Thief 4 CPU benchmarks

[Gikaseixas]

Consider the lack of L3 cache on the A10 another reason why the performance compared to the FX 8150 was impressive. It has nothing to do with FX 8350. That's what I got from Headfoot's post. Poor attempt to undermine the Piledriver OC scaling compared to Bulldozer... Fail

 

[frozentundra123456]

Not sure how one can call the game well threaded or a good reflection on amds architecture when both tech spot and game.gpu show an i3 faster than FX8350.

 

[davie jambo]

So Techspots figures are bollocks then , seeing you get higher fps with a 8350 & 7950 than their test with a 8350 and R9 290x

 

[AtenRa]

Well i dont get higher fps than techspot at default in Max settings, but my 1GHz HD7950 gets almost the same fps as 290X. Thats not suppose to happen at GPU bound settings.

 

[SPBHM]

Consider the lack of L3 cache on the A10 another reason why the performance compared to the FX 8150 was impressive. It has nothing to do with FX 8350. That's what I got from Headfoot's post. Poor attempt to undermine the Piledriver OC scaling compared to Bulldozer... Fail

exactly, even without the l3 it can keep up with the 8150, and the 8350 with l3 and more modules is 20% or so faster... but as I said, I don't know if the difference (between the 8350 and 5800K) to the 8350 is a sign of "decently threaded", or just the l3 (or both) since this game seems to perform the same on i5s and i7s.

So Techspots figures are bollocks then , seeing you get higher fps with a 8350 & 7950 than their test with a 8350 and R9 290x

not sure what you are trying to say here

this benchmark is GPU and CPU bound depending on the place and CPU,

Techspot 1200P "maximum" with the 8350 and 290x = 47FPS
Techspot 1200P "maximum" with the 8350@4.5GHz and 290x = 53FPS
PClab 1080P "Very high" with the 8350 and 290x = 45
PClab 1080P "Very high" with the 8350@5GHz and 290x = 54
AtenRa 1080P "maximum" with the 8350 and 7950 OC = 45

looks CPU limited, faster VGA not helping.

 

[AtenRa]

not sure what you are trying to say here

this benchmark is GPU and CPU bound depending on the place and CPU,

Techspot 1200P "maximum" with the 8350 and 290x = 47FPS
Techspot 1200P "maximum" with the 8350@4.5GHz and 290x = 53FPS
PClab 1080P "Very high" with the 8350 and 290x = 45
PClab 1080P "Very high" with the 8350@5GHz and 290x = 54
AtenRa 1080P "maximum" with the 8350 and 7950 OC = 45

looks CPU limited, faster VGA not helping.

Well, if i reduce the GPU frequency to 810MHz and Memory at 1250MHz with the Core i7 3770K OC to 4444MHz i get 41,5fps(picture bellow).

Now, FX8350 at default with the HD7950 @ 1GHz is faster producing 45,6fps. Using the R9 290X even with the FX8350 at default it should have more than 50fps that davie jambo got with his HD7970 1GHz card.

At maximum settings the game is very GPU bound, the 290X even with the FX8350 should have produce more than 47fps. I strongly believe Techspot may accidentally used double or triple buffering at vsync when using the FX8350.

 

[davie jambo]

I see what your saying about it being CPU bound , but that still does not explain why on lesser hardware two of us have got better benchmarks than they did.

How much more powerful is a R9 290x over a 7970 ? 10-15% ?

 

[Headfoot]

I know, but I assumed you think the game is "decently threaded" because of the difference between the 8350 and 5800K, but as I said, you also need to consider the l3 cache.

You assumed incorrectly.

I looked at the CPU charts in the top right of the screen, seeing usage across more than 2 cores as my basis for saying it is "decently threaded." More than 2 cores is decently threaded in my book, if not well-threaded. IMO, single-thread modern games are crap, 2 thread games are almost inexcusable for major publishers (eg Skyrim) and 3-4 threads is decent. 6-8 threads is pushing the envelope.

Like Gika said earlier, the fact that there is no L3 cache on the apu and that it still catches up with the 8150 really highlights how much of a turd Bulldozer was until Piledriver came around

 

[Headfoot]

Consider the lack of L3 cache on the A10 another reason why the performance compared to the FX 8150 was impressive. It has nothing to do with FX 8350. That's what I got from Headfoot's post. Poor attempt to undermine the Piledriver OC scaling compared to Bulldozer... Fail

Thank you, this is exactly what I was saying

 

[AtenRa]

First impressions with AMD Catalyst 14.3 Beta

Feature Highlights of The AMD Catalyst™ 14.3 Beta V1.0 Driver for Windows®

• • Thief:
    • AMD Mantle and AMD True Audio support
    • Improves stuttering observed in CrossFire mode
  • Call of Duty: Ghosts: QUAD CrossFire profile update – improves level load times
  • Audio issues observed when using CrossFire configurations (and V-sync enabled) have been resolved
  • BattleField 4: V-sync issues observed on CrossFire configurations (with Mantle enabled) have been resolved

System 2
Core i7 3770K (Default Clocks)
ASUS Maximus V Gene
2x 4GB DDR-3 1600MHz 9-9-9 Kingston LoVo
Win 8.1 64Bit

HD7950 @ 1GHz

Video Settings set to High, No SSAA, No FXAA, 16x AF
Cat 14.2 = 56,10 fps
Cat 14.3 = 65,10 fps

Video Settings set to Max, 16x AF
Cat 14.2 = 47,70 fps
Cat 14.3 = 49,60 fps

 

[AtenRa]

I have updated post 79 with two more CPUs, The Kaveri A10-7700K and Core 2 Quad Q9450 both at default and OC.

Edit : systems 5 and 6 added (31-3-2014)

System 5
A10-7700K (Default Clocks)
ASROCK FM2A88X-ITX+
2x 4GB DDR-3 2133MHz 11-12-11
Win 8.1 64Bit

OC @ 4.3GHz, NB = 2200MHz

System 6
Core 2 Quad Q9450 (Default Clocks 2,66GHz)
ASUS Rampage Formula
2x 2GB DDR-2 1066MHz
Win 7 64Bit

OC @ 3.2GHz, DDR-2 @ 1003MHz

 

[SPBHM]

I have updated post 79 with two more CPUs, The Kaveri A10-7700K and Core 2 Quad Q9450 both at default and OC.

Edit : systems 5 and 6 added (31-3-2014)

System 5
A10-7700K (Default Clocks)
ASROCK FM2A88X-ITX+
2x 4GB DDR-3 2133MHz 11-12-11
Win 8.1 64Bit

OC @ 4.3GHz, NB = 2200MHz

System 6
Core 2 Quad Q9450 (Default Clocks 2,66GHz)
ASUS Rampage Formula
2x 2GB DDR-2 1066MHz
Win 7 64Bit

OC @ 3.2GHz, DDR-2 @ 1003MHz

C2Q is pretty slow, you made these tests using mantle or not?

3.2GHz c2q should be a good candidate for Mantle testing in this game and BF4.

 

[AtenRa]

D3D with catalyst 14.2, i will try mantle later.

 

[parvadomus]

I have updated post 79 with two more CPUs, The Kaveri A10-7700K and Core 2 Quad Q9450 both at default and OC.

Edit : systems 5 and 6 added (31-3-2014)

System 5
A10-7700K (Default Clocks)
ASROCK FM2A88X-ITX+
2x 4GB DDR-3 2133MHz 11-12-11
Win 8.1 64Bit

OC @ 4.3GHz, NB = 2200MHz

System 6
Core 2 Quad Q9450 (Default Clocks 2,66GHz)
ASUS Rampage Formula
2x 2GB DDR-2 1066MHz
Win 7 64Bit

OC @ 3.2GHz, DDR-2 @ 1003MHz

So kaveri gained nothing over piledriver.. did you check if it throttles?

 

[AtenRa]

So kaveri gained nothing over piledriver.. did you check if it throttles?

I dont think so but i will recheck.

 

[AtenRa]

Ok I managed to use the same Power supply (1000W Be-Quiet 85+) with every system.

But i will have to clear a few specs.
Core i7 920 is using an Intel SSD 320 120GB. Win 7 64bit
Core i7 3770K is also using a OCZ Vertex 4 256GB SSD. Win 8.1 64bit
Core 2 Quad Q9450 is using 2x 76GB Raptors in RAID 0. Win 7 64bit

Rest of the systems are using a single SeaGate 1TB SATA-6 drive with Win 8.1 64bit.

Edit: Also to point out that Core i7 3770K is using a Corsair H80i and FX8350 is using WaterCooling as well.

I have used a Kill A Watt to measure the Maximum Power usage of each system. This is the maximum Power i show in the device, it is not the average or the power usage each system used for the entire benchmark run.

1080p Maximum Settings at Default clocks.

1080p Maximum Settings - Overclocked.

 

[Hitman928]

Thanks for the tests AtenRa, I appreciate the time you put into it for the comparisons. Looking forward to Mantle tests as I still have a C2Q box that would be cool to use with new games going forward, even if it's just mantle games, that would be enough for that box.

 

[inf64]

Thanks for the benchmarks AtenRa, nicely done :thumbsup:

 

[Enigmoid]

Ok I managed to use the same Power supply (1000W Be-Quiet 85+) with every system.

But i will have to clear a few specs.
Core i7 920 is using an Intel SSD 320 120GB. Win 7 64bit
Core i7 3770K is also using a OCZ Vertex 4 256GB SSD. Win 8.1 64bit
Core 2 Quad Q9450 is using 2x 76GB Raptors in RAID 0. Win 7 64bit

Rest of the systems are using a single SeaGate 1TB SATA-6 drive with Win 8.1 64bit.

Edit: Also to point out that Core i7 3770K is using a Corsair H80i and FX8350 is using WaterCooling as well.

I have used a Kill A Watt to measure the Maximum Power usage of each system. This is the maximum Power i show in the device, it is not the average or the power usage each system used for the entire benchmark run.

1080p Maximum Settings at Default clocks.

1080p Maximum Settings - Overclocked.

Nice.

Do you know approximately the idle powers of the systems?

 

[AtenRa]

A little teaser,

A10-7700K default

Video Settings set to High, No SSAA, No FXAA, 16x AF
Cat 14.2 = 39,10 fps
Cat 14.3 Mantle = 58,3 fps

Video Settings set to Max, 16x AF
Cat 14.2 = 37,40 fps
Cat 14.3 Mantle = 46,10 fps

Nice.

Do you know approximately the idle powers of the systems?

Well after the first results with Mantle I have decided to make a full review, so im re-benching every system with Catalyst 14.3 in D3D and Mantle. I will also include power usage in D3D/Mantle and Idle state.

 

[AtenRa]

OK, i got it ready

Thief : Mantle CPU Scaling and Power evaluation

Sorry for the automatic translation, if you require any inquiry pls fell free to ask me here.

For more graphs and power measurements click the link above.

 

[Enigmoid]

Nice job.

The i3 looks pretty amazing.

 

[jpiniero]

The Pentium results are strange, but otherwise it is a huge improvement for everything but the 3770. Look at the Core 2 Quad, that would be a huge improvement in playability.

 

[SPBHM]

The Pentium results are strange, but otherwise it is a huge improvement for everything but the 3770. Look at the Core 2 Quad, that would be a huge improvement in playability.

mantle helped almost every CPU considerably, look at the minimum...

about the Pentium, PClab also got poor gains with it,
mantle and this game are not working well with only 2 threads I think.

http://pclab.pl/art56897-3.html

BF4 Mantle worked a lot better for the Pentium.

 

[frozentundra123456]

The Pentium results are strange, but otherwise it is a huge improvement for everything but the 3770. Look at the Core 2 Quad, that would be a huge improvement in playability.

I have a feeling it is approaching gpu limit with the 3770k.