AMD - Bulldozer Chitchat

[watercooled]

Super Pi

So, we have early access to this brand new, highly anticipated CPU, so lets run a 20 year old, completely obsolete benchmark which isn't even remotely relevant to any programs which you're likely to use or which will require much CPU power...

Maybe I'm the only one annoyed by the continued use of such a pointless benchmark?

Nothing against you BTW CAT, just an observation of the first look benchmarks for CPUs using it so frequently.

[CAT-THE-FIFTH]

Like I said Bulldozer and Piledriver lack an X87 unit,so the fact that Trinity is outperforming Llano is a good sign. In more modern applications single thread performance should be better than 19% for Trinity. Considering that the TDP is the same as Llano and the process is the same,at least a 20% improvement in single thread performance and around 30% to 40% improvement in GPU performance looks decent.

Edit!!

I had a look at some numbers for the A8-3870K:

http://www.anandtech.com/show/5626/i...re-i7-3770k/11

It seems with newer drivers the HD6550D is now very close to an HD5570 or a GT440. Even a 25% improvement in GPU performance would make the A10-5800K IGP faster than any of those cards.

[Terbinator]

Snip...

Card names aside, Dirt 3 @ full settings/ '768'p is the top-tier console graphics on one piece of silicon, competing cars aside.

If i was to build a gaming PC as cheap as possible, one that i'd be happy to call a gaming PC, Trinity would be the spine of it.

[CAT-THE-FIFTH]

I made a typo in my calculations - the A10 is around 29% faster in the SM3.0 score,not 36% as I said before.

[CAT-THE-FIFTH]

Off topic.

Can one of you look at this:

http://semiaccurate.com/forums/showt...=5021&page=132

My maths is correct right?? Or am I loosing my mind??

[Random_guy]

The confusion is coming because of the units, you were talking about speed but measuring it in seconds, when it should be work done per second. This means you have to effectively use the opposite pivot point the one you'd normally use if you just want to work with the time.

So in the case of the benchmark it's the same amount of work done so just call it 100 units of work.

Call the first time (for the A8) t1, and the second (for the A10) t2. Then the A8's speed would be:

(100 / t1)

And the A10's:

(100 / t2)

Then we can do a say the A10's speed is (((100 / t2) / (100 / t1)) * 100)% of the A8's speed, which cancels to ((t1 / t2) * 100)%.

So for your numbers it becomes (29.219 / 23.775) * 100 = 123%, so A10's speed is 123% of A8's speed i.e. the A10 was 23% faster than the A8.

This will seem odd to a lot of people because normally you would divide by the value you are setting to 100%, in this case the A8, but as I said, it's reversed because speed is not measured in seconds.

*disclaimer, it's currently 0045 and I've had a couple of glasses of wine, so the above is probably either incorrect, overly patronising in tone, or both

[CAT-THE-FIFTH]

Move to other thread.

[Golden Dragoon]

The confusion is coming because of the units, you were talking about speed but measuring it in seconds, when it should be work done per second. This means you have to effectively use the opposite pivot point the one you'd normally use if you just want to work with the time.

So in the case of the benchmark it's the same amount of work done so just call it 100 units of work.

Call the first time (for the A8) t1, and the second (for the A10) t2. Then the A8's speed would be:

(100 / t1)

And the A10's:

(100 / t2)

Then we can do a say the A10's speed is (((100 / t2) / (100 / t1)) * 100)% of the A8's speed, which cancels to ((t1 / t2) * 100)%.

So for your numbers it becomes (29.219 / 23.775) * 100 = 123%, so A10's speed is 123% of A8's speed i.e. the A10 was 23% faster than the A8.

This will seem odd to a lot of people because normally you would divide by the value you are setting to 100%, in this case the A8, but as I said, it's reversed because speed is not measured in seconds.

*disclaimer, it's currently 0045 and I've had a couple of glasses of wine, so the above is probably either incorrect, overly patronising in tone, or both

Your math seems correct there, and you aren't being patronising, many people struggle when working out percentages of this nature, how many review sites have you seen that say x is faster than y by z% at the bottom of these charts?

[DanceswithUnix]

So, we have early access to this brand new, highly anticipated CPU, so lets run a 20 year old, completely obsolete benchmark which isn't even remotely relevant to any programs which you're likely to use or which will require much CPU power...

Maybe I'm the only one annoyed by the continued use of such a pointless benchmark?

Nothing against you BTW CAT, just an observation of the first look benchmarks for CPUs using it so frequently.

So are you implying that 3dmark06 is a valid and useful benchmark?

[watercooled]

Not at all, in fact not many are, especially between platforms as the 3dmark programs are, as I've said, compiled with an Intel compiler. But at least it uses more than one thread and some recent instructions rather than plain x86 INT and some x87 which is actually emulated on many modern CPUs as its rarely used because of far better alternatives with new instruction sets.

[kalniel]

But at least it uses more than one thread and some recent instructions rather than plain x86 INT and some x87 which is actually emulated on many modern CPUs as its rarely used because of far better alternatives with new instruction sets.

But I thought x87 was cutting edge? Why else would it be used for things like PhysX?

[watercooled]

Because it's obviously the most efficient possible way of doing things and can't possibly be purely to get you to buy an Nvidia GPU.

[scaryjim]

... can't possibly be purely to get you to buy an Nvidia GPU.

The irony of "NVidia GPU" being turned into a skimlink in the above sentence is truely mind blowing...

[Golden Dragoon]

Didn't they try to argue that 64bit precision wasn't enough, and 128bit+ was too much, but the 80bit precision of x87 was just right?

[watercooled]

They've spouted plenty of laughable excuses over the years, a CPU is perfectly capable of running optimised PhysX code, they just make it deliberately inefficient both so you need their GPU, and I suppose it gives the impression the GPUs are incredibly fast or something. I could be wrong but I don't think it's even multithreaded for CPU...

[CAT-THE-FIFTH]

Trinity A10-4600M clockspeeds:

http://www.sibvez.ru/shop/UID_13494.html

It seems with laptops already advertising Trinity CPUs already being released,Trinity should soon be out.

The A10-4600M has a 35W TDP AFAIK.