Ivy Bridge Benchmarks

[MayImilae]

Quote:I get the same speed on a Sandy Bridge 2600K running at 4.5Ghz.

Well, I figured out part of the mystery. My motherboard is turboing automatically to 4181mhz (probably only one core running that). Still, it's running pretty low voltage, and the temperatures aren't even at 30C per core, according to speedfan. I'm impressed. It's going to be awesome to crank it up tomorrow. Anyone know of a better monitor(s) for clockspeed than CPU-Z? I need something that can show the clockspeed of each core.

Now, for the NSMBW turboed scores. World 1-1, right at the start (not moving, but after killing the goomba)

DX9
Internal Resolution: 1x Native
Scaled EFB Copy: Off
EFB to Ram (cached)
Tex Cache - Safe
DSP LLE (on thread)
Result - 64fps
(I had a really weird glitch here. If I kept it at fullscreen, it would race at 118fps, but coins, ? blocks, etc would be stuck. Opening the graphics config panel fixed it, and dropped it down to 64fps)

DX9
Internal Resolution: 1x Native
Scaled EFB Copy: Off
EFB to Texture
Tex Cache - Fast
DSP HLE
Result - 200fps

[ExtremeDude2]

(05-08-2012, 06:13 PM)MaJoR Wrote: (I had a really weird glitch here. If I kept it at fullscreen, it would race at 118fps, but coins, ? blocks, etc would be stuck. Opening the graphics config panel fixed it, and dropped it down to 64fps)

I suspect that RAM didn't get switched at at first, then when you opened it, it re-freshed and turned on

[bret emerald]

(05-08-2012, 06:13 PM)MaJoR Wrote:

Quote:I get the same speed on a Sandy Bridge 2600K running at 4.5Ghz.

Well, I figured out part of the mystery. My motherboard is turboing automatically to 4181mhz (probably only one core running that). Still, it's running pretty low voltage, and the temperatures aren't even at 30C per core, according to speedfan. I'm impressed. It's going to be awesome to crank it up tomorrow. Anyone know of a better monitor(s) for clockspeed than CPU-Z? I need something that can show the clockspeed of each core.

Thanks for providing these very comprehensive test results, but I think that testing like this is better done with turbo boost disabled completely so that you know exactly what the clockspeed is (then you won't need to use cpu-z to monitor). With turbo boost clock frequency can often alternate between different values, and some motherboards may use different default turbo power limits (as seems to be the case here). I know you don't want to overclock yet which is quite understandable but you can probably push the multiplier up to 39 or 40 without any voltage adjustments at all, or just testing at the default 34 would be perfect as well.

I'm sure you don't want to repeat all the tests but you can probably leave out the d3d11 mario galaxy ones as they can often vary depending on video card/drivers etc. Perhaps for NSMBW just HLE with efb to ram(no cache) would be adequate.

Hope I don't sound too ungrateful, I know the testing must have taken some time.

[IcemanSR]

i agree,
all due respect to your time in bench's
but using turbo ON is just wrong when getting precise results
my 2500k gets turbo boosted to 4.2ghz on all cores

[MayImilae]

Yea yea yea, I should have watched the clocks when doing the benchmarks. But I was up front that turbo was on. I wasn't aware that my motherboard would crank it up so far automatically. Though I admit it still impressed me. Running at under 4.2ghz it was matching a 4.5ghz 2600K? NICE. And it didn't even hit 40C in the stress test. The safe ceiling is 90C, so I have TONS of breathing room.

So, about that turbo. I finally had a chance to study the board some, and I'm figuring it out. And it's uh, different. Asus went to this AI Tweak system during Sandy Bridge, and it definitely has a learning curve. It does everything through turbo. You have to have turbo enabled to get anywhere with this board, but it is able to push well beyond the CPU's normal turbo limitations. It's weird, but it works, and it has all the options of a traditional base+multiplier overclock so, whatever. *shrug* I'll do more testing tomorrow, now that I know how it operates.

[MayImilae]

*I hate making a double post but it's been a while and this is new information*

Alright guys. So far I've raised it up to 4.8ghz. With voltage on auto, which is just, INSANE, considering this is ivy bridge. 4.8ghz with 1.5v, ON AIR, and it's within safe ranges and perfectly stable. Just, WOW. This chip is AWESOME. If I get voltages under control I could possibly hit 5ghz. This is better than I ever hoped, and better than most 3570Ks I've seen! So, I have alot more things that I need to learn to fully understand this chip and this board, but in the meantime, here's some benchmarks.

Core i5 3570K @ 4.5ghz

Cinebench 11.5
OpenGL - 62.20fps
CPU - 7.39pts
CPU (single core) - 1.89pts

Mario Galaxy (observatory start point, no movement)
DX9
Internal Resolution: 1x Native
Scaled EFB Copy: Off
EFB to Ram
Tex Cache - Safe
DSP LLE (on thread)
Result - 63fps


Core i5 3570K @ 4.8ghz

Cinebench 11.5
OpenGL - 68.62fps
CPU - 7.89pts
CPU (single core) - 2.03pts

Mario Galaxy (observatory start point, no movement)
DX9
Internal Resolution: 1x Native
Scaled EFB Copy: Off
EFB to Ram
Tex Cache - Safe
DSP LLE (on thread)
Result - 67fps

Mario Galaxy (observatory start point, no movement)
DX9
Internal Resolution: 3x Native
Scaled EFB Copy: On
EFB to Ram
Tex Cache - Safe
DSP LLE (on thread)
Result - 60-61fps

[skid]

I would call 67fps, "fast" no matter which way I look at it. Looking forward to the NSMBW bench.

[Squall Leonhart]

Ivy bridge has too many performance REGRESSIONS for my taste.

Slower in vba-m than a 2600k (by some amount) and numerous other games show it performs equal to, or less than a 920D0

Pokemon Fire Red - Intro (Charizard/Flames)

skid_au @ 3.8ghz 2600k
917% - 1063% unsync'd audio
1082 - 1335 with turbo pressed

Squall 3.8Ghz 920D0
480-534% unsync'd audio
917-1070% with turbo pressed

Gigaherz 3.7Ghz 3770k
700-800% unsync'd audio
1080-1200% with turbo pressed

(05-06-2012, 11:49 PM)dannzen Wrote: jeah... changing the mobo without reinstall is tricky...
regedit tweaks and driver deinstall...

i think i did it one time for aa customer

No it isn't

Vista and 7 are just fine with being dropped into a new mainboard/cpu as long as the MSAHCI driver is enabled.

having raid enabled before the switch might pose some problems.....

[Duke Nukem]

(05-13-2012, 09:19 AM)Squall Leonhart Wrote: Ivy bridge has too many performance REGRESSIONS for my taste.

Slower in vba-m than a 2600k (by some amount) and numerous other games show it performs equal to, or less than a 920D0

Pokemon Fire Red - Intro (Charizard/Flames)

skid_au @ 3.8ghz 2600k
917% - 1063% unsync'd audio
1082 - 1335 with turbo pressed

Squall 3.8Ghz 920D0
480-534% unsync'd audio
917-1070% with turbo pressed

Gigaherz 3.7Ghz 3770k
700-800% unsync'd audio
1080-1200% with turbo pressed

Hmm for me it's the opposite, I think it's faster and not slower.

Dolphin, bsnes, DeSmuME, SSF, PCSX2 is all running much smoother now with the 3770K.
Heck I even think XEBRA, ScummVM, Project64 and Fusion have all gained from it as well.

I don't know though it could all be in my head, but so far from what I have tested I haven't been exactly disappointed.

It's a joy booting up the PC these days with literally nothing that stops you from enjoying the games.
And I don't miss the 2600K one bit.

[bret emerald]

Based on the benchmarks here and comparing with my own, I think Ivy Bridge performance in Dolphin can be summarised as follows.

When using efb copies to texture or to ram (cache enabled), performance is identical between Ivy Bridge and Sandy Bridge.

When efb copies to ram(no cache) is used Ivy Bridge is significantly faster (~10%) when using the same ram speeds. Are there many games that require efb to ram(no cache) to be used? Off the top of my head I can only think of NSMBW (so that coins spin).