Bulldozer performance figures are in

[Justintime]

Looking decent.. high hopes for AMD!

 

[Martini1991]

So it's a 1/3 faster give or take than a 2600K in the multithreaded stuff, now the 2600K runs 8 threads with HT which very roughly equates to a 50% boost over 4 threads only. So it's roughly the same speed thread per (non HT) thread on the 2600K. If the BD is running 4.2GHz and the SB presumably 3.8GHz it's not that far away clock for clock for single threaded performance.

No where near 50% increase, how the hell you came up with that, I'm bemused.Compare the 2500k and the 2600k at the same clocks.
Hyperthreading each generation will become less and less due to efficeny being better in core performance.

I don't know how you've managed to gauge it to being close to SB clock for clock, the Fritz results have it on par with Deneb.

 

[CAT-THE-FIFTH]

I don't know how you've managed to gauge it to being close to SB clock for clock, the Fritz results have it on par with Deneb.

Which means nothing in lightly threaded applications due to the resource sharing of each module. None of these results gives any indication of lightly threaded performance.

The first pass video encoding results are 36% higher than the Core i7 2600k indicating that lightly threaded performance is higher.

 

[Martini1991]

Which means nothing in lightly threaded applications due to the resource sharing of each module. None of these results give any indication of lightly threaded performance.

The first pass video encoding results are 36% higher than the Core i7 2600k indicating that lightly threaded performance is higher.

Which is promising.

BD looks like it's either going to be good in an app, or terrible.

 

[Final8y]

Which is promising.

BD looks like it's either going to be good in an app, or terrible.

What apps will it be terrible at ?

 

[Martini1991]

What apps will it be terrible at ?

Well, Fritz ?
If what people say about when all cores being loaded, it taking a performance hit, rendering etc would be affected?

That's assuming that benchmark is legit.

 

[DragonQ]

So it's a 1/3 faster give or take than a 2600K in the multithreaded stuff, now the 2600K runs 8 threads with HT which very roughly equates to a 50% boost over 4 threads only. So it's roughly the same speed thread per (non HT) thread on the 2600K. If the BD is running 4.2GHz and the SB presumably 3.8GHz it's not that far away clock for clock for single threaded performance.

Nope, typically 25%. x264 takes good advantage of HyperThreading but still only manages a 25-30% performance increase. x264 presumably hasn't been optimised for Bulldozer yet but it probably can be.

So in this comparison, an i7-2600K without HyperThreading would get about 77 fps, making 4 Bulldozer modules (8 Zambezi cores) 77% faster than 4 Sandy Bridge cores. Much better than the 30% boost provided by HyperThreading and in the same die space (IIRC).

However, as I said before, there's no clock information here so these numbers are worthless.


Personally I will be going Zambezi if it is able to run my specific code (just about) as fast as Sandy Bridge. I suspect it will be close but we'll see. AVX implementation might be important, plus I'd be using an Intel compiler so that might affect things.

 

[Final8y]

Well, Fritz ?
If what people say about when all cores being loaded, it taking a performance hit, rendering etc would be affected?

So because of Fritz the same will go for every program that loads all the cores & that has been confirmed with the simple task of running other apps that loads all the cores ?

 

[Martini1991]

So because of Fritz the same will go for every program that loads all the cores & that has been confirmed with the simple task of running other apps that loads all the cores ?

I don't know?
If the logic applied, is when all cores are loaded, it would do so regardless of app?

 

[Final8y]

I don't know?
If the logic applied, is when all cores are loaded, it would do so regardless of app?

Only if it was the fact its because of all cores being loaded & not an issue with BD & Fritz & that can not be known without running other full loading apps & that's what they should have done as soon as they noticed strange behaviour.

 

[drunkenmaster]

2 BD cores share fetch&decode hardware so depending on the app they will take a hit if both cores on a single module are active.

They officially share some hardware per core, that doesn't mean it will take a hit. If separately it could fetch/decode two instructions per core, and then they put two cores together and save space but making it share the ability to decode 3 instructions per clock, then it would impact performance, if it still has 2 instruction decode/fetch per clock then it wouldn't lose performance.

Just because things are pushed together on the die doesn't mean they are shared and performance is lost.

A lot of die space can be saved simply by having one communication path rather than two, and ultimately 99% of the design of Bulldozer is to add a second core to a module for as little die space impact as possible.

9 fps more on the second pass, isnt that much really being as it has twice as many cores, would have to be cheap to compete.

I wish people would stop saying that, firstly if its faster than a 2600K in most situations and the same price, who cares. This is NOT a $1000 core, or a $500 one, its said to be around $320 putting it squarely in 2600k pricing region.

Number of cores has become irrelevant, this isn't 15 years ago with one or two issues per core, its the Nvidia vs AMD era, core architecture is so different now that core vs core comparision is NOT valid.

A 2600k can run 4 issues per core, on 4 cores, or on a perfect clock it can essentially execute 16 instructions. Bulldozer is 8 cores, 2 issue cores, and on a perfect clock it can execute..... 16 instructions. Bulldozer cores are significantly smaller than Phenom cores, and AFAIK look set to be smaller than Sandy cores. AMD have been competitive price/performance for, well, ever. The big problem right now is that Intel are generally a process ahead and can make cores at a lower cost than AMD but charge more, Phenom cores are pretty damn big AND on a higher process meaning to be competitive on price performance, profit per core vs Intel is a joke. Hence a small die strategy, the same one that has helped AMD.

Make no mistake, a Bulldozer with 4 x 4 issue cores to compete with Intel wouldn't be much smaller, if ANY smaller at all. Its a design choice, nothing more or less. 2 issues isn't better than 4, and 4 isn't better than 2, they are just different options. People expecting twice the performance of a 2600k because its got twice as many cores, are the same people who would expect a car with twice the maximum speed out of a v8 engine over a v4, its not how life works.


The biggest thing for AMD is SIZE not performance, if they get performance aswell, that is exceptional engineering, Bulldozer was, and always has been first and foremost about making AMD's architecture more efficient per mm2.

As ATI and now AMD can tell you, performance per mm2 directly translates to profit margin on a product, performance per mm2 is what makes Intel billions a year, not outright performance. Look at Nvidia to see what ultimate performance on a massive design gets you, not much, look at Intel chip sizes vs AMD and realise their chips cost half as much to make, yet sell for more.

 

[Martini1991]

@ Drunken.
All well and good having 8 cores, but nothing will use it, hence why the 1100T sucks against the 2500k. This will be the case with BD against SB for the majority.
Those who need the cores are in the minority.
AMD is outpaced price/performance wise right now.. Everything they have gets its arse handed to it by a cheaper SB CPU.
Number of cores is not irrelevant, you'd have to be an idiot to think that.... When software can use CPU cores like games use GPU cores, THEN it'll become irrelevant.

EDIT : Since when was bulldozer about effiency? It was ALWAYS about performance....

 

[CAT-THE-FIFTH]

@ Drunken.
All well and good having 8 cores, but nothing will use it, hence why the 1100T sucks against the 2500k.

In video encoding with software like HandBrake this is not the case and especially so at stock speeds since most Intel and AMD CPUs sold are not overclocked.

 

[Martini1991]

In video encoding with software like HandBrake this is not the case and especially so at stock speeds since most Intel and AMD CPUs sold are not overclocked.

One case. It does after all, have 6 cores

 

[clocka]

186W TDP? Blimey.

they are using an old version of CPU-Z. v1.58 adds support for bulldozer and preliminary support for sandybridge-E.

 

[CAT-THE-FIFTH]

The Phenom II X6 does well in rendering too and if you intend to run multiple virtual machines.

Even, if a 4 module Bulldozer has the performance of eight Phenom II cores in highly multi-threaded applications it will be very fast.

I suspect Bulldozer will have decent performance in lightly threaded applications. Even the Phenom II had a decent IPC improvement over the Phenom and Athlon 64.

 

[Martini1991]

The Phenom II X6 does well in rendering too and if you intend to run multiple virtual machines.

Even, if a 4 module Bulldozer has the performance of eight Phenom II cores in highly multi-threaded applications it will be very fast.

I suspect Bulldozer will have decent performance in lightly threaded applications. Even the Phenom II had a decent IPC improvement over the Phenom and Athlon 64.

As it should, it's 6 cores.
If AMD get a module to execute a thread, then they're onto a winner.

 

[CAT-THE-FIFTH]

As it should, it's 6 cores.
If AMD get a module to execute a thread, then they're onto a winner.

Which makes the Core i5 a relatively poor CPU for those sorts of things.

 

[Martini1991]

Which makes the Core i5 a poorrelatively CPU for those sorts of things

Barely though.
In other cases the 1100T gets owned by a 2120.
http://www.anandtech.com/bench/Product/288?vs=203 Look at the rendering stuff, the 2500k is beating the 1100T.

 

[CAT-THE-FIFTH]

Barely though.
In other cases the 1100T gets owned by a 2120.

It is still faster though. It makes the Core i5 worse for those applications. If you need to run those kinds of applications it is a better choice.

The Core i3 is only fine for gaming IMHO. The Core i3 2120 can produce simliar framerates(or near enough) in many games to a stock Core i5. This means it also makes the Core i5 look overpriced too in many cases since most people don't overclock.

For the purposes I mentioned in the previous posts the Core i3 is usually slower. I have one myself and the AMD Athlon II X4 and Phenom II X4 are faster for video encoding and rendering.Things like image editing are far more bottlencked by lack of RAM and disk speed.

http://www.anandtech.com/bench/Product/288?vs=203 Look at the rendering stuff, the 2500k is beating the 1100T.

It seems that many reviews consider the Phenom II X6 faster for rendering:

http://techgage.com/article/intels_sandy_bridge_revealed_core_i5-2500k_i7-2600k_reviewed/6

http://www.tomshardware.com/reviews/sandy-bridge-core-i7-2600k-core-i5-2500k,2833-15.html

http://www.hardware.fr/articles/815-16/3d-studio-max-cinema-4d.html

http://techreport.com/articles.x/20188/14