Bulldozers Weak/Strong points?

[Vesku]

Aren't the pre-Sandybridge i7 chips hot running? It's just they are excellent processors, where as the P4s were hot without excellent all around performance to go along with it.

I think that's one of the reasons there is a lot of questions and speculation. There's the question of "will this be AMD's P4" lingering in the back of people's heads. Hope it's at the very least a good chip for the server market.

 

[Topweasel]

I hope for IPC it can beat nehalem, but I don't think it will.

Doesn't need to. Without know lots of things, base CPU speed, Single and Dual core Turbo mode speeds, the affect of the reduced L2 latency, and the affect of some of the shared computing functions on single thread applications. It could in theory be the best CPU ever invented and not be that much better per "core" then a P4.

 

[Topweasel]

Aren't the pre-Sandybridge i7 chips hot running? It's just they are excellent processors, where as the P4s were hot without excellent all around performance to go along with it.

I think that's one of the reasons there is a lot of questions and speculation. There's the question of "will this be AMD's P4" lingering in the back of people's heads. Hope it's at the very least a good chip for the server market.

I would say the Phenom is AMD's P4. The original was talked up and up and up. Came out disappointing, couldn't beat hardware already released (Core 2 Duo). Each improvement they made to it Intel would have something to beat it.

So the question really is. Is this AMD's Core 2 Duo or AMD's Cyrix III.

 

[formulav8]

Didn't AMD already have a very strong FPU compared to Core2 and newer Intel?

Compared to the Core2 and newer the PhII biggest weakeness is Integer performance. Have they mentioned anything about the Int changes? :\

 

[gdansk]

Didn't AMD already have a very strong FPU compared to Core2 and newer Intel?

Compared to the Core2 and newer the PhII biggest weakeness is Integer performance. Have they mentioned anything about the Int changes? :\

Phenom II did score better in float-point workloads than it did comparatively in other workloads, for example Cinebench and POVRay. Sandy Bridge muddles the results because it can use AVX (I don't know if these programs do, however).

Yes, the optimization guide (section 2.1) gives some general characteristics:

• Four-way AMD64 instruction decoding (This is a theoretical limit.)
• Dynamic scheduling and speculative execution
• Two-way integer execution
• Two-way address generation

Additionally, in section 2.10, AMD mentions:

2.10.1 Integer Schedule
The scheduler can receive and schedule up to four micro-ops (μops) in a dispatch group per cycle.
The scheduler tracks operand availability and dependency information as part of its task of issuing
μops to be executed. It also assures that older μops which have been waiting for operands are
executed in a timely manner. The scheduler also manages register mapping and renaming.
2.10.2 Integer Execution Unit
There are four integer execution units per core. Two units which handle all arithmetic, logical and
shift operations (EX). And two which handle address generation and simple ALU operations
(AGLU). Figure 2 shows a block diagram for one integer cluster. There are two such integer clusters
per compute unit

From this we can generally grok that integer performance has been "redesigned". They now have 2 AGLU and 2 ALU rather than 3 ALU and 2 AGU (well 3, but the third AGU is not used IIRC). But they actually have 4 pipelines now, but they list it clearly as a theoretical limit. The AGLU sounds particularly interesting, but I cannot find exactly what operations it can or does perform.

Additionally in regards to FPU performance the guide mentions (section 2.7):

The AMD Family 15h processor floating point unit (FPU) was designed to provide four times the raw
FADD and FMUL bandwidth as the original AMD Opteron

Given that it is designed for 4 times the performance but shared* between two cores, it sounds like they've designed for each core to have twice as much as performance as the previous generation. To me, Bulldozer sounds like a float-crunching beast.

* I'm not sure if this is the right word, as one core can use the entire 256-bit pipelile or both cores can use half the FPU. So it doesn't have to be shared (in either case really).

 

[IntelUser2000]

Phenom II did score better in float-point workloads than it did comparatively in other workloads, for example Cinebench and POVRay. Sandy Bridge muddles the results because it can use AVX (I don't know if these programs do, however).

Not really. The gains are within the expectations of non-vector based enhancements. I don't think any programs existed that used AVX at launch, other than those that measure FLOPS. AVX optimized applications should show far greater gain.

The AMD Family 15h processor floating point unit (FPU) was designed to provide four times the raw
FADD and FMUL bandwidth as the original AMD Opteron.

The bold seems to be the key word. FP throughput was doubled with Agena/Barcelona with 128-bit SSE units. Bulldozer adds FMA for another 2x.

Original Opteron: 64-bit
Agena/Barcelona: 128-bit(2x)
Bulldozer: 128-bit FMA(2x from Agena/Barcelona)

 

[gdansk]

Not really. The gains are within the expectations of non-vector based enhancements. I don't think any programs existed that used AVX at launch, other than those that measure FLOPS. AVX optimized applications should show far greater gain.

I should say could muddle. It appears that they're all using the same instructions at this point. Of course, AVX based metrics will be more applicable when Bulldozer comes along so that both can be using the same instructions.

The bold seems to be the key word. FP throughput was doubled with Agena/Barcelona with 128-bit SSE units. Bulldozer adds FMA for another 2x.

Original Opteron: 64-bit
Agena/Barcelona: 128-bit(2x)
Bulldozer: 128-bit FMA(2x from Agena/Barcelona)

Ah, I missed that qualifier. Well it still has 4(x2) FPU versus the 6 in Thuban. Including any improvements in the FPU it should still be more than 33% more effective. Does anyone have an idea how the FMA4 will effect performance?

 

[JFAMD]

Keep in mind that unless all of your apps are recompiled, you don't get to take advantage of AVX. Which means that the FP comparison put AMD clearly in the lead.

And on an AVX-128 instruction Intel requires customers to pad the top register (128-256) with zeroes. This essentially means that even though it is a 256-bit pipe, a 128-bit instruction consumes the same pipe.

FP benchmarks are going to be interesting because they will all be optimized around 256-bit executions but this is going to be an area where real life performance could vary wildly from benchmarks.

On the AMD side we can handle 2 128-bit executions simultaneously (one per FMAC) and we can mix AVX and SSE. Our 128-bit FLOPs will be essentially the same as our 256-bit FLOPs, so whether it is new or old code, it should perform the same.

So our real life performance will be a lot closer to the benchmarks.

 

[Artista]

So with all the fanboy crud aside and wild speculation aside is it worth waiting for BD to upgrade vs buying a sandy bridge now?

Basically is BD THE next evolutionary step in CPU technology or just a alleged small improvment?

(I would hate to buy SB now and in two months BD is mopping the floor with a SB.)

 

[dma0991]

So with all the fanboy crud aside and wild speculation aside is it worth waiting for BD to upgrade vs buying a sandy bridge now?

Basically is BD THE next evolutionary step in CPU technology or just a alleged small improvment?

(I would hate to buy SB now and in two months BD is mopping the floor with a SB.)

You could just wait less than 2 months more to decide whether to go BD or SB. If BD does mop the floor then get BD. If it doesn't at least you can get SB at reduced price from the current to compete with BD. A win win situation for both parties. :sneaky:

 

[Vette73]

So with all the fanboy crud aside and wild speculation aside is it worth waiting for BD to upgrade vs buying a sandy bridge now?

Basically is BD THE next evolutionary step in CPU technology or just a alleged small improvment?

(I would hate to buy SB now and in two months BD is mopping the floor with a SB.)

I would wait not only to see what BD does, but also the Z68 chipset should be out from intel as well.

So more options and probable lowwer prices.

 

[chewietobbacca]

AMD = A Motherload of Disappointment

AMD will be a step or ten behind Intel in terms of performance. AMD is budget oriented though. If anything is good about the Bulldozer then, shouldn't there be a glint of hope or leak about their upcoming first lineup and their benches. There is none. So, here's to a disappointing launch for Bulldozer.

AMD also claimed in that "one" supposedly leaked slide that their 8-core proc is 50% faster than Intels quad core of last year. You don't say! 8-core is 50% faster than intels 4-core.

Intel all the way until AMD get's their shenanigans together.

You clearly weren't around in K8's heyday when it stomped Intel all over the place

Are you joking?

How many AMD chipsets/cpus required countless driver updates to get working with windows. I seem to remember.. all of them. Yes AMD eventually gets stable but takes a while.

It almost sounds like you're mixing GPUs w/ CPUs up

And by that logic, because Intel screwed up the chipset for Sandy Bridge, we should worry about their next release right?

 

[Topweasel]

Basically is BD THE next evolutionary step in CPU technology or just a alleged small improvment?

Its a completely new way to build CPU architecture. That much is clear. Even if BD isn't the quickest out of the gate, remember back to the Geforce FX series. Panned and hated for its heat and performance issue, but it was their completely new architecture design that setup almost 3 years of straight dominance in the 7k,8k, and 9k. series.

BD can be the next evolutionary step, one that may need to be made, but that doesn't always mean instant performance pay-off.

 

[Idontcare]

So with all the fanboy crud aside and wild speculation aside is it worth waiting for BD to upgrade vs buying a sandy bridge now?

Without question at this point if you are considering an upgrade to SB and you have the luxury of being able to wait a couple months then I would wait a couple months.

Even if worst-case happens and BD doesn't smash SB it will be sure to cause downward pressure on SB's prices.

Basically is BD THE next evolutionary step in CPU technology or just a alleged small improvment?

(I would hate to buy SB now and in two months BD is mopping the floor with a SB.)

Microarchitecture-wise, bulldozer is a huge change, totally new core from the ground up. It won't just be a small improvement.

That said...the big unknown here is GloFo's 32nm process tech. That determines clockspeeds, yields, and power-consumption. So we have no way of knowing whether GloFo is going to do good by AMD or if they are going to be the achilles heel in the equation.

Unfortunately we do know GloFo went gate-first integration for HKMG which is known (as in "science" known, process technology and all that) to enable denser chips (same design by smaller die size) but at the expense of having lower drive currents (less clockspeed potential) compared to a gate-last integration.

So, while we don't know the specific parametrics of GloFo's 32nm HKMG SOI we do know enough to know that we should not be expecting to be surprised to the upside when it comes to clockspeeds and power-consumption comparisons to Intel's 32nm HKMG.

Nevertheless I remain steadfastly optimistic that the boys and girls in green have really pulled another K7 Athlon out of the hat here and we are going to see some 3.6GHz stock clock 8-core parts (turbo'ing up to 4.2GHz). Crossing my fingers on this one, really want to see it happen.

 

[tijag]

Without question at this point if you are considering an upgrade to SB and you have the luxury of being able to wait a couple months then I would wait a couple months.

Even if worst-case happens and BD doesn't smash SB it will be sure to cause downward pressure on SB's prices.



Microarchitecture-wise, bulldozer is a huge change, totally new core from the ground up. It won't just be a small improvement.

That said...the big unknown here is GloFo's 32nm process tech. That determines clockspeeds, yields, and power-consumption. So we have no way of knowing whether GloFo is going to do good by AMD or if they are going to be the achilles heel in the equation.

Unfortunately we do know GloFo went gate-first integration for HKMG which is known (as in "science" known, process technology and all that) to enable denser chips (same design by smaller die size) but at the expense of having lower drive currents (less clockspeed potential) compared to a gate-last integration.

So, while we don't know the specific parametrics of GloFo's 32nm HKMG SOI we do know enough to know that we should not be expecting to be surprised to the upside when it comes to clockspeeds and power-consumption comparisons to Intel's 32nm HKMG.

Nevertheless I remain steadfastly optimistic that the boys and girls in green have really pulled another K7 Athlon out of the hat here and we are going to see some 3.6GHz stock clock 8-core parts (turbo'ing up to 4.2GHz). Crossing my fingers on this one, really want to see it happen.

Since pretty much 100% of 2500k chips will easily OC to 4.2 ghz standard, for BD to be a 'win' in the consumer space the IPC is going to have to be on par with Sandy Bridge.

If BD's top bin is 3.6ghz with turbo to 4.2ghz, intel can easily compete with that, perhaps with a new chip with higher TDP. Since we all expect Sandy Bridge to maintain a lead in single threaded IPC over BD, they wouldn't even need to hit 4.2ghz. It'd probably be like 3.8ghz.

Honestly, that would disappoint me. For BD to be a 'win' for AMD, IMHO, BD needs to be faster than SB. Ivy bridge will be coming out, with further clock speed gains, and probably some very small IPC tweeks.

Ideally BD would be faster than SB, and then IB would top BD. Then we would expect the next BD, which will probably be released as an APU, to compete with whatever is coming out after IB.

 

[hamunaptra]

I believe BD will be one hell of a cpu when it comes to dealing with all sorts of different code at the same time.
After reading some in depth discussions of its uarch it would seem that benchmarks that only throw one type of code at a cpu might not be so favorable on BD, but real world apps and benchmarks that have many different load / code types would be executed more efficiently.

Also, another thing to keep in mind.
The FPU is not just 2 pipes, there are 2 128bit int MMX / SSE pipes as well, so mixed MMX/SSE w/ int&fpu instructions goin through the flexfpu should theoretically be shear monstrous amazing amount of throughput! =P

And yes, it will be very interesting to see what those 2 AGLU's are capable of doing as far as basic operations that are unkown of at this point.

 

[formulav8]

Lots of good spec info. So I guess all thats left to do is wait.. and wait...

 

[exar333]

Too early to call without real benches. If a somewhat in-depth preview/review doesn't appear before the end of April, either the release date is in jeopardy or the performance is well below expectations. Already I am a little nervous with the lack of benches.

I just looked at AT and Anand did a pretty nice preview of the Nehalem tech 5 months before it launched! BD is no where to be seen and it is LESS than this far away. Silence is usually worrisome.

 

[Idontcare]

I just looked at AT and Anand did a pretty nice preview of the Nehalem tech 5 months before it launched! BD is no where to be seen and it is LESS than this far away. Silence is usually worrisome.

That's a difference between Intel and AMD that exists outside the scope of product maturity.

Look at the history of AMD previews by AnandTech, AMD just manages their shop differently and they don't really let the cat get out of the bag.

Just saying I don't think we can read too much into the absence of bulldozer leaks and previews because that is just an AMD thing in general.

 

[Riek]

Too early to call without real benches. If a somewhat in-depth preview/review doesn't appear before the end of April, either the release date is in jeopardy or the performance is well below expectations. Already I am a little nervous with the lack of benches.

I just looked at AT and Anand did a pretty nice preview of the Nehalem tech 5 months before it launched! BD is no where to be seen and it is LESS than this far away. Silence is usually worrisome.

While i understand your concern, AMD is continiously doing those things. With bad and with good, they keep things a secret as long as possible. Remember the time of the K7, there were some benchmarks in advance, but they crippled them to suprise. We speculated a long time about the 68xx and 69xx series and the only true rumours came days before. A week before everybody had the idea of 1900sp. They turned out pretty damn good. On the other hand we had barcelona failure which most people remember, which was also only disclosed at launch. Either way AMD has a habbit of this while going both ways.

But side remark: the latest AMD products (Brazos and 6950/6970/6990) were all doing what they were expected to do, yet they were all overhyped. So i wouldn't expect Bulldozer to bring us to the future, i do expect it to be competitive in the area it is positioned (super high end).

 

[formulav8]

While i understand your concern, AMD is continiously doing those things. With bad and with good, they keep things a secret as long as possible. Remember the time of the K7, there were some benchmarks in advance, but they crippled them to suprise. We speculated a long time about the 68xx and 69xx series and the only true rumours came days before. A week before everybody had the idea of 1900sp. They turned out pretty damn good. On the other hand we had barcelona failure which most people remember, which was also only disclosed at launch. Either way AMD has a habbit of this while going both ways.

But side remark: the latest AMD products (Brazos and 6950/6970/6990) were all doing what they were expected to do, yet they were all overhyped. So i wouldn't expect Bulldozer to bring us to the future, i do expect it to be competitive in the area it is positioned (super high end).

Actually AMD was quite vocal about Phenom. They not only mentioned alot of the specs well before hand but also performance expectations and such. But the difference then was all of the Core2 talk that AMD is doomed, blah blah stuff. So that may be why they were more vocal with Phenom.

On a side note: Phenom looked quite good on paper, what happened to it in the real world...


Jason

 

[Soleron]

Actually AMD was quite vocal about Phenom. They not only mentioned alot of the specs well before hand but also performance expectations and such. But the difference then was all of the Core2 talk that AMD is doomed, blah blah stuff. So that may be why they were more vocal with Phenom.

On a side note: Phenom looked quite good on paper, what happened to it in the real world...

If Phenom had launched at the likely design clockspeed of 3.0GHz it would have been competitive. Athlon 64 was getting 3.2GHz on 90nm, it should have been possible.

 

[formulav8]

If Phenom had launched at the likely design clockspeed of 3.0GHz it would have been competitive. Athlon 64 was getting 3.2GHz on 90nm, it should have been possible.

I believe they were expecting at least 2.6ghz. They actually sent out many pre-release samples to tech sites with 2.6ghz versions before the l3 cache bug came to light.

 

[bryanW1995]

^this.

Like previous AMDs it will

1. Run hotter
2. have faster FPU
3. Be cheaper than Intel's current offering
4. Cause problems with about 5% of all existing windows software on launch day.
5. Crush in the server market.

I have a p2 x6 and a boatload of intel cpus. The x6 runs cooler and has no windows software issues. I don't recall having any issues with windows on previous amd cpus that I've owned, either.

 

[Idontcare]

Actually AMD was quite vocal about Phenom. They not only mentioned alot of the specs well before hand but also performance expectations and such. But the difference then was all of the Core2 talk that AMD is doomed, blah blah stuff. So that may be why they were more vocal with Phenom.

On a side note: Phenom looked quite good on paper, what happened to it in the real world...


Jason

I won't challenge your recollection of history as I freely admit we are pushing up against the hairy edge of the limits of my trustable memory (phenom was what, 2007? 2008?) but the mental "note to self" I have filed away in the back of my mind was that the "hype" by AMD was limited to that of Barcelona and the server aspects of the chip.

IIRC, which is a long-shot in this case, Phenom the desktop consumer processor was not hyped by AMD, they were quite mum about it.

Charlie D (at theINQ then) was their cheerleader at the time with his famous "dancing in the aisles" articles that kept hyping the rumors and expectations of Phenom every other week.

The skant few concerns I have regarding Bulldozer are all related to GloFo process tech concerns based on a few comments made to me by people in the know over the year. I don't put too much concern in those comments because even if they are/were dead accurate at the time they were merely static snapshots of a very dynamic and fast-paced process of iterative improvements that are always going on.

As a process tech aficionado I just want to own a CPU made on 32nm HKMG w/SOI. Beyond the computing experience provided by the product there is a deeper connection I would have to such a chip, and I'm waiting on pins and needles to satisfy that nerdy geeky need. I feel the same way about FinFET technology and am patiently waiting for it to come into production as well.