S/A: "AMD outs bulldozer based orochi die"

[bryanW1995]

Are your really that far out there . AMD has nothing to compete with intel on the desktop . Nothing but price . As far as blaming the software venders get real once they can do it correctly Intel will have all the real core they need to do the Job . AMD64 is a perfect example of a hype lie . The same as offerring more than 6 threads on the desk top will see it slowly but surely. Intel and AMD are both guilty here. But you do have to build the hardware befor the software to realize the gain . So you believe the egg came before the chicken . Not so the chicken had to lay the egg and than mother those eggs and keep them warm so as to hatch. I suppose you will argue the incubator was befor the egg.

you lost me at "befor"

 

[ModestGamer]

lol.

nemesis was talking about a claim made by amd marketing (don't remember whom it was but I know it wasn't john) that barcelona would beat intel's cpu by 40%. it did this clock/clock in that one isolated benchmark iirc, but amd had clocking issues and some other snafus with barcelona. intel's 10ghz claim was even fluffier, but they've produced since then so it doesn't come up as much as the "40% faster than intel" claim these days. if BD knocks SB around a bit but intel makes ridiculous claims beforehand then the pendulum will again swing the other direction no doubt.

@modest gamer: really? amd's most expensive cpu is the 1090t. intel's equivalently priced cpu is the i7 950. let's see what anandtech bench thinks. hmmm, 24-3 with a few ties. overall say ~ 10% faster for i7 950? and that's with 4 cores + ht instead of 6 cores. also, oc headroom is a bit higher on i7 950 but it's clocked ~ 5 % lower by default, so if anything oc'ing gives the i7 even more of an advantage. so now intel has a CLEAR lead at anthing over $200, not exactly what an amd-leaning enthusiast wants to see/hear.

seriously. the 1090t is on average $25 cheaper at every vendor. Theres fud afoot and I'm not gonna spend all night chasing it. Your on your own for that one.

Are your really that far out there . AMD has nothing to compete with intel on the desktop . Nothing but price . As far as blaming the software venders get real once they can do it correctly Intel will have all the real core they need to do the Job . AMD64 is a perfect example of a hype lie . The same as offerring more than 6 threads on the desk top will see it slowly but surely. Intel and AMD are both guilty here. But you do have to build the hardware befor the software to realize the gain . So you believe the egg came before the chicken . Not so the chicken had to lay the egg and than mother those eggs and keep them warm so as to hatch. I suppose you will argue the incubator was befor the egg.

Really AMD has nothing to compete with intel ? Where do you guys come up with this stuff. they compete in plenty of places and the cpu performance seperation claims aren't that great to matter.Wow I wonder what operating systems did when we had dual pentium 2 and pentium 3 and dual socket 478 systems ? Wow how could they have ever known that multi core or cpu solutions were goingto come along.

Geez.

 

[Wreckage]

Before anyone sticks a fork in bulldozer, might I remind everyone that currently none of the schedules have changed in the last 60 days, so this new round of pessimistic rumors are pretty much that, rumors.

I am going to be in vegas twice before the end of the year. Maybe I can find a sports book where I can place a bet on bulldozer. Based on the current rumors, I could make a killing off of the under.

Well, I'm looking forward to Bulldozer. Hopefully everything is on track. I went to an AMD tech tour a few years back and really enjoyed it. I hope they do something similar soon as you really get a ton of information and can question people with close knowledge of the product.

 

[piesquared]

Are your really that far out there . AMD has nothing to compete with intel on the desktop . Nothing but price . As far as blaming the software venders get real once they can do it correctly Intel will have all the real core they need to do the Job . AMD64 is a perfect example of a hype lie . The same as offerring more than 6 threads on the desk top will see it slowly but surely. Intel and AMD are both guilty here. But you do have to build the hardware befor the software to realize the gain . So you believe the egg came before the chicken . Not so the chicken had to lay the egg and than mother those eggs and keep them warm so as to hatch. I suppose you will argue the incubator was befor the egg.

That's the most ridiculous thing i've read in a long time. What the hell else are products supposed to compete on if not price?? Surely you realize that's how the free market works... Should we just throw money out the window and fire up the midieval barter system to aquire goods and services now?

Does an i3 competetive with an i7? Nope. Is it priced accordingly? Yup.
By your line of thinking, it's completely reasonable to throw everything out of the market from both companies below the 980 because nothing can compete with it. How bizarre.

 

[Dark Shroud]

AMD has Intel beat in at least one area, I can't buy an Intel physical quad core at $100 or even $150.

 

[tatertot]

Before anyone sticks a fork in bulldozer, might I remind everyone that currently none of the schedules have changed in the last 60 days, so this new round of pessimistic rumors are pretty much that, rumors.

I am going to be in vegas twice before the end of the year. Maybe I can find a sports book where I can place a bet on bulldozer. Based on the current rumors, I could make a killing off of the under.

I've read through what you wrote there, and what I don't see is any assertion that Charlie is actually wrong in his claim of a Q4 2011 Bulldozer launch for servers. Saying rumors are rumors is the same sort of thing... rumors are sometimes correct.

"On track", "the schedule hasn't changed", etc, mean nothing when the public window stands at "sometime in 2011". Charlie's claim is within that window.

 

[JFAMD]

As someone that knows that the official position right now is 2011 launch, I can't comment on any quarter. No matter how people try to trick me. When we have quarter granularity from an official source it will be online in a few minutes.

Until then, no comment on quarter.

 

[ModestGamer]

As someone that knows that the official position right now is 2011 launch, I can't comment on any quarter. No matter how people try to trick me. When we have quarter granularity from an official source it will be online in a few minutes.

Until then, no comment on quarter.

Just keep us updated on the client/server part launch. Appreciate all the info we can get.Performance data would be great. I am looking to upgrade in 2011 so it would be nice to plan ahead.

 

[ModestGamer]

I've read through what you wrote there, and what I don't see is any assertion that Charlie is actually wrong in his claim of a Q4 2011 Bulldozer launch for servers. Saying rumors are rumors is the same sort of thing... rumors are sometimes correct.

"On track", "the schedule hasn't changed", etc, mean nothing when the public window stands at "sometime in 2011". Charlie's claim is within that window.

they might launch it q1 but they aren't commenting. Leave it at that.

 

[Ancalagon44]

Since Barcelona:

Shanghai - 3 months early, higher than expected clock speed

With an enormous performance deficit and limited overclocking compared to equivalent Intel chips, as well as the infamous TLB bug?

Dont get me wrong, I'm definitely an AMD fan, being on my third AMD processor in a row, but Shanghai was a disaster for AMD. Istanbul was a lot better, but my guess is that due to its die size, it has to be sold at a much lower margin than an equivalent Intel CPU in order to be competitive.

So... I guess its not AMD's timing thats off, its that a lot of us long for the old days of the Athlon 64 when Intel just could not compete on performance. Its nice having the budget gaming performance crown, yes, but I hope AMD has higher aspirations than that.

 

[DrMrLordX]

With an enormous performance deficit and limited overclocking compared to equivalent Intel chips, as well as the infamous TLB bug?

Shanghai was C2 stepping and did not have the TLB bug. You are thinking of Barcelona. Also, server/workstation CPUs (Opterons) are not billed on their overclocking abilities, or lack thereof.

 

[Ancalagon44]

Shanghai was C2 stepping and did not have the TLB bug. You are thinking of Barcelona. Also, server/workstation CPUs (Opterons) are not billed on their overclocking abilities, or lack thereof.

No, but they did underperform compared available Intel parts, and the situation was even worse on the desktop where AMD's HyperTransport bus didnt give it much of an advantage.

 

[DrMrLordX]

No, but they did underperform compared available Intel parts, and the situation was even worse on the desktop where AMD's HyperTransport bus didnt give it much of an advantage.

I'm sure AMD is still crying in its collective beer over Shanghai. Still, it had no TLB bug, and overclocking (or a lack thereof) was not one of its weak points. It was better than Barcelona, and, to return to what JFAMD said, it was ahead of schedule, so I guess it did okay.

So can we go back to discussing Bulldozer/Orochi now?

 

[Triskain]

No, but they did underperform compared available Intel parts, and the situation was even worse on the desktop where AMD's HyperTransport bus didnt give it much of an advantage.

I remember Anandtech's own Johan de Gelas saying something else about Shanghai:

Right now, it is clear that the latest AMD Opteron is in the lead. We are really at the pivotal moment in time. No matter how good the current Xeon "Harpertown" and "Dunnington" architectures are, they lose too many battles due to the platform they are running on.

Quoted from this comparison.

Of course that only lasted until Gainestown was released but until then Shanghai undoubtetly was the best in the server segment.

And now back to Bulldozer please.

 

[DrMrLordX]

And now back to Bulldozer please.

Absolutely. This brings an interesting point to mind . . . AMD used to have a lead in the server market in the 4p+ market (and to a lesser extent in the 2p market) thanks to a platform that lent itself better to the server market than what Intel had (HT > FSB). Is Bulldozer going to bring anything new to the platform table that will help AMD regain their edge in the enterprise sector? We've spent so much time talking about the modules that nobody seems to be talking about the platform.

 

[Triskain]

Bulldozer will utilize the same G34 platform as Magny-Cours without any changes. That's alright because:

1.) The G34 Platform is pretty decent.
2.) AMD has enough validation to do for Bulldozer, adding a new platform into the picture would be a nightmare.

The AMD roadmaps say that a new platfrom will be introduced in 2012 but it doesn't say if that will entail a new socket. If AMD wants to integrate PCIe into the CPU like Intel does then a new socket will be necessary.

 

[Idontcare]

Absolutely. This brings an interesting point to mind . . . AMD used to have a lead in the server market in the 4p+ market (and to a lesser extent in the 2p market) thanks to a platform that lent itself better to the server market than what Intel had (HT > FSB). Is Bulldozer going to bring anything new to the platform table that will help AMD regain their edge in the enterprise sector? We've spent so much time talking about the modules that nobody seems to be talking about the platform.

^ that's because AMD has only really educated on the core-stuff regarding modules.

All the uncore stuff that takes care of platform interfacing hasn't really be made overly public yet as far as I know. Yes we know socket info and that infers platform, but we don't know if BD enables anything new within the infrastructure of existing platforms or even in future platforms.

It is mostly the realm of conjecture once you step outside the module.

 

[JFAMD]

There will be lots of enhancements to the memory controller, significant advantages there. But no details until launch. The big message there is that there will be a big improvement in memory throughput, and that improvement is not tied to higher speed memory per se. If you put in the same speed memory you would see a pretty good jump in throughput.

 

[busydude]

There will be lots of enhancements to the memory controller, significant advantages there. But no details until launch. The big message there is that there will be a big improvement in memory throughput, and that improvement is not tied to higher speed memory per se. If you put in the same speed memory you would see a pretty good jump in throughput.

Yay, for better northbridge. :thumbsup:

 

[Idontcare]

I reads that to mean moar channels!

 

[Sp12]

Well, JFAMD, that blog kinda answer my question, but not quite.

Assuming a dual-threaded app, how does the performance scaling in a bulldozer module compare to the scaling inconsistencies we saw in the C2Q series between threads on the same die vs threads on the other dualie die?

Assuming a purely single threaded app, does cannibalizing the shared resources of a module to one integer core have any performance benefits? How significant? I know the second core has 80% of the performance of two separate dies, but I'm interested in the performance loss in single-threaded designs in a module.

 

[JFAMD]

Here are some simple numbers (for demonstration only):

1 thread on 1 core: 100%
2 threads on 1 core (HT): 120%
2 threads on 2 cores in 1 module: 180%
2 threads on 2 cores in 2 modules: ~200%

I say ~200% because Amdahl's law prevents it from really being 200%, but you get the general idea.

As you can see, when you start getting more threads. an HT system has more of a penalty for putting multiple threads on the same core.

 

[cbn]

Well, JFAMD, that blog kinda answer my question, but not quite.

Assuming a dual-threaded app, how does the performance scaling in a bulldozer module compare to the scaling inconsistencies we saw in the C2Q series between threads on the same die vs threads on the other dualie die?

Assuming a purely single threaded app, does cannibalizing the shared resources of a module to one integer core have any performance benefits? How significant? I know the second core has 80% of the performance of two separate dies, but I'm interested in the performance loss in single-threaded designs in a module.

JFAMD just gave estimates for what I believe are the server parts.

I just wonder how this will change if AMD decides to include turbo on one or both cores in the consumer Bulldozer modules?

 

[cbn]

http://www.xbitlabs.com/news/cpu/di...e_Advantage_of_Performance_Architectures.html

Some more information.

P.S. I thought the following comment was interesting.

At least in initial designs, multi-core chips will consist of several modules, each of which will have two independent integer cores (that will share fetch, decode and L2 functionality) with dedicated schedulers and two 128-bit FMAC pipes with one FP scheduler.

 

[jvroig]

I just wonder how this will change if AMD decides to include turbo on one or both cores in the consumer Bulldozer modules?

Bulldozer does have turbo. "Real turbo" as Anand called it, as opposed to the more simplistic TurboCORE found in Thuban.