AMD Bulldozer Simple Facts Please!!!!!

I recommend you upgrade to the latest Phenom II X6. It's not much more over the fastest Phenom II X4 and it will be the last upgrade for your motherboard I think.

Bulldozer is socket AM3+ so we can't plug them into an AM3 board but can plug in an AM3 CPU into an AM3 mobo.

Bulldozer has not been revealed to have HT, but it does have clusters of cores which share common parts of cores called modules. Each AMD module has 2 cores.

It will be launched 1st half of next year.

Feed me...

I estimate that Bulldozer will achieve parity with Intel's Nehalem arch.

That's all.

I think Sandy Bridge will beat Bulldozer, but Bulldozer will be competitive and in roughly the same spot as Thuban, although I think AMD will breach the $300 barrier with Bulldozer.

I think a lot of people will be disappointed if Bulldozer is no better than Thuban; I know I will.
I'm thinking it will edge past Nehalem, get similarly beaten by Sandy Bridge, but cost a lot less, and perhaps, just perhaps, people will wake up and realize that a mid to high end prior generation (e.g. X4 or i5) is still all the CPU that 95% of people need (and 90% of those could probably use a X2 or C2D).

If the 12.5% performance increase is right then SB will be faster.

Then again we have to wait until BD is released. As far as we know its been deleayed.

I think people are still kind of underestimating Bulldozer, for two reasons:
1. I don't think they understand how the cores on a bulldozer chip work, since i hear many people saying it may beat Sandy in multi threaded tasks but definitely not single threaded tasks. I don't think thats a fair assumption since in single threaded tasks, one bulldozer core has access to all resources in the module, and gets to use both 128 bit FMACS to equal Sandy Bridges 256 bit FPU. Also, if you have a program that uses 4 cores, and you have an 8 core (4 module) Bulldozer chip for example, instead of two modules being active using all cores, only one core in each of the four modules will be active, resulting in 4 beefy cores.

2. This ones simple. Since the 1.25 bil settlement, AMD has said they will use all of intels available instruction sets (including AVX) in addition to new AMD instruction sets. This includes instructions for Fused Multiply- Add, which intel isnt implementing until Haswell.

As for sources, check the settlement info and any other source of Bulldozer info. I didnt use a specific source, this is all stuff i remember reading.

Good idea.

I;m wondering why hasn't some moderator merge all those threads about AMD BD together, like how mousempnkey merged all those thread about HD6Ks,


I hope not though, it'll get really confusing

That's because I don't stop in here as much as I used to but it's an interesting thought.

The problem with all of that math is the "50% faster with 33% more cores" statement is about a fully utilized server processor in server workloads, 12 threads pegged vs. 16 cores pegged.

Imagine how long it takes you to get to work in rush hour traffic. Then think about that same road if there is light traffic or you are the only car on the road. Big difference.

I think people will be very happy and very surprised with the performance, but as pointed out above, it is best for everyone to wait until the actual results are out because all of the speculation is wrong. People just don't understand the architecture well enough to make an accurate assessment.

You kinda missed this part:

I am not trusting anything. I can pull all the numbers to speculate. Thats what I did. I never said it was right or concrete. Thats why for the time being, until a third party has BD to test and show us what it can do Sandy Bridge is the next top end CPU performance wise since we have concrete info from Anand.

Same goes for you JF-AMD. I never trust anything straight from the company source. No offense but its hard to trust people who work for the company. Its like the reason why I got Windows 7. Not because Microsoft touted it to be great. but because I had a few friends at my job who had the Beta and said it was much better than Vista.

The last time we got was the over hyped 40% from Barcelona. That turned out to be worse than Kentsfield. Thats why I am not trusting anything yet. I am even not 100% on Anands review since it was done in controled areas and with unfinalized hardware.

I know you as a person may not, but you do not have any control over AMDs marketing department. They will spin what they need to to get sales..

Thats why I said I am waiting for BD results. I have seen sources claiming this or that and otehrs who say it will beat SB. None of it is 100% concrete. Right now I have concrete info showing SB as being better than Nehalem. So with what I have, SB is showing to be the best CPU someone can buy for performance in Q4 2010/Q1 2011. When we have BD results, if they are better than SB I will state that. Of course it would be nice to know WHEN we can expect BD since rumors float around on the web that it wont have to face SB, more that it will face Ivy Bridge instead.

And I have yet to make up my mind since I wont be building a new system until late 2011. More towards 22nm since I want a system with SATA 6.0, USB 3.0 and possibly light peak (I love fiber based tech).

Hey JF ... glad to see you here.

AFG I think they will manage a 9W special from what I saw elswhere.

ENTP btw.

2 things:

1. I am the director of product marketing, so I actually am in control of AMD's marketing when it comes to any claim about performance. I am the one that would decide what we say about performance. We are more conservative than you would think.

2. The 40% claim was from the head of the server division. It was in reference to STREAM memory throughput, not a client claim on client benchmarks. I think too many people interpreted that statement to be 40% better client performance on client benchmarks, it was never intended to be that.

Well I can see a resurgence in Notebook sales and we all might finally have 4hrs of real battery use without having to resort to the sheer blasphemy of using a Netboot.

I played with one of those woefull EEPC things for a week and threw it away ... literally ... the Atoms are rubbish.

I gave it to a work mate who celebrated ... then played with it for a week and quietly filed it into a drawer ... next to the financial calculator with the printer, the old stamps, and sundry items of no real value or interest.

A historical anomaly ... must be an aweful lot of idiots out there ... now they are buying i-pads I guess ... what next month?

Seriously, I want something portable that won't grind to a halt when the graphics gets rough and that means something weighing like a boat anchor at present ... and connected to mains power.

Lets see what AMD's offers ... I can't see Intel matching the combined CPU / GPU power ... unless they buy NVidia.

I keep predicting that ... um ... and getting it wrong.

I actually use a Macbook for about 4-6 hours battery (4 wi-fi, 6 energy saver). Had it since 2007. Nifty little piece of $2000 hardware I didn't pay for.

Netbooks are rubbish. Anything in general with a 10" or smaller screen is rubbish.

You might want to take a look at the new CULVs Reynod. Ain't much, but it's the closest thing to your sliced cheese... (you won't buy a Mac will you?)

What will be a dealbreaker will definitely be a good OpenGL driver.

Atoms doing Photoshop aren't that bad if they use CS4+ with a good OpenGL GPU.

Netbooks tend to have slightly more battery life, but believe me, there's no way I'm ever going to use one for school, it's so slow for Adobe CS4.

/hugs Reynod

I recall the 40% claim. It was an outright claim. It was not specified that AMD were referencing STREAM memory throughput. It is like an advertising claim written on an ad... except those on an add require an asterix and small print writing.

AMD is by no means the only company guilty of this (or product type).
http://news.cnet.com/AMD-Go-to-Bar [...] 53bcontent

I'm still waiting for Wireless Products that can achieve anything near their "theoretical" throughput. 54Mbps... BS! 108Mbps? BS! And don't get me started on Wireless N. Hell you could be right near or a tad further then near (as recommended) and still never hit anything close to those numbers.

+1 for the hybrid graphics. Since you DO work for the Server division, would you like to tell if workstation boards/chip sets will support such a feature? I think a lot of the folks that run workstation GPUs would like this....

edit: mobile workstations; not desktop/2P workstations.

Also, will Bulldozer allow one to completely turn off the 2nd (or more) CPU in a 2P (or more) set up along with turning off un-needed cores of CPU1/0.

I wonder how good M$ will be on all this?

So let me get this straight, AMD considers some of the 2P stuff as workstations? Wouldn't all 2P stuff technically be under "server" as they use the server chipsets,etc? Exactly what does AMD consider to be "server" and "workstation"?

Btw, when I was referring to "workstation" I was referring to a 2P set up with a workstation GPU. Sorry for any confusion.

Make sense. Hopefully, the Linux community will be able to implement things like this.... I'm pretty damn sure Microsoft will do it.

Kaspro, its completely off topic, if you want make a thread regarding this

A. I don't work in client.

B. That question has been answered on many forums already. Bulldozer will not be supported in AM3.

One more question to JF-AMD:
In regards to Stream, I find that there are not many plug-ins/software that use OpenCL/Stream,etc currently on the AMD side. On the other hand, there are quite a notable collection of CUDA enabled plugins/software available with tangible benifits. For example, CS5 Mercury playback engine. Another example is n-Tracer,iray,etc for 3DSMax. Is AMD doing much to promote OpenCL,etc? I see nVida promoting Tesla/CUDA however, I don't see AMD doing anything like this currently. We (a friend and I, do note I didn't write the code; it's a CFD sim if you wanted to know) have found that CUDA is significantly easier to use than OpenCL/CirectCompute due to the lack of documentation,etc for OpenCL, AMD may want to look in to improving this.

What about the HPC crowd? Surely they would be driven by the CPU and the related systems (esp. CPU >> RAM subsystem)?