AMD Ontario APU pictured,die size ~77mm^2

[Calmatory]

But if your cpu has only two decoders it doesn't mean that it has an equal IPC to cpu with 4 decoders when executes code with ILP <= 2. A simple example (code with sequence of 4 arithmetic operations):
a = b + c
a = a + d
e = g + h
e = e + f
Cpu with 4 decoders can execute first and third instructions in the same cycle, while cpu with 2 decoders will need one more cycle for that. Of cause in reality things are a bit more complex because of OutOfOrder buffer but again, i really doubt bobcat has bigger OOO instruction window then Conroe/Athlon64.

Two words: vectorization and SIMD.

[kl0012]

Read this

Firstly, it does not contradict what i said. Second, i don't see a reason to measure IPC in KDE, GNOME e.t.c. These are window managers spending most of the time in the halt state.

[Mechanical Man]

Firstly, it does not contradict what i said. Second, i don't see a reason to measure IPC in KDE, GNOME e.t.c. These are window managers spending most of the time in the halt state.

No, it was not meant to contradict all, it was just meant to give more wide view on matter. It clearly shows that it depends on type of program used. Remember bobcat is meant for netbooks/low power notebooks, think what kind of programs that target uses. Yes it is clear that there is cases where decode stage can bottleneck and where l2 can bottleneck. But for intended target market those kind of tasks should be minimum.

[informal]

Hans slightly updated the comparison image of Bobcat and Atom :

A picture is worth a thousand words

[nn_step]

Firstly, it does not contradict what i said. Second, i don't see a reason to measure IPC in KDE, GNOME e.t.c. These are window managers spending most of the time in the halt state.

I was using the largest open source code bases.

Now, if you would like; I'd be more than happy to test any open source software or project you deem more fit.

[Chumbucket843]

BTW, Bonic benchmark keeps at min memory access.

while we are on the subject, BOINC benchmark is whetstone and drhystone which are useless. for one linux scores much higher than windows on the same system. there is controversy over using this bench as the basis of points. some projects award points as WU time * boinc bench score, others use a really old computer such as a PIII machine as a base and then multiply it by a speed up factor. this whole system gets screwed up when you run non-deterministic algorithms because the wu may run until some criterion is met.

[kl0012]

Two words: vectorization and SIMD.

Vectorization is good, but it is not a panacea. Replace third operation in my example with "mul", "and", "shift", "test" or "sub" and SIMD wont help (while these ops are still independent). But my point was simple - as far as some cpu has a bigger pool of uops available for execution, so the cpu's OoO logic has a better chance to explore ILP. This is way i'm surprised by bobcat results (if these are real). I would guess that they have used a loop buffer, but such a buffer would consume a lot of space on the cpu die.

[rcofell]

Just finally realized I can see bumps on the die shot, meaning it's likely a shot at the top level interconnect. The regular structure over the GPU portion must just be power/gnd rails.

Two words: vectorization and SIMD.

However his example is emphasizing how the actual instructions are ordered (for dynamic OoO execution, see VLIW for statically done by compiler), he could've noted those were SIMD operations there instead

while we are on the subject, BOINC benchmark is whetstone and drhystone which are useless. for one linux scores much higher than windows on the same system. there is controversy over using this bench as the basis of points. some projects award points as WU time * boinc bench score, others use a really old computer such as a PIII machine as a base and then multiply it by a speed up factor. this whole system gets screwed up when you run non-deterministic algorithms because the wu may run until some criterion is met.

Yeah, the BOINC benchmark is just a purely synthetic benchmark that has little correlation to actual real world performance.

While this is sort of off-topic, the whole notion of having a universal credit scheme based off of an arbitrarily chosen synthetic benchmark makes consistency [between machines] very hard to find; considering each processor has different real world performance characteristics, along with the fact many projects run completely different algorithms... And this is before even bringing SSE into the mess (benchmark doesn't use it, and doing so only further distorts the results), along with the other issues up that alley...

It would be nice if each project made a benchmark representing their algorithm instead and that all got normalized for cross-project comparison

[kl0012]

I was using the largest open source code bases.

Now, if you would like; I'd be more than happy to test any open source software or project you deem more fit.

Well if you have a free time to play with it...
http://sourceforge.net/ - chose whatever you want.

[duploxxx]

damn bobcat has a very small core size and dense cache

80SP so that is ati 5470 performance...

time for a decent netbook that can do it all if the graphics can downclock enough to safe power when needed.

[-Boris-]

Is it just me or is anyone else interested in overclocking that little toy?

[STEvil]

extremely interested...

[-Boris-]

extremely interested...

A nice OC-board and LN2.

[madcho]

A nice OC-board and LN2.

You think you will able to beat nehalem to super pi ?

[-Boris-]

You think you will able to beat nehalem to super pi ?

Super Pi isn't that heavy.

[madcho]

Super Pi isn't that heavy.

when you want push hard some chips with ln², you do super pi, 3Dmarks to break records.

So i guess, he think this chip will be able to break nehalem best time with ln².

I think this is a incredible ship to do fast cheap NAS server.

[-Boris-]

when you want push hard some chips with ln², you do super pi, 3Dmarks to break records.

So i guess, he think this chip will be able to break nehalem best time with ln².

I think this is a incredible ship to do fast cheap NAS server.

I'm hoping that the higher power version will be able to run 1080P in software.
I don't trust hardware acceleration that much. Hopefully I can throw away my current HTPC, witch crappy nVidia chipset. Miss my old AMD 690G.

[Mechanical Man]

I'm hoping that the higher power version will be able to run 1080P in software.
I don't trust hardware acceleration that much. Hopefully I can throw away my current HTPC, witch crappy nVidia chipset. Miss my old AMD 690G.

My Ol&trusty 3870 handles hd decoding with no prob. So I cant see why would Bobcats decode be any worse (it should be better).

[-Boris-]

My Ol&trusty 3870 handles hd decoding with no prob. So I cant see why would Bobcats decode be any worse (it should be better).

Because you are only using the processor when you do it in software.

[madcho]

you can do it in sofware too i think boris.

[-Boris-]

you can do it in sofware too i think boris.

I'm not to sure. I've seen people have problems with C2D at 2.5GHz+.

[madcho]

I'm not to sure. I've seen people have problems with C2D at 2.5GHz+.

C2D 2.5ghz ... have they this inside their computer ? :

[Mechanical Man]

Because you are only using the processor when you do it in software.

My point was for your untrust of hw decode, not about sw decode speed. I dont think bocat can be fast enough for 1080p sw decode. Only with big maybe its fastest and most consuming version, but i would not put my bets on that.

[-Boris-]

C2D 2.5ghz ... have they this inside their computer ? :

No, on my old X2 @ 2.2GHz only WMV worked smoothly at 1080P, some formats was way to slow.

My point was for your untrust of hw decode, not about sw decode speed. I dont think bocat can be fast enough for 1080p sw decode. Only with big maybe its fastest and most consuming version, but i would not put my bets on that.

My distrust for hardware decoding has nothing to do with speed, it has to do with support. I want something that can play all formats in all containers. As far as I know ATi still don't have support for pixelmapped 1080P in Vista or Win 7.

[Mechanical Man]

No, on my old X2 @ 2.2GHz only WMV worked smoothly at 1080P, some formats was way to slow.



My distrust for hardware decoding has nothing to do with speed, it has to do with support. I want something that can play all formats in all containers. As far as I know ATi still don't have support for pixelmapped 1080P in Vista or Win 7.

Is pixelmapping not property of tv not gfx card? Atleast that is my understanding (if we speak about same thing, in otherwords overscan).