AMD Bulldozer Finally!

[Martini1991]

What is interesting though, is whether it goes ;

1 thread = 1
2 threads = 1.8
3 threads = 2.8
4 threads = 3.6
5 threads = 4.6
6 threads = 5.4
7 threads = 6.4
8 threads = 7.2

Or

1 thread = 1
2 threads = 2
3 threads = 3
4 threads = 4
5 threads = 4.8
6 threads = 5.6
7 threads = 6.4
8 threads = 7.2

With 1 being base BD "core" performance.

Also, if a BD core was say, 10% faster than a Phenom II core, so a Phenom II core is 0.9, an 8 core Phenom II with a base performance of 0.9 would equate 7.2.

If I'd said this months ago, you'd all have gone mental

 

[muon]

Just so we're clear the 80% used is just a made up number right? If not, source??

 

[Martini1991]

Just so we're clear the 80% used is just a made up number right? If not, source??

AMD's own slides, the 80% isn't a made up number.

I was exhausted at the time, funnily enough what I was writing about 0.8 will likely be correct.

 

[chaotic_uk]

at the end of the day will it be worth people with the 1090T/1100T upgrading to these cpu's ?

 

[gareth170]

at the end of the day will it be worth people with the 1090T/1100T upgrading to these cpu's ?

if u use fully multithreading apps then yes.

im getting one either way.

 

[Sin_Chase]

if u use fully multithreading apps then yes.

im getting one either way.

Even if an i7 kerb stomps the best model in multithreaded apps?

(Hypothetical question)

 

[Martini1991]

Even if an i7 kerb stomps the best model in multithreaded apps?

(Hypothetical question)

To be fair, Gareth has a reason for AMD loyalty.

if u use fully multithreading apps then yes.

im getting one either way.

How many are actually out?
The whole encoding thing, I don't buy either simply because of Intels Z68 stuff which allows encoding off the IGP, although this isn't in all applications, but it'd leave an 8150 for dust.

 

[gareth170]

To be fair, Gareth has a reason for AMD loyalty.

.

, haha

 

[Martini1991]

Even if you didn't have AMD loyalty, you'd still get the 8150.

I love how the amount of flaming I received for my opinion has diminished

 

[gareth170]

How many are actually out?
The whole encoding thing, I don't buy either simply because of Intels Z68 stuff which allows encoding off the IGP, although this isn't in all applications, but it'd leave an 8150 for dust.

not many tbh.

i've found using my gpu for encoding gives slower result

 

[gareth170]

Even if you didn't have AMD loyalty, you'd still get the 8150.

I love how the amount of flaming I received for my opinion has diminished

can't really answer that but i would have different hardware

 

[Owenb]

What is interesting though, is whether it goes ;

1 thread = 1
2 threads = 1.8
3 threads = 2.8
4 threads = 3.6
5 threads = 4.6
6 threads = 5.4
7 threads = 6.4
8 threads = 7.2

Or

1 thread = 1
2 threads = 2
3 threads = 3
4 threads = 4
5 threads = 4.8
6 threads = 5.6
7 threads = 6.4
8 threads = 7.2

With 1 being base BD "core" performance.

Also, if a BD core was say, 10% faster than a Phenom II core, so a Phenom II core is 0.9, an 8 core Phenom II with a base performance of 0.9 would equate 7.2.

If I'd said this months ago, you'd all have gone mental

I have been wondering that myself. Hopefully the answer will be both. If out and out performance is important, e.g games, then it will be the 2nd option, but if it's performance per watt then the first.

People need to understand that there is a lot of change going on at AMD. This is the first iteration of this architecture, they no longer have a fab in-house and it's the first time they have done a HKMG chip(I am ignoring Llano as the time difference is so small that I doubt they learnt anything that could be transferred to BD). I don't see this as a stellar chip but from what we have seen, which might be all BS, it looks reasonable competitive and is certainly a step in the right direction

 

[pmc25]

As I said before, there is absolutely no chance of scaling being linear or consistent from one application to another.

The argument is entirely pointless at the moment.

 

[muon]

As I said before, there is absolutely no chance of scaling being linear or consistent from one application to another.

The argument is entirely pointless at the moment.

Well it does matter. If theoretical scaling is less than 1x then, real world scaling will be impacted.

 

[Martini1991]

As I said before, there is absolutely no chance of scaling being linear or consistent from one application to another.

The argument is entirely pointless at the moment.

That's like saying everything is pointless.
We all know that those figures are only rough estimates, but they're based on what's there, and more than likely, will be the case.

 

[Kraven]

AMD need to release these soon, I want to order and build before BF3!

 

[Martini1991]

Well it does matter. If theoretical scaling is less than 1x then, real world scaling will be impacted.

And we're seeing real world benchmarks too, they're the leaked ones.

 

[Mr Krugga]

why do that when you've got users on forums including me that are willing to do that? also users results cover awide range of rigs

Why choose one or another?

If you're smart enough, you'd be making an informed decision based on results coming from as many unbiased sources as possible.

You already said that you'd be buying the Bulldozer CPU regardless of its performance in which case I'm inclined to say you're more biased than 99% of proper reviewers out there.

And it's ******** that an average user can make a fair comparison between specific hardware. Most will only have one or two rigs at hand and won't be able to put them side by side in the same environment.

 

[gurusan]

Fanboys will be fanboys. They think it's some sort of moral struggle yet really they are just consumers like the rest of us....

 

[drunkenmaster]

Well it does matter. If theoretical scaling is less than 1x then, real world scaling will be impacted.

It doesn't matter, because EVERY chip has real world scaling below 1x, on top of this I can't remember the quote exactly that Martini is talking about but I'm fairly sure AMD have NOT said the second thread on a module will only offer 80% of the performance, which only makes it more ridiculous.

IIRC this quote from AMD is from AGES ago

For the record, AMD says that a single Bulldozer module has around 80 per cent of the performance of two conventional CPU cores

This in no way indicates what Martini is saying AT ALL. AMD is saying they COULD have made a bigger conventional core that would have only been 20% faster on each core, but would have been big enough that they couldn't come close to fitting 8 cores in, or they make 8 smaller cores, its as simple as that.

I haven't seen a single suggestion that the second core only provides 80% scaling, thats not even slightly hinted at from that statement.

Its a "what if we went the conventional route" comparison, and 4 cores with 100% performance, or 8 cores with 80% of the performance each = 60% performance increase IN THE SAME DIE SPACE, with this method.

Its a throw away comment anyway, because they didn't actually design a 10-20% larger core and find out the actual IPC they could get out of it. Theory and practice often aren't close and going after the diminishing returns of single thread performance its unlikely to be that big a loss anyway.

again this theoretical core wasn't a Phenom 2 core as they've already said that Bulldozer IPC has increased.


Anyway, the point being that, from a Phenom 1/Q6600, up to 990x, interlagos and Bulldozer, no chip scales 100% from one thread to the next. Some workloads come pretty close, some come no where near.

Take almost any chip with one thread instead of 4/6/8 and it will have more L3 cache dedicated to itself, more memory bandwidth, all HDD requests all working on one thread. All chips lose performance when you throw in more threads, that has never been different, why is such a big deal being made of it even if the quote did say scaling was lost with more threads, that is entirely normal.