SiSoft Leaks AMD FX "Zambezi" Scores: Worse Than Intel Core i7

[Natfly]

I won't address particular points,but you are wrong. Each bulldozer core is a real core in every sense of the word.

Absolutely not, a Bulldozer "module" is a core. It is a core that implements CMT and duplicates some resources to more efficiently run two hardware threads. But it still is one core, go look up the definition of a processor core (hint: it includes reading/decoding instructions) and even AMD has admitted this saying a module has the performance of 1.8x actual cores. The fact that its core has two int clusters and supports two threads in hardware doesn't make it a dual core, go read some papers on CMT and even then a "module" is called a core. I'd even wager that AMD's engineering team called them cores before the marketing people injected the word "module."

 

[nonameo]

Absolutely not, a Bulldozer "module" is a core. It is a core that implements CMT and duplicates some resources to more efficiently run two hardware threads. But it still is one core, go look up the definition of a processor core (hint: it includes reading/decoding instructions) and even AMD has admitted this saying a module has the performance of 1.8x actual cores. The fact that its core has two int clusters and supports two threads in hardware doesn't make it a dual core, go read some papers on CMT and even then a "module" is called a core. I'd even wager that AMD's engineering team called them cores before the marketing people injected the word "module."

The question is, does a module perform more like one core or two? We'll find out when bulldozer is released. After all, what we care about is performance, not semantics.

 

[inf64]

Module is called an optimized dual core,since that's what it is. It contains 2 tightly linked cores that share some stuff and has some stuff dedicated. They share latency tolerant parts and don't share latency sensitive parts(like integer cores). Each of the 8 threads in 8C bulldozer processor will have a piece of dedicated HW that it runs on. In case of integer workloads it's an (integer) core,in case of fully multithreaded FP workloads it's one 128bit FMAC/one integer SIMD pipeline within FlexFP co-processor..

 

[AtenRa]

Absolutely not, a Bulldozer "module" is a core. It is a core that implements CMT and duplicates some resources to more efficiently run two hardware threads. But it still is one core, go look up the definition of a processor core (hint: it includes reading/decoding instructions) and even AMD has admitted this saying a module has the performance of 1.8x actual cores. The fact that its core has two int clusters and supports two threads in hardware doesn't make it a dual core, go read some papers on CMT and even then a "module" is called a core. I'd even wager that AMD's engineering team called them cores before the marketing people injected the word "module."

No words, just pics

http://www.anandtech.com/show/3865/amd-bobcat-bulldozer-hot-chips-presentations-online

 

[BlueBlazer]

No words, just pics

Another quote, this time from AIDA64 author here.....

The problem comes if you want to prevaricate in the AMD thing, the Bulldozer (ie, specifically the Zambezi) architecture is also much closer all in a four-core Phenom II X4 to like a 6 or 8-core Sandy Bridge-E . The Sandy Bridge, this is me'g supports HyperThreading, which is able to forward additional advantage in many situations.

But if the Zambezi 8 core processor would be true, it would not be much of a chance against The Sandy Bridge.

And not by accident is not the Sandy Bridge (in the current, four-core version, LGA1155) is known, but the upcoming 6 to 8-core generation. Bar, record prices in the Zambezi illegal activities will be finally positioned ellenfeleul LGA1155 AMD, Sandy Bridge, this remains a lack of competition

 

[lau808]

um... can someone translate that to proper english? lol

 

[Tsavo]

um... can someone translate that to proper english? lol

Everyone is stepping on the spins, while we own the Bulldozer architecture tour of the problem. In stepping in does not help. Orajellel can be compensated, but there is marred by dissipation. Many seeds can be (not) be compensated, but again, the dissipation kozbeszol. The B0 and B2 and was not nearly as much change happened, as many expect. It is now accept that the Bulldozer ratyi and then recalled the October surprise will be less

 

[BlueBlazer]

um... can someone translate that to proper english? lol

I think its pretty "readable" (even though the translation is bad) Breakdown....

- Bulldozer architecture (specifically Zambezi) is closer to 4 core Phenom II, than to 6 or 8-core Sandy Bridge
- Even if Zambezi has 8 real cores, still has no chance against Sandy Bridge
- Besides Sandy Bridge 4 cores, there will be upcoming 6 and 8-core generation
- Zambezi pricing will be positioned against LGA1155 processors
- Lack of competition for Sandy bridge

Correct me if I'm wrong.

 

[3DVagabond]

um... can someone translate that to proper english? lol

"Darmok and Jalad at Tanagra".

 

[bronxzv]

Absolutely not, a Bulldozer "module" is a core.

AFAIK a module (marketing lingo) was called a core (CPU architect lingo) in the patents and an integer core (marketing lingo) was called an integer cluster (CPU architect lingo), also I remember seing (a long long time ago) a comment by Andy Glew on comp.arch saying how bad the new marketing nomenclature is from his POV. The net effect will be that Bulldozer cores will look weak (more cores needed for a given performance) much like the Pentium 4 IPC was looking weak (more GHz required for a given performance) when Willamette launched.

Especially when comparing with SNB it's definitely more sensible to compare a module (aka "MCMT core") to an hyperthreaded core (aka "SMT core"). For example the shared L2$ and hardware prefetcher will have similar impact on thread afinity optimizations.

Moreover a Bulldozer module (aka "2 cores") will be not even on par with a single SNB core for some codes, the worst case for Bulldozer being AVX-256 code (not using FMA4) where a module will have less throughput than a single SNB core (assuming well balanced vmulp*/vaddp* code)

 

[beginner99]

I also think it's astonishing that a company with a market cap of less than $5B can stay within one or two generations of a company with a market cap of over $100B.

I agree. Still it's an unfair comparison because intel does much more than just CPU and GPUs. So IMHO you can't really compare these 2 figures 1 to 1.
Meaning the actually money following into cpu creation is probably not 20x times higher than at AMD.

 

[Black96ws6]

I'm still holding out hope, however another thing people seem to be forgetting is JF-AMD's own comment on this fiasco, which is:

Originally Posted by JF-AMD
I said that the benchmarks are "not representative", not inaccurate.

Am I the only one that caught this? The last part of that sentence should scare the crap out of both AMD and Intel fans.

Because how much can be fixed with stepping changes, unless there were major bugs found and fixed?

If AMD doesn't deliver, our hobby will suffer.

 

[nonameo]

I'm still holding out hope, however another thing people seem to be forgetting is JF-AMD's own comment on this fiasco, which is:



Am I the only one that caught this? The last part of that sentence should scare the crap out of both AMD and Intel fans.

Because how much can be fixed with stepping changes, unless there were major bugs found and fixed?

If AMD doesn't deliver, our hobby will suffer.

Yes, I have been wondering as well how much can change in a bios/driver release or a stepping change. I mean, look back to AM2 or phenom II stepping changes. How much changed from C2 to C3? or from C3 to thuban? I see drivers changing things more just because of the way GPUs react to driver changes, but that market moves a LOT faster than the CPU market, and it looks like bulldozer needs more gain than that anyway. Really, I just don't see how more than minor changes in performance can be expected(without increases in clockspeed)


But hey, here's to northbridge overclocking! perhaps it will be more relevant than ever.

 

[drizek]

Not inaccurate doesn't necessarily mean accurate.

Also, if bulldozer didn't suck, it would have have ben out 3 months ago. there is obviously something wrong with it.

 

[radaja]

"Darmok and Jalad at Tanagra".

OMG,these last few months i often felt like Picard trying to read these damn google translated pages.this made me laugh hard!!!

 

[GaiaHunter]

Not inaccurate doesn't necessarily mean accurate.

Also, if bulldozer didn't suck, it would have have ben out 3 months ago. there is obviously something wrong with it.

Something wrong doesn't mean the architecture or even the specific processor sucks.

Clearly there are problems, possible various, from GF 32nm manufacturing problems to bugs in the architecture and/or platform.

Look at the example of the X1800XT - it wasn't a bad performing card per se, just very late due to problems. The refresh of it, the X1900XT, on the other hand was great.

On the other hand it is possible it is more like a HD2900. Still that crappy architecture was the foundation of the very good HD4000 and HD5000 series.

 

[LOL_Wut_Axel]

I also find it amazing that after Athlon XP+, Athlon 64, Athlon X2/FX, AMD twirled thumbs and barely improved IPC over A64 in Phenom I/II. Basically, they underestimated Intel by pretty much rolling over Athlon 64's architecture into Phenom II. When AMD took risks and was focused on innovation, it was not far behind Intel (and sometimes ahead). The intellectual capital (of engineers) was there.

I can only blame poor management of Ruiz and an insane premium that AMD paid for ATI in the $5.4B acquisition......Remember that they purchased ATI using a leveraged buyout, and the interest on the debt killed any positive cash flow AMD had to reinvest into R&D. So essentially, the buyout of ATI took the wind out of AMD's sails, and 5 years later it seems it still has not recovered. The fact that GPU profits/revenues started to shrink in recent years as users shifted to gaming on consoles/tablets/smartphones isn't helping AMD's graphics division either.

Fusion... that's the only big reason why AMD acquired ATI. As you can see, it's definitely paying off now, even if it's five years later. AMD has sold around 12 million APUs in less than a year.

As far as architectures go, you're definitely right. AMD is still using a 2003 architecture in 2011. Given that, if Bulldozer has made no improvements on it, it'll suck from day one.

 

[Zstream]

..

 

[nonameo]

I heard from a little birdy that this thing is very good in sql performance. One's that have to do many calculations within the search. I'm guessing major players like facebook can just pop this cpu in or am I wrong?

Only if they already have amd though.

 

[RussianSensation]

Fusion... that's the only big reason why AMD acquired ATI. As you can see, it's definitely paying off now, even if it's five years later. AMD has sold around 12 million APUs in less than a year.

Ok, but why buy ATI at the peak for $5.4B?? I believe that was > 20% price premium. Good management would have assessed what was more important in the short term. They were well aware of how uncompetitive Phenom I was going to be before they released it. So they knew they'd lose market share in the next 2-3 years in all 3 key markets, making the ATI acquisition extremely costly from a cash flow perspective. It would have been better to invest more into a new CPU architecture, make more cash flow from it, and buy ATI much later. AMD really wasn't in a position to afford ATI buyout easily at the time, but that's exactly what they decided to do.....esp. in a market with relatively high borrowing interest rates at the time.

There was no rush to buy ATI at the time (more point on this below). It took 5 years of technological innovation since 2006 (and by that I mean manufacturing node to fall to 32nm) and ATI's own architectural changes in Cypress/Barts, etc. to maximize performance/watt before Fusion could be possible. In other words, AMD could have delayed the purchase of ATI into late 2007/2008 and still got Fusion out the door in 2011, while making $$ from their 20-25% server market share at the time (had Phenom I/II were more successful if they invested more $ rather than rolling over A64 tech).

Excerpt:

http://gear.ign.com/articles/720/720756p1.html

Key points:

1) Extremely poor timing of buyout on the part of AMD IF they actually did a lot of planning and due diligence prior to buyout considering the stock was trading <$13 just 1 year before buyout:

"ATI stock jumped significantly in early trading based upon the news. The company's stock, which went as low as $12.40 per share in the last year, is now trading at $22.37, up nearly 20% on the day."

2) No one wanted to buy ATI (and esp. not for $5.4-5.8B), so why the rush and the huge 24% price premium?

"AMD investors worry that the ATI deal may have cost the company too dearly. No other companies bid on ATI".

3) Unreasonable multiple for buyout on a going-concern basis:

"ATI's worth as a company on the stock exchange is equivalent to 28 times its projected earnings next year, while AMD's ratio is only 15 times. Companies in the Standard & Poor's 500 trade at 14 times--a benchmark for more conservative investors. As such, ATI is considered a risky investment for AMD."

4) Despite point #4 with 28x PE ratio, Intel immediately reduced ATI's revenue after buyout announcement. This means the buyout price was already inflated after that:

"Intel responded to the news today by terminating ATI's (now essentially AMD's) license to produce chipsets for Intel CPUs."

5) Buyout occurred during a volatile/risky time for AMD's CPU business as they were only trying to transition to a far inferior Phenom I/II (and AMD's management for sure knew Phenom I was going to flop since Core 2 Duo were available and it was really easy to gauge that a C2Q would mop a Phenom I since C2D crushed X2 processors.....so they knew!):

"Investors are also worrying about the ongoing battle between AMD and Intel in general CPU sales. Both companies are producing more chips than are currently in demand and are locked in a struggle to continuously outdo the other in chip cost and performance. Both companies are racing to release the next generation of quad-core CPUs and are relying heavily upon strong sales in the fourth-quarter to make up for a slow summer."

If you look at AMD stock now as a whole it's worth 4.51B in market cap, or less than what AMD paid for ATI just 5 years ago.

 

[Arkaign]

um... can someone translate that to proper english? lol

The problem comes if you want tergiversar in the thing d' AMD, l' structure of what east will niveladora (IE, specifically Zambezi) is also much nearer all in a phenomenon II X4 to the four-heart to have taste of bridge an E of Sandy of 6 or 8 bases. The bridge of Sandy, this one is me' ; G supports HyperThreading, which can give additional advantage in many situations. But if the processor of the base of Zambezi 8 were truth, it would not be much d' an occasion against the bridge of Sandy. AND this n' is not accidentally Sandy who is known the bridge (in the current, the version of the four-heart, LGA1155), only the generation close to 6 to 8 bases. The bar, prices of register in Zambezi which the illegal activities will be l' ellenfeleul finally placed LGA1155 AMD, bridge of Sandy, this one is still a lack of the competition But yew the Zambezi the 8 hearts would processor ve true, it would not ve much off has chance against The Sandy Bridge. And not by accident is not the Sandy Bridge (in the current, furnace-core version, LGA1155) I known, goal the upcoming 6 to 8-hearts generation. Bar, recording in the prices illegal Zambezi activities will ve finally positioned ellenfeleul LGA1155 AMD, Sandy Bridge, this remains has lack off competition But yew the Zambezi 8 core processor would Be true, it would not Be much off has chance against The Sandy Bridge. And not by accident is not the Sandy Bridge (in the current, furnace-core version, LGA1155) is known, goal the upcoming 6 to 8-core generation. Bar, record prices in the illegal Zambezi activities will Be finally positioned ellenfeleul LGA1155 AMD, Sandy Bridge, this remains has lack off competition

 

[Gigantopithecus]

I heard from a little birdy that this thing is very good in sql performance. One's that have to do many calculations within the search. I'm guessing major players like facebook can just pop this cpu in or am I wrong?

I think with this comment you're hitting on what I see happening to CPUs long-term: increased niche specialization. There are sufficiently large markets for SQL specialists as well as other widely-utilized enterprise software that it makes sense to design chips to run one kind of software especially well. General usage CPUs have reached the point where $50 chips are sufficient for 95% of consumers, so it seems to me that fat profit margins really only exist at the enterprise level. Just a thought.

 

[Idontcare]

4) Despite point #4 with 28x PE ratio, Intel immediately reduced ATI's revenue after buyout announcement. This means the buyout price was already inflated after that:

"Intel responded to the news today by terminating ATI's (now essentially AMD's) license to produce chipsets for Intel CPUs."

I had forgotten all about this.

It was definitely a cherry deal for ATI's shareholders.

Anytime you manage to sell high, that is going to mean someone else is buying high.

AMD's M&A activity though is questionable, they seem to find themselves buying high and selling low.

Qualcomm laughed all the way to the bank on AMD's dime as well.

The sad part is that so did Hector.

 

[lol123]

I think with this comment you're hitting on what I see happening to CPUs long-term: increased niche specialization. There are sufficiently large markets for SQL specialists as well as other widely-utilized enterprise software that it makes sense to design chips to run one kind of software especially well. General usage CPUs have reached the point where $50 chips are sufficient for 95&#37; of consumers, so it seems to me that fat profit margins really only exist at the enterprise level. Just a thought.

That might make sense for the smaller players like AMD and Nvidia who can't compete anymore (in the case of AMD, Nvidia of course never have) across the whole board of applications, but Intel and IBM still make decidedly general purpose processors and reap by far the largest profits. And if you look at Intel you actually have the opposite trend going on with the enterprise-oriented Itanium processors slowly being phased out in favor of x86 across the entire spectrum.

 

[sequoia464]

Why do people buy the motherboard before the CPU performance is definitively known?

Did you get duped by the hype the AMDroids are always furiously spinning?

Let's just say I chose poorly. Had a 2500K and a Z68 board that I returned. I'll ride this out for another month or two now to see what happens.

Given a choice I will allways buy AMD though. Corporate ethics thing with me, I'm the guy that will drive across town to avoid going into a Wal-Mart.

Having said that, If AMD doesn't deliver I will toss this brand new shiny Asus Sabertooth and get an Intel setup. Hate the UEFI bios anyway.