More Bulldozer Benchmarks (Supposedly from AMD)

[aigomorla]

Lightwave is easy to build clusters for. I have six 2U quad-core Xeons on my desk, which cost about $700 per node with ECC memory. Lightwave also scales linearly with core count. Cheap AMD boards accept non-regestered ECC. Xeons on a server board do not overclock. 48 decent cores > 24 Gen.1 Nehalem cores. I don't pretend expertise in watercooling, but I do with clusters. I can see Zambezi being the basis of a cheap farm. I'd likely gain performance AND money right now by selling off my Xeons and Supermicro boards, and that includes the price of six Intel NICs.

Daimon

Edit: I'm pretty sure a guy doesn't get fired for having new guidelines for existing customers looking to build cheap nodes, and I don't subscribe to the idea that I should be locked into an Intel platform, as I have been since 2006.

I am my IT department; rendering is a hobby.

while i totally agree with you, however they UPGRADED...

that isnt an upgrade... P4 -> AM3+ yes upgrade... C2D -> AM3+ yes another upgrade...
Lynnfield -> AM3 what do you honestly call that? I would of upgraded the cpu to a i7-800 series as it would be more merit.

its a platform shift to a lower system unless its MC.

 

[Zstream]

its a platform shift to a lower system unless its MC.

You know this, how?

 

[aigomorla]

You know this, how?

RAWR... here we go again....

OK if you honestly think AMD can scale with intel even with bulldozer.

The 800 series is a lynnfield LGA1156 socket version of the i7 920, down to hyperthreading ON.
8 working threads on lynnfield vs 8 working threads on the BD.
Thats what it boils down to, and intel has better arch when it comes to encoding, and rendering.
The X6 + Magoney is the only exception, but the cost to get a complete AM3+ package vs, a IT directed CPU upgrade?
Magoney is Socket G34 not Socket AM3+.

Are you honestly going to tell me AM3+ is an UPGRADE vs lynnfield?

 

[Zstream]

RAWR... here we go again....

OK if you honestly think AMD can scale with intel even with bulldozer.

The 800 series is a lynnfield LGA1156 socket version of the i7 920, down to hyperthreading ON.
8 working threads on lynnfield vs 8 working threads on the BD.
Thats what it boils down to, and intel has better arch when it comes to encoding, and rendering.
The X6 + Magoney is the only exception, but the cost to get a complete AM3+ package vs, a IT directed CPU upgrade?
Magoney is Socket G34 not Socket AM3+.

Are you honestly going to tell me AM3+ is an UPGRADE vs lynnfield?

.

I apologize to aigomorla for the false accusation

 

[3DVagabond]

Hey! 8 cores can almost 4 with SMT! Awesome results!!

Somehow I think the person writing this post...

it was predictable since the disclosure of the Bulldozer memory hierarchy :
http://www.realworldtech.com/page.cfm?ArticleID=RWT082610181333&p=8

one common bottleneck for high throughput multithreaded code is the L2$ bandwidth, on paper a Bulldozer module has more or less the same L2$ bandwidth than one SNB core (+ more L2$ usage due to the two small L1D write-through caches vs. the write-back shared L1D on SNB) so it's predictable that 4 BD modules will have roughly the same speed than 4 SNB cores on benchmarks that saturate the L2$ bandwidth (x264 come to mind).

Now, IMHO both designs have a lackluster L2$ throughput, simply not enough bandwidth to handle well optimized AVX-256 code (+ FMA4 on Bulldozer)

Anyway, in other words one BD "core" is more or less half a SNB core. If AMD really intents to advertise the fact that they have "more cores" I hope reviewers will resurect the (pretty good) AMD's "True Performance Initiative" paper.

...isn't going to get this response.

 

[aigomorla]

If you just logged into my account I'm going to be pissed.

Either way, you don't know the performance.

WTF do you mean if i logged onto your account? (i dont have any power in this section and in fact for this comment my mod hat is completely off.)

No you know what im honestly tired of all this bulldozer spam thats going around and people yelling fake benchmarks fake benchmarks when all the fake benchmarks point to the same crap.

If 1/100 start saying something.. it can be false.
If 2/100 start saying something it can still be false.
If 75/100 people start saying something and the 25 who represent the company or are die hard fans of the company deny...

Im sorry, 75 people saying something holds a lot of weight and it looks like the 25 people are HIDING SOMETHING.

and if out of the 25 people that are hiding something loses 2 key members during a launch it smells to me like they are trying to sweep something under the RUG very fast.


So from what ive seen and from 5 credible sources your still saying benchmarks are fraud and fake, from those 5 public websites even.

No im going with more AMD your trying to HIDE SHIT BECAUSE YOU GOT CAUGHT WITH YOUR PANTS DOWN.
And watch when REAL benchmarks happen and it shows whats current, its going to be more of a PR nightmare then what AMD is doing NOW.
Why? cuz they just told all the public web sites which probably gains a lot of weight that they are all FAKE and inaccurate.

But hey they fired their key members so they can always TOSS the blame on the people who left....
This is what it smells like to me... blame the two guys who left in all the mess ups and dely the crap out of the product so no one can validate whose telling the truth.



Seriously people.. when i threw a gulftown preview did ANYONE yell fake?
Why is everyone yelling FAKE when people who are more crediable then i am are throwing BULLDOZER reviews.

So are you guys telling me that my Gulftown reviews were honestly FAKE?? No.. better yet.. intel didnt TUNE them properly?
No they were totally inaccurate correct? Had to take them with a complete grain of salt....

BULLSHIT....

Thats what i say with Bulldozer BULLSHIT.


<sorry i snapped... and thats it for me and bulldozer>

 

[Zstream]

Anger management much

I never said AMD was this or that. In fact I don't think anyone really has. I stated that you don't know the benchmarks. Why are you getting upset over a cpu?

 

[Atreidin]

So if a lie spreads and is repeated by people over and over again, it suddenly becomes the truth?

A statement doesn't magically become true simply because more and more people are repeating it. That's dumb logic.

 

[aigomorla]

I never said AMD was this or that. In fact I don't think anyone really has. I stated that you don't know the benchmarks. Why are you getting upset over a cpu?

cuz a lot of my friends are being targeted at by people now for producing "fake" results, when i know that they are about as BLUNT as one can get.

And to be honest im tired of this bullshit they are doing.
Its Phenom launch all over agian.. delays more delays..
To be honest im so fed up i dont think im going to even move my friend on a X6 anymore... im gonna tell her she is screwed on her 965.

So if a lie spreads and is repeated by people over and over again, it suddenly becomes the truth?

A statement doesn't magically become true simply because more and more people are repeating it. That's dumb logic.

Lies?

Once again... if the people who presented the SAME reviews had the SAME exact results... and have a history of NOT BSing reviews... whose honestly the one starting rumors?

And if the company is holding the chips because they DONT want the scores validated... isnt that the kind of scenario it is now?

Once u call someone a public liar... even false, you know how hard it is to erase?
You think this is fair for AMD?
And when the scores are validated... do you think AMD will say sorry to us?

No they will just blame the two who left and shift the topic on llano.

 

[bronxzv]

Somehow I think the person writing this post...
...isn't going to get this response.

My understanding was that the poster was ironical and was mocking at the BD 8 cores that are almost as fast as 4 SNB cores in AMD's cherry picked benchmarks, now maybe I got him wrong?

 

[3DVagabond]

My understanding was that the poster was ironical and was mocking at the BD 8 cores that are almost as fast as 4 SNB cores in AMD's cherry picked benchmarks, now maybe I got him wrong?

Your response was fine. :thumbsup: The original post was so lowbrow though that it's not worthy of your thoughtful response, was my point.

 

[bronxzv]

Your response was fine. :thumbsup: The original post was so lowbrow though that it's not worthy of your thoughtful response, was my point.

I have to confess that my primary goal was to post my thoughts on the L2 cache bandwidth, I selected this post pretty much randomly

 

[BlueBlazer]

Those slides are from AMD's Tech Day of 31 August 2011,

How could this slip by me? First the slides had been cropped (at the bottom) to hide the dates. And then it daunted me on this particular slide......

Can anyone remember that 8.4GHz "Guiness World Record" date? Here's the clue. :hmm:

 

[bronxzv]

Out of curiosity I have checked the AMD site for "True Performance Initiative" and I got this old press release:

http://www.amd.com/us/press-releases/Pages/Press_Release_10473.aspx

adapted to the current situation it may read like that:

"

True Performance Initiative
Through the True Performance Initiative, or TPI, AMD is setting out to assist customers in understanding the benefits of PC performance. TPI also will help define a new, more accurate measure of processor performance for standard applications. [...] Historically, x86 microprocessors have improved both instructions (work) per clock per core and number of cores compared to older generations. However, this is not true with some processors today. Therefore, the number of cores cannot be solely relied upon as a measure of system performance.
"Consumers are beginning to understand that the true indicator of performance is how fast their applications run, not the number of cores of their processor. Core count measures only how many cylinders the engine has, not how much torque it delivers [...]

"

 

[Blitzvogel]

Out of curiosity I have checked the AMD site for "True Performance Initiative" and I got this old press release:

http://www.amd.com/us/press-releases/Pages/Press_Release_10473.aspx

adapted to the current situation it may read like that:

"

True Performance Initiative
Through the True Performance Initiative, or TPI, AMD is setting out to assist customers in understanding the benefits of PC performance. TPI also will help define a new, more accurate measure of processor performance for standard applications. [...] Historically, x86 microprocessors have improved both instructions (work) per clock per core and number of cores compared to older generations. However, this is not true with some processors today. Therefore, the number of cores cannot be solely relied upon as a measure of system performance.
"Consumers are beginning to understand that the true indicator of performance is how fast their applications run, not the number of cores of their processor. Core count measures only how many cylinders the engine has, not how much torque it delivers [...]

"

I know AMD wants to look honest here, but don't they realize that chances are that Intel is going to catch them hardcore here?

 

[trollolo]

i don't understand why people are saying that these slides should be dismissed because amd.com isn't acknowledging them. they're leaked, just like the iphone 4 was, and how those slides from verizons droid rollout were

 

[3DVagabond]

i don't understand why people are saying that these slides should be dismissed because amd.com isn't acknowledging them. they're leaked, just like the iphone 4 was, and how those slides from verizons droid rollout were

Because we've seen to many "leaks" that are nothing but made up Fud. They can't be just accepted at face value because they might be more Fud.

 

[sequoia464]

Because we've seen to many "leaks" that are nothing but made up Fud. They can't be just accepted at face value because they might be more Fud.

Haven't seen any news here for a while - how is the rebuilding coming along?

 

[3DVagabond]

Haven't seen any news here for a while - how is the rebuilding coming along?

Rebuilding? LOL We're still in the demolition stage. Well, that and holding onto the walls as we're still suffering a lot of after shocks. NZ is small with not a lot of resources to throw at this. Kiwi's (I'm American) are a very proud bunch too. Most other countries would be screaming for foreign aide. NZ's going to do it on their own though. Most predictions are 5yrs. to get back to some sort of normalcy. Even after that I'm sure a lot of things will end up downscaled because of modern costs of building.

Thanks for asking, mate. :thumbsup:

 

[dac7nco]

Aigo,

this is a hobby. I'll look at real benchmarks first, of course, but the prospect of spending a few grand on an ECC-supported platform with a highly-overclockable unlocked CPU matters a great deal to me, especially if I can carry over the memory. I expect Bulldozer to be a dog in terms of core efficiency, but I do expect it to be a boost in an application that scales well to THOUSANDS of cores. I could build a 2U Bulldozer node for $400, including NIC, and that's what I'm looking at: $2,400 for forty-eight so-so cores. My dual-5645 SR2 system was a pig as well, but I don't miss it one bit: Once Quicksync came around, I had no need for it. For me it's about buying the most per-dollar, and AMDs clusterfuck of a CPU may actually meet my narrow needs.

Daimon

 

[3DVagabond]

Aigo,

this is a hobby. I'll look at real benchmarks first, of course, but the prospect of spending a few grand on an ECC-supported platform with a highly-overclockable unlocked CPU matters a great deal to me, especially if I can carry over the memory. I expect Bulldozer to be a dog in terms of core efficiency, but I do expect it to be a boost in an application that scales well to THOUSANDS of cores. I could build a 2U Bulldozer node for $400, including NIC, and that's what I'm looking at: $2,400 for forty-eight so-so cores. My dual-5645 SR2 system was a pig as well, but I don't miss it one bit: Once Quicksync came around, I had no need for it. For me it's about buying the most per-dollar, and AMDs clusterfuck of a CPU may actually meet my narrow needs.

Daimon

Bulldozer might well meet a good number more peoples needs as well. We'll have to wait and see of course. I don't know why so many people are ready to pronounce it dead before it's even released?

 

[Olikan]

i don't understand why people are saying that these slides should be dismissed because amd.com isn't acknowledging them. they're leaked, just like the iphone 4 was, and how those slides from verizons droid rollout were

because...

no roadmaps
no architecture details
many benchmarks errors
slides are "ugly"

or amd fail hard at making slides, or they are fake...

 

[Vesku]

Actually winning some and losing some compared to the 2600K makes a lot of sense.

8 threads * .85 (Nehalem level IPC) * .8 (cost of BD cluster) = 5.44 "SB cores"

+/- a bit and it is consistent with current close to release performance rumors.

So it should meet my personal requirements, here's to hoping.

 

[MisterMac]

Aigo,

this is a hobby. I'll look at real benchmarks first, of course, but the prospect of spending a few grand on an ECC-supported platform with a highly-overclockable unlocked CPU matters a great deal to me, especially if I can carry over the memory. I expect Bulldozer to be a dog in terms of core efficiency, but I do expect it to be a boost in an application that scales well to THOUSANDS of cores. I could build a 2U Bulldozer node for $400, including NIC, and that's what I'm looking at: $2,400 for forty-eight so-so cores. My dual-5645 SR2 system was a pig as well, but I don't miss it one bit: Once Quicksync came around, I had no need for it. For me it's about buying the most per-dollar, and AMDs clusterfuck of a CPU may actually meet my narrow needs.

Daimon

This.

This exactly.


It'a a serverchip.


I have some servers running some own software with X3440 procs in them.
(Nehalem)
It scales based on systeminfo. threads availeble.
Even if i went older and got CORE 2 architecture Xeon's i'd never ever MAX a core(with 2 threads running) - hell i won't even come close to 50% utilization.

I'd never dream of replacing my home machine with AMD for my gaming needs.
It' sucks BEEP BEEP for any constantly intensive processing needs where utilization is 80%+.

But a 16 threaded interlagos chip? dual interlagos chip even!
32 threads for the price of 8 Intel's?

As long as their minimumly half of a Xeon Westmere/Nehalem and priced accordingly, bring it on...bring it on now!

 

[Rubycon]

Where's the proof of that system running LinX for two weeks straight with no errors?

How could this slip by me? First the slides had been cropped (at the bottom) to hide the dates. And then it daunted me on this particular slide......

Can anyone remember that 8.4GHz "Guiness World Record" date? Here's the clue. :hmm: