we have to eat crow about Fury CrossFire since we have not tested it yet nor commented on its performance?
 
Previous to any Fury CrossFire testing we are going to go back and fully test Fury and Fury X at 4K resolution, even though we do not suggest any Team Green or Red single GPU cards to our readers for 4K gaming. Hopefully that will keep all the whiners off our backs for a while. It will be great to spend all the resources on that to say card A is faster than card B at 4K resolution but we do not suggest either of those be used for 4K gaming. Makes sense to me...
 
Conman said:
Please don't waste time on whiners when you could spend it on new content.
Click to expand...

No, when it continually gets spun by these people that we are actively going out of our way to show one or the other look bad, I feel as though we need to answer those people, whether or not they have actually thought out what they are saying.

It is what it is. We do react to our readers, good, bad, or otherwise.
 
Conman said:
Please don't waste time on whiners when you could spend it on new content.
Click to expand...

I've been called worse :D

It's a good call, Kyle. Plus, you will already have the single card data to evaluate scaling performance for your 4k Fury crossfire review.
 
CPO_Sharky said:
I've been called worse :D

It's a good call, Kyle. Plus, you will already have the single card data to evaluate scaling performance for your 4k Fury crossfire review.
Click to expand...

Unless the drivers change, or unless a game gets patched....every time that happens, all that old data goes out the window. We do not show you old data ever, not like some other sites will do.
 
CPO_Sharky said:
At the risk of sound corny, that means a lot. Major props to you and your staff, you've earned a new fan here.
Click to expand...

That is pretty much the way we have done it for the last 15 years. Nothing new here. But thanks.

If we are not producing content that our readers care about reading, analyzing and arguing about, then we are not doing our jobs.

That said, I think the 4K testing will be a huge waste of time when it comes to actually suggesting single-GPU cards for 4K....but maybe we will be amazed. I am glad we have put it off for a bit as we got that new driver last week, so now there will be none of that "wait for the real drivers" whining bullshit.
 
DPI said:
Oh they'll find a way. Someone will always try to rationalize that "the methodology must be flawed" when the results don't fit their expectations. And that goes for both camps.

So it'll be either "a new beta driver just popped up on an FTP somewhere 5 minutes ago, WAI DIDNT U GUYS YOOZ THAT?" or "Why did you guys use the newest unstable driver, when the older driver has better performance for XYZ game?". So in the end all you can do is be the adults in the room and use your best judgment.
Click to expand...

Which it seems [H]ard is doing, but you, not so much.......
 
You must log in or register to reply here.
Back
Top