Rumour: Bulldozer 50% Faster than Core i7 and Phenom II.

[beginner99]

No, there are 2 FMACs and on any cycle you could have a 128-bit FP instruction scheduled on each of the FMACs or you could merge the FMACs together to execute a single 256-bit AVX execution.

So what is missing to make a module 2 full cores?

I mean you say Integer behaves like 2 cores and FP also behaves like 2 cores. Obviously there must be some kind of issue/bottleneck not to call a module 2 full cores.

You only get 1 AVX per clock but that seems about it.
But can 2 threads each schedule a 128-bit FP or can 1 thread schedule 2 per clock?

 

[JFAMD]

There is a ton of duplicated circuitry in the front end that is identical per core and has low utilization.

2 threads can schedule 2 FMACs per cycle.

 

[nemesismk2]

if the Rumour: Bulldozer 50% Faster than Core i7 and Phenom II is true then it's exciting but I am waiting for some hard facts and benchmarks to decide if it's true

 

[tokie]

I think it is fully possible that AMD will sell a ~$200 Bulldozer that will be faster at stock than a ~$200 Sandy Bridge.

Intel has always been more expensive for comparable performance. The ace Intel has up its sleeve however is tons of frequency headroom. I don't think AMD will have that right away with a new architecture on a new process.

 

[veri745]

So what is missing to make a module 2 full cores?

I mean you say Integer behaves like 2 cores and FP also behaves like 2 cores. Obviously there must be some kind of issue/bottleneck not to call a module 2 full cores.

You only get 1 AVX per clock but that seems about it.
But can 2 threads each schedule a 128-bit FP or can 1 thread schedule 2 per clock?

You can see the BD block diagram in Anand's article.

The Fetch and Decode blocks are shared between cores, as well as the L2 cache. The FP block shares the scheduler, and if the FMACs are working on two different threads, then they can only use 128 bits each, instead of combining for 256-bit instructions (which means worst case performance is pretty much what exists in Deneb).

 

[PlasmaBomb]

Marked... going to read through this before commenting...

 

[JFAMD]

You can see the BD block diagram in Anand's article.

The Fetch and Decode blocks are shared between cores, as well as the L2 cache. The FP block shares the scheduler, and if the FMACs are working on two different threads, then they can only use 128 bits each, instead of combining for 256-bit instructions (which means worst case performance is pretty much what exists in Deneb).

There is a lot of data that came out after that article. I would be careful about how much you view it as the gospel vs. his impressions.

This was before hot chips, before analysts day and before any of the bulldozer blogs.

 

[hamunaptra]

Yeah read this article instead:
http://www.realworldtech.com/page.cfm?ArticleID=RWT082610181333

much more up to date and MUCH more in depth. Also, even since this article, it has been stated that the FPU can be ganged together easily to do 256bit instructions when needed.

 

[nemesismk2]

has begun, these core wars

that was funny

 

[out.of.order]

on the same website today they show the slide under NDA where is described the Bulldozer estimated performance

http://www.donanimhaber.com/islemci...cileri-icin-ilk-resmi-performans-dokumani.htm

i have counted the number of pixels in every rectangle, here the results



they used this benchmarks:

Media: PCMark tv & movie subtest only
Render: Cinebench 11.5
Gaming: 3DMark06 cpu only

based on 1100T reviews i can estimate:

PCMark TV&Movies
1100T 6435
i7-2600K 7703
estimated Bulldozer 7974

Cinebench 11.5
1100T 5.88
i7-2600K 6.89
i7-980x 8.91
estimated Bulldozer 10.98

3DMark06 cpu only
1100T 5986
i7-2600K 6608
estimated Bulldozer 8810

 

[Edrick]

Why are the AMD fanboys getting excited over the fact that a 8 core 2011 CPU can beat an Intel 4 core 2008 CPU?

 

[out.of.order]

Why are the AMD fanboys getting excited over the fact that a 8 core 2011 CPU can beat an Intel 4 core 2008 CPU?

don't know... you should ask to an amd fanboy

 

[Riek]

Why are the AMD fanboys getting excited over the fact that a 8 core 2011 CPU can beat an Intel 4 core 2008 CPU?

one reason could be that a score of ~11 in 11.5 is 50% faster than SB 2600K. (11vs 7)
That the gaming 3dmark 2006 score is higher than an overclocked SB2600K ( 8810 vs SB@4.3GHz -> 8109)
http://www.guru3d.com/article/core-i5-2500k-and-core-i7-2600k-review/16

Or those are at least things that might explain that...

 

[IntelUser2000]

out.of.order:

The score you put for 2600K for 3DMark06 CPU sounds like its an overclocked score. 3DMark06 is pretty multi-threaded for the CPU component.

 

[out.of.order]

out.of.order:

The score you put for 2600K for 3DMark06 CPU sounds like its an overclocked score. 3DMark06 is pretty multi-threaded for the CPU component.

DOH!!!

corrected

thanks

 

[Edrick]

one reason could be that a score of ~11 in 11.5 is 50% faster than SB 2600K. (11vs 7)
That the gaming 3dmark 2006 score is higher than an overclocked SB2600K ( 8810 vs SB@4.3GHz -> 8109)

1. All still estimated.
2. Still 8 core vs 4 core.
3. Price will be based on performance, so it will cost more than a 2600K.
4. No word on speeds yet.

Don't get me wrong, it is a huge step for AMD no question. But I do not think it is the "second coming" as some people tend to think.

 

[out.of.order]

1. All still estimated.
2. Still 8 core vs 4 core.
3. Price will be based on performance, so it will cost more than a 2600K.
4. No word on speeds yet.

Don't get me wrong, it is a huge step for AMD no question. But I do not think it is the "second coming" as some people tend to think.

2. 8 "half" cores vs. 8 threads. remember what JF said: if two cores give 100% one module give 80% of throughput

4. speed doesn't matter... i think the comparison is at same TDP

 

[Riek]

1. All still estimated.
2. Still 8 core vs 4 core.
3. Price will be based on performance, so it will cost more than a 2600K.
4. No word on speeds yet.

Don't get me wrong, it is a huge step for AMD no question. But I do not think it is the "second coming" as some people tend to think.

ofcourse:
1) there is nothing we know but estimates, it is either that or not talk about it at all
2) yes and 8T vs 8T. Core count doesn't matter if it delivers the performance it needs. Intel has a 6core which is also beaten in all those applications which is also on 32nm. (also referring to 1 again )
3) ofcourse price will be based on performance, that doesn't really matter know since we don't know either.
4) indeed, also referring to 3 that we also don't know the price or the performance yet

Either way if those results are in the range what will happen, BD will compete with 6core SB pretty well in all those applications, which considering where they are know would be a small miracle. But again the slides seem fake, which is unfortunate since i know many people would like to know what to expect.

 

[Edrick]

2. 8 "half" cores vs. 8 threads. remember what JF said: if two cores give 100% one module give 80% of throughput

4. speed doesn't matter... i think the comparison is at same TDP

They are not half cores according to JF. They are 8 true cores.

And there is no way that an 8 core BD is @ 95W (which SB is at). Most likely 130W. So it can not be comparing at the same TDP. Now an i7 950 is at 130W, so that is more likely.

 

[out.of.order]

They are not half cores according to JF. They are 8 true cores.

And there is no way that an 8 core BD is @ 95W (which SB is at). Most likely 130W. So it can not be comparing at the same TDP.

i7-950 130W TDP
1100T 125W TDP

the comparison is with that 2 processors

i know SB is at 95W and the HD2000 waste ~15W at full load... SB 4x is also much smaller than the ~310 mm2 Bulldozer

 

[Edrick]

i know SB is at 95W and the HD2000 waste ~15W at full load... SB 4x is also much smaller than the ~310 mm2 Bulldozer

Exactly. That is why comparing the 2 is like comparing apples to oranges.

 

[ElFenix]

arguing about core count is like arguing that a 300 hp 4 cylinder engine is better than a 300 hp V8 despite the fact that both cars get the same 0-60 and the same fuel mileage.

 

[Edrick]

arguing about core count is like arguing that a 300 hp 4 cylinder engine is better than a 300 hp V8 despite the fact that both cars get the same 0-60 and the same fuel mileage.

Sure, if we had 8 core and 4 core CPUs whic had the same die size, same TDW, same speed, same price, same IMC, etc. But we don't.

 

[Castiel]

8 Cores going to be faster then 4 cores.. WOW who would have ever thought that? Not sure where they pulled that graph for gaming from since maybe 2 games take advantage of more then 4 cores. I want to see some Cinebench single threaded benchmarks.

 

[Edrick]

Notice that they say i7 950 on the right, yet on the left they say package 1156?