AMD's dx 11 CPU overhead is essentially gone (at least on Polaris)

[ybriK]

How did this even turn into the Faildriver architecture debate. When it's proven overtime that it simply cannot compete against any modern Intel CPU even against its' competition (Sandy Bridge). Worth mentioning also the significantly higher TDP whether stock or OC'd.

 

 

 

 

 

 

[Biggerboot]

@AMD

[Herman Mcpootis]

i just tested GTA 5 and metro last light redux on 16.12.1 and 17.1.2 drivers. the GTA 5 results seems to be generally several fps higher on the new drivers, metro was generally the same

[Morgan MLGman]

12 hours ago, Bananasplit_00 said:

nope, Vega wont be cheap like the Fury is, a new R9 Fury is $350, thats what i paid for my 970! Vega is probably going to be hitting close to what Pascal is doing and therefor priced similarly

An R9 Fury costs 284$ atm: http://pcpartpicker.com/product/bt8H99/xfx-video-card-r9fury4tf9

Though I've seen those go for as low as 239$ (Sapphire Nitro)

11 hours ago, Droidbot said:

You realise that the FX processor line is 6 years old, right? The 8350 bottlenecks the RX 480 and 1060 by a good margin even compared to a 26/2700k...

Actually, it doesn't in newer AAA titles. I had an FX-8350 @4,5GHz with an R9 290X (heavily OC'd) which is around this level of performance, didn't encounter bottlenecks even in CPU intensive games such as ROTTR/The Witcher 3, the GPU worked @100% without any issues.

Of course the performance wasn't what it would've been with an i7-7700K @5GHz, but the GPU wasn't held back in terms of dropping usage from 100% etc.

12 hours ago, Bananasplit_00 said:

this driver update would push up those FPS numbers if it affected the Fury too, makeing it even more of a great choice for < DX12 games, which most titles are currently. the Fury is increadably cheap used, around $200, because people think its as much of a dissapointment as at launch

The R9 Fury already can reach GTX 1070 performance levels if overclocked, I'm in the process of testing and benchmarking (I'm creating a video because some users actually asked me to after seeing 3DMark scores my setup does), if on this latest driver the performance is even better then I'm gonna have to re-do the benchmarks I already did and add those older scores (done on 17.1.1) to the current 17.1.2 driver. This is really an amazing card, I bought it at the time it costed less than a 1060 6GB and the same as a 480 8GB here in Poland and I've got no regrets, couldn't have chosen any better for the price.

12 hours ago, Majestic said:

But Fiji is still crap.

And why is that exactly? If I can reach performance of a 1070 FE at stock (in DX11) for less than I'd pay for a 1060 6GB and completely demolish any card within the price range in DX12 then why is Fiji crap?

12 hours ago, ivan134 said:

Seriously, go back and read the OP and watch the entire video because I'm going to start ignoring anything else you write here if you don't.

 

• Overwatch is in the video.
• 480 is the highest end GCN 4 GPU available.
• This test had not been done on older GCN cards and I asked people to post their own results.

Stop arguing for the sake of arguing.

I'm gonna do a video comparing scores with an older driver as well as regular FPS numbers with my R9 Fury (as I mentioned above), definitely gonna tag you if you're interested in the performance results, I'm very curious myself whether the driver update affects older GCN versions such as Fiji, although I don't see why it wouldn't as Fiji is still technically the highest-end AMD's lineup (until Vega releases, that is)

Though I'm not sure whether I have any noticeable CPU overhead in my system  as my CPU doesn't hold back the card at all (i7-6700K @4,7GHz), so the difference might be less noticeable than if I had an i3 or a low-end i5

10 hours ago, Dabombinable said:

An i5 6500 is better in all tasks than your 8370

You make good points but this one isn't 100% true. There are multithreaded tasks in which the 8370 wins with an i5-6500 by a significant margin as it's up to 34% faster in multithreaded workloads. Of course it's also around 32% slower in singlethreaded ones, I'm not forgetting that.  That doesn't change the fact that there just are certain scenarios in which an 8370 would win by quite a bit. Funnily enough, my old FX-8350 @4,5 was a tiny bit faster in Cinebench than my previous i7-3770. Also, a fun fact: your i5 does exactly half of a 4,7-4,8GHz i7-6700K in terms of Cinebench multithreaded score

[Bananasplit_00]

22 minutes ago, Morgan MLGman said:

An R9 Fury costs 284$ atm: http://pcpartpicker.com/product/bt8H99/xfx-video-card-r9fury4tf9

Though I've seen those go for as low as 239$ (Sapphire Nitro)

Actually, it doesn't in newer AAA titles. I had an FX-8350 @4,5GHz with an R9 290X (heavily OC'd) which is around this level of performance, didn't encounter bottlenecks even in CPU intensive games such as ROTTR/The Witcher 3, the GPU worked @100% without any issues.

Of course the performance wasn't what it would've been with an i7-7700K @5GHz, but the GPU wasn't held back in terms of dropping usage from 100% etc.

The R9 Fury already can reach GTX 1070 performance levels if overclocked, I'm in the process of testing and benchmarking (I'm creating a video because some users actually asked me to after seeing 3DMark scores my setup does), if on this latest driver the performance is even better then I'm gonna have to re-do the benchmarks I already did and add those older scores (done on 17.1.1) to the current 17.1.2 driver. This is really an amazing card, I bought it at the time it costed less than a 1060 6GB and the same as a 480 8GB here in Poland and I've got no regrets, couldn't have chosen any better for the price.

And why is that exactly? If I can reach performance of a 1070 FE at stock (in DX11) for less than I'd pay for a 1060 6GB and completely demolish any card within the price range in DX12 then why is Fiji crap?

I'm gonna do a video comparing scores with an older driver as well as regular FPS numbers with my R9 Fury (as I mentioned above), definitely gonna tag you if you're interested in the performance results, I'm very curious myself whether the driver update affects older GCN versions such as Fiji, although I don't see why it wouldn't as Fiji is still technically the highest-end AMD's lineup (until Vega releases, that is)

Though I'm not sure whether I have any noticeable CPU overhead in my system  as my CPU doesn't hold back the card at all (i7-6700K @4,7GHz), so the difference might be less noticeable than if I had an i3 or a low-end i5

You make good points but this one isn't 100% true. There are multithreaded tasks in which the 8370 wins with an i5-6500 by a significant margin as it's up to 34% faster in multithreaded workloads. Of course it's also around 32% slower in singlethreaded ones, I'm not forgetting that.  That doesn't change the fact that there just are certain scenarios in which an 8370 would win by quite a bit. Funnily enough, my old FX-8350 @4,5 was a tiny bit faster in Cinebench than my previous i7-3770.

cheapest Fury where i live is $350 sadly, if it was $250 id already have one lol, and yah the Fury is pretty amazing, people dont seem to get that. id love to see the vid so tag meh

[Majestic]

34 minutes ago, Morgan MLGman said:

And why is that exactly? If I can reach performance of a 1070 FE at stock (in DX11) for less than I'd pay for a 1060 6GB and completely demolish any card within the price range in DX12 then why is Fiji crap?

In terms of driver overhead. The Fury X is almost always next to the 480 and 980, whilst being much faster hardware.

That's driver overhead.

[Morgan MLGman]

30 minutes ago, Majestic said:

In terms of driver overhead. The Fury X is almost always next to the 480 and 980, whilst being much faster hardware.

That's driver overhead.

I agree that there is an issue here with driver overhead, though I disagree that it's next to the 480 and the 980, it's 9% slower than a 1070 on drivers 16.12.2 and the gap might be even lower with the latest 17.1.2 if it also improves performance on Fiji. Even my non-X Fury leaps ahead of an RX 480 by a significant margin, a 1500MHz/2250MHz RX 480 (which is a very good overclock) does around 15300 GPU score in FireStrike, my Fury at 1160MHz/550MHz does 17300 which is a stock 1070 FE score.

[Nena Trinity]

10 hours ago, Dabombinable said:

I just maxed everything out - and why would I set GTA V to load a single core? I'm frustrated as it that Skyrim switches between using 4 cores evenly and 1 core heavily (the FPS suddenly drop+it stutters badly). And faulty i5? PMSL, actually doing a comparison with my 4790K downclocked with HT disabled, it turns out that its performance was actually normal, also-face it, single threaded an FX 8350 is terrible (and its multi threaded scaling as well).

Let me guess you have never used the launch options to force Skyrim to use the desired amount of cores right?

 

Also GTA V does that since all settings affecting CPU goes directly on one thread... (tough its even worse for a i5-4440 because GTA V uses all cores quite a bit combined with those damned settings your core would obviously be overloaded...)

 

Edit: Overall single-threaded does not matter anymore for AAA titles... (if it did a 5GHz G3258 could play all!)

[Dabombinable]

2 hours ago, Morgan MLGman said:

You make good points but this one isn't 100% true. There are multithreaded tasks in which the 8370 wins with an i5-6500 by a significant margin as it's up to 34% faster in multithreaded workloads. Of course it's also around 32% slower in singlethreaded ones, I'm not forgetting that.  That doesn't change the fact that there just are certain scenarios in which an 8370 would win by quite a bit. Funnily enough, my old FX-8350 @4,5 was a tiny bit faster in Cinebench than my previous i7-3770. Also, a fun fact: your i5 does exactly half of a 4,7-4,8GHz i7-6700K in terms of Cinebench multithreaded score

Fine-an FX 8*** would be better for servers which aren't used 24/7, and file compression. And don't forget that the i5 is kind of clocked low-and is matched by a 2007 Xeon X5450 @ around 4.5GHz.

[Morgan MLGman]

7 minutes ago, Dabombinable said:

Fine-an FX 8*** would be better for servers which aren't used 24/7, and file compression. And don't forget that the i5 is kind of clocked low-and is matched by a 2007 Xeon X5450 @ around 4.5GHz.

Of course, remember that I'm not saying it's a better CPU to get, I'm just saying that those CPUs for certain scenario uses might actually be decent, especially if bought for a good price, you mentioned you could've gotten an 8350 + 990FX Sabertooth for 100$... That's a killer-deal imo

[Dabombinable]

1 hour ago, Nena360 said:

Let me guess you have never used the launch options to force Skyrim to use the desired amount of cores right?

 

Also GTA V does that since all settings affecting CPU goes directly on one thread... (tough its even worse for a i5-4440 because GTA V uses all cores quite a bit combined with those damned settings your core would obviously be overloaded...)

 

Edit: Overall single-threaded does not matter anymore for AAA titles... (if it did a 5GHz G3258 could play all!)

I do not like messing around with a game's multi threading as in the past, only a few have actually taken well to being forced to run on more cores/threads than they were designed for-all others get frequent CTD, corrupted saves and glitches. At most I simply change the affinity so that Skyrim uses 2 cores 2 threads (and then I only do that when I'm running overclocked to 4.7-4.8GHz)

[Nena Trinity]

2 minutes ago, Dabombinable said:

I do not like messing around with a game's multi threading as in the past, only a few have actually taken well to being forced to run on more cores/threads than they were designed for-all others get frequent CTD, corrupted saves and glitches. At most I simply change the affinity so that Skyrim uses 2 cores 2 threads (and then I only do that when I'm running overclocked to 4.7-4.8GHz)

Hmm I see but I normally set this in the launch options (( -cpuCount=4 -exThreads=3 -high -novid ))

[Dabombinable]

3 minutes ago, Morgan MLGman said:

Of course, remember that I'm not saying it's a better CPU to get, I'm just saying that those CPUs for certain scenario uses might actually be decent, especially if bought for a good price, you mentioned you could've gotten an 8350 + 990FX Sabertooth for 100$... That's a killer-deal imo

Yeah...but when I have multiple motherboards to repair (most of which are as far as I can tell, literally irreplaceable), and a constantly growing supply of CPU+RAM+HDD (I get given my families old stuff to tinker with-luckily they don't like OEM machines), I really can't afford to add much more clutter to my room+workshop. Even though I've got an FX 8350 cooler that's ATM useless to me (if I could only adapt it for SS7).

[Majestic]

 

@ivan134 Here they once again point out that the 480 is having issues on the G4560 when the 1060 isn't. In certain scenario's.

[Nena Trinity]

14 minutes ago, Majestic said:

 

@ivan134 Here they once again point out that the 480 is having issues on the G4560 when the 1060 isn't. In certain scenario's.

Ouch!

[-rascal-]

Going to run some testing with ARMA III regards to this later today...

I am thinking to using Rainbow Six Siege, but I believe it is a more GPU heavy game...but who knows.

Metro 2033 is just garbage...

 

Rocket League maybe (I'm getting mixed results on whether it is a CPU or GPU heavy game).

 

EDIT:

@ivan134
Programs/games I have decided to run this test with...so far:

• ARMA III (already did some tests)
• Rainbow Six Siege (did some tests)
• Resident Evil 5 (did some tests)
• Deus Ex Mankind Divided
• DiRT 3
• Bioshock Infinite
• Cinebench
• 3DMark Firestrike
• Unigine Valley

I wanted to test Rocket League, and DOOM, but I don't know if I can get FOUR consistent runs for EACH driver set since there are no in-game benchmark tools. I could probably fire up one of the Crysis (I've got Warhead, 2, and 3) games...and see how that goes. StarCraft II was another I was planning to test...again could be difficult to do.

 

I am doing 4 runs for per program/game, per driver.

As for which drivers I am using for comparison...

• Crimson 15.12
• Crimson 16.7.3
• ReLive 16.12.2 (currently installed on my X99 system)
• ReLive 17.1.2