15.4 vs 1018.1 "Vbs" Benchmarks, Quad and Dual Core

Which missing crossfire profiles? Tell me. Yes the improvements are quite good, GTA V is much more smooth. (Among other games) Moved from 15.4b as well, you have nothing to worry about.

And pill, it's not the fps that counts the most it's the ms (frametime) in these drivers (and they're quite significant).

And yeah klaupe pretty much explained it.

 

Well what do u expect, it's a dual core? What's the moral of the story? Buy a quad? I don't get it...

 

There is no moral, this driver improves overhead/drawcalls for dx11 multithreaded applications, so if you have a quad core / +HT, you'll benefit. Signifcantly reduces the ms.

So if you don't have these drivers yet, I'd suggest you move on to them fast.

That's the moral of the story or point or whatever.

 

Pill, check out the quad core deviation, it is huge. When comparing dual to dual core the new driver is actually worse frame time, but on quad it is twice more fluid

 

Where do I find this vbs driver and does it work on 7?

 

may be this driver are "overall" best of 15.4, but for example with wolfenstein old blood i have artifacts and texture flashing, mortal kombat X stuttering and microfreeze... 15.4 contains optimization for latest games that this driver does not contain so try your games and see the results. Tested on 7870 XT boost edition 1150/1500mhz (tahiti le) on i5-4440.

 

Well DX12 is optimized for at least 4 cores so you should be comparing 4/6/8 cores.
2-4 won't really prove anything other than dual cores aren't fit for the job.....which isn't news.

seems a moot point tbh as what gamer will pair up a dual core with a decent GPU ?


Anyway GTA V has just finished installing so I must run along....have fun.

Gonna run it on W10 if anyone wants a report later...

 

PrMinisterGR, first of all, thanks for taking the time to bench these! Your efforts are really appreciated.

Since we are trying to reach the best way to compare results, my 2¢:

• Try to graph everything starting at zero, or at least keep the y-range in a significant interval, like 10+ fps or ms. Otherwise the graphs are just like a magnifying glass of the bar tips, which do not represent meaningful information. Remember that when someone looks at a bar, intuitively compares the areas between bars to gauge results, so a bar with double the area would represent a "two times" better result. This is why starting at zero is always a good idea.
• Do not use an inverted y-axis for the millisecond graphs, as this representation hints at negative values, where the biggest bar would be the most negative value and thus the "lesser" value.
• In the latency/frame plots, please use a degree of transparency for the closest plot, so that it doesn't cover the other plot.
• I would prefer VAR.S instead of STDEV.S for the variance results, but that's just my personal opinion (I think our own perception of variability is definetely non-linear, that's why we are so sensitive to stuttering).
• It could also be argued that all values should be rounded to no decimals. Are sub-fps differences really perceptive? Or microseconds? As a rule of thumb I would round the FPS to integers and just keep one decimal in latency results.

 

Compare Dying Light QUAD CORE deviation of 15.4, to QUAD CORE deviation of 1018.1. It´s almost doubled.

I don´t know how to explain any further.

 

I think Pill slightly misunderstood.. I think he thought there was some point between the dual core and quad yet the point we're actually trying to show here is just the overhead/drawcall improvements in multithreaded dx11 applications with the newest win10 drivers (running on 8.1) lol. Nothing else, nothing more.

Pill, did you run the game on 7/8.1 as well? Would've had a comparison point (although it may not have been as prominent for you as it is for me because I'm running crossfire, but still, something should be seen).

I also have texture flashing and blue checkers in old blood, must be the drivers (didn't have this with the 15.4bs), Mortal Kombat I don't know, don't play that anyways.

 

BTW., found space for a WIN 10 partition, will bench a little bit tomorrow

 

Yaay, lol, maybe your pCars thing will get solved then, certainly hope so. Would be cool.

 

me too, with 15.4 old blood runs fine.

 

I did that so that the differences could be perceived easier. I'l probably follow Vbs' advice and put everything on a zero axis. In the comment for every bench you can actually see the felt performance too.

I'll try to borrow the Steam Account from a friend who has it. I can't make any promises. I used the Benchmark function in RadeonPro, and for the games that RadeonPro wasn't working (Project CARS) I used the benchmark function in FRAPS.

1-2ms in the 99% of the frametimes is actually a huge gap that you can feel in the game. The framerates are the same for most games, the frame variation is usually better for the 1018.1 driver in a quad core scenario. I haven't posted anything in any other forum except this one.

Any chance you might give some actual numbers? 8-12fps sounds huge, I never really found differences that big. The frametimes for some games were much smoother though.

The moral of the story is that these drivers actually worsen the dual core + ht performance (which is crucial for cheaper systems), but they are better at smoother frame deliver for quad and up, compared to 15.4.

I'll do that for all the framerate graphs.

That is the purpose of it. The smaller bar is the "greater" value, as it indicates less frame "step" variation (in the standard deviation test), and more smoothness in the 99% test.

I will incorporate all of these. I am still not precisely sure about the standard deviation vs variance thing, to be honest. You might be right about linearity, I'll try to keep both values.

Thank you for the feedback, now I need time (and endless uploads to imgur).

 

When i promise i deliver.

When i installed the 1018.1 Drivers i didn't notice any boost.
Yesterday i installed Windows 10 and i did use the same replay for the benchmark
So the events was exactly the same between Windows 8.1 and Windows 10.

And like a user of here of Guru3D, i too recieve a boost in performance using the Windows 10.
I can't imagine if SMS and AMD even fix the game in order to get even better framerate

 

PrMinisterGR, you write the most knowledgeable, detailed and comprehensive reviews.
I truly respect and applaud your efforts.

 

The guy who started 2Cores+HT vs 4Cores +HT should be shot.

j/k......

Srsly... the interesting issue for those who have 4670 and 4770 is 4Cores vs 4Cores+HT.
How many fps is 4670 owner losing for not having HT, should he upgrade.
4770 owner would like to see HT off/on results.
No i7 owner will run with 2Cores+HT... wth.

The issue for i3 owner is whether he should upgrade to i5. He's not about to upgrade to $300 i7 when $190 CPU gives 90% of gaming perf.

Lets recap: No one gives a flying hoot about i3 vs i7.
It's a waste of benching time which is better spent in other areas. Just a suggestion.

 

please do let me know how the performance for gta5 is on windows 10

thank you in advance mate

 

This is an awesome finding! Unfortunately I don't have a spare partition for Windows 10 at this moment. Any change you might use FRAPS to record the runs, and send me the excel files?. I would love to see a frametime plot between the drivers in Win 8.1 and Win 10.

The only recap I can give you is for you to read the DX11 Overhead thread, this is where this thread derives from. The reason I run this benchmarks with a simulated i3 was to study the driver overhead with lower CPUs. If you have anything constructive to say, apart from belittling all of us without any kind of reason, please do so.

EDIT: I have updated the original post following Vbs' math suggestions, and the clarity suggestions from others.

 

+1

Your framerates are below 60 fps, I'm guessing you're GPU-bound? Cause if you're already maxing the GPU usage it'd make sense why you're not seeing ~10 fps more.

I tested the game with 290X and 970 in GTA V and I could clearly see the difference. Looking at Trevor's house in the desert for example, I'd get 60 fps with the 970, 50 fps with the 290X and it wasn't fully utilized so it's not GPU's fault for sure.