AMD to launch desktop Bulldozer "Zambezi" at E3 Show in L.A. (June 7-9)

[[XC] hipno650]

Since we have no real Zambezi benchmark number it's unknown how it will perform.It may be a total flop for all we know . But AMD knew a 1+ year how i7 Gulftown performs and they still call Zambezi their high performance part aimed at enthusiasts .They also revived FX branding.This at least suggests the performance should be close to what Gulftown offers.It may not be able to beat it completely but as I said before AMD just needs to be competitive. Intel will have 3.46Ghz Gulftown when Zambezi launches so it will be a hard task to equal this monster,let alone beat it.

I would think the same thing....... BUT I do remember the much hyped (lots of hype from AMD as well) Phenom line of CPU's which turned out to be absolute trash..... also from what i have seen so far in the naming Scheme the FX line of CPU's will pretty much just replace the Phenom 2 name in an attempt to build off the past beastly name.

I think AMD will make some HUGE mutlithreading per improvements and the integrated GPU will be a huge plus over Intel but I can't possibly see how they will come close to equalling or beating Intel in low threaded perf. like say 1-4 cores of workload which represents the majority of even enthusiast load. I have yet to see a single game use more then 3 cores PROPERLY. most conversion prog or photo/music editing progs won't use more than 4 cores... other average use again under 4 cores. In which case AMD might have a tough time beating SB....

they will be very competitive in rendering and DC type computing (which is very important to me BTW) so that will be interesting.

What it might come down to for us will be Overclockablity of the 8 core parts. SB OC's so freaking well it's crazy..... if you don't get 4.8ghz out of a 2600k your doing something terribly wrong and 5GHZ seems really common.... can AMD make up for that much of a clock speed disadvantage????? also considering they have poor so much effort into the on DIE GPU you really need to wonder how many other improvements can be made on the same Die size and TDP....

only time will tell. I really hope the June launch is true.... rumours of it being pushed back are just.... bad again reminds me of the Phemon launch.... but im hopeful

[-Boris-]

wow so much nonsense in your latest posts. u should read your own posts again and u might see what i point at You should also read more on bulldozer design if you like the concept.

Well, point out what's incorrect. And I know quite much about BD's design. I would say most of what's know publically.

I don't accept avoiding answers like "I'll let you figure that out yourself." Point something out.

I would think the same thing....... BUT I do remember the much hyped (lots of hype from AMD as well) Phenom line of CPU's which turned out to be absolute trash..... also from what i have seen so far in the naming Scheme the FX line of CPU's will pretty much just replace the Phenom 2 name in an attempt to build off the past beastly name.

I think AMD will make some HUGE mutlithreading per improvements and the integrated GPU will be a huge plus over Intel but I can't possibly see how they will come close to equalling or beating Intel in low threaded perf. like say 1-4 cores of workload which represents the majority of even enthusiast load. I have yet to see a single game use more then 3 cores PROPERLY. most conversion prog or photo/music editing progs won't use more than 4 cores... other average use again under 4 cores. In which case AMD might have a tough time beating SB....

One of the best things with BD is that not only is the frequency increased with fewer threads, but IPC per thread is increased as well.
If you have one hexacore i7 and one hypothetical hexacore BD with equal performance at 6 threads, I believe that BD will have quite a performance advantage if both are running only one thread. All the i7 gets is more cache and bandwidth to that thread and higher frequency, while BD also gets higher IPC.

[marten_larsson]

Hipno, I hope you realize that BD doesn't come with integrated graphics, not untul 2012 at least.

[Hornet331]

You know, I've always wondered why I've been so angry with AMD since buying their stock many years ago. Turns out that all this time, they could have charged more for their processors - but they didn't! What a bunch of idiots. Instead of keeping their margins artificially low, AMD could've been paying out a big fat dividend to keep me happy...right?


So...your position is that AMD should price its chips the same as intel's chips of equivalent performance? That's your idea of a smart business move for AMD?!? Really?!?!

Lol wth your talking, I think you forget the P4 vs Athlon time (2004-early 2006). The slowest dualcore was 360$ (x2-3800) and they charged what it was worth, and suprisingly they made there best profits ever.

No read again what I wrote, if the product has the same performance it should price it a bit lower, but if it has better performance it should be price at least equal or higher depending on how much better performance it offers.

I like how many still think amd is a charity and offers cpu so cheap because they are nice....

[Kristers Bensin]

Just one thing. 40% better performance is not the same thing as 40% better bench results. A 5GHz processor don't get twice as good results as a 2.5GHz processor. So if 980X has 40% better results than Phenom X6, it's probably has much higher performance advantage.
BD needs much more than 50% better performance to compete with 990X.

People will once again accuse a company of lying about performance just because thay don't realize these things. I don't say you don't understand this, but there are many others out there.

"40% better performance is not the same thing as 40% better bench results"... WTH do u mean by that?

"So if 980X has 40% better results than Phenom X6, it's probably has much higher performance advantage." Huh? Maybe my english is bad, but i dont understand what u mean by that.

BD will still need more than 50% higher performance than Phenom II to compete with 980X. Of course there are differences between loads. But if 980X score 50% higher in benches than Phenom II, then BD needs maybe 70%+ higher performance to get these 50% higher results.

A 4GHz chip has twice the performance of a 2GHz chip, but might only have 40% higher FPS. You can't just look at benches and think that they accurately show the differences in performance for a single part of the system.

The difference between Phenom II X6 1100T and 990X is what i belive about 35-45% in general benching. So if they can get 30-50% better performance with Zambezi, they can compete at some level. Remember Zambezi have 2 more cores than 990X.

Irrelevant.

When AMD says 50% higher performance they are talking about total throughput, that's what they have said all the time. And I doubt that they mean some average benches. If I know AMD right they are talking about numbers pretty close to best case scenario. And a rumored benches is not something we should base performance estimates on.

I bet that AMD will have a very nice processor and I will by it, but I dont think it will compete with the best six core i7s. Especially not the SB ones.

No, they said at some interview that Zambezi will have about 30-50% better performance over Phenom II X6, not throughput. What they also said is that a Zambezi module will give 180% throughput of a real dual core, wich has about 195% throughput if it scales accordingly.

The most interresting part is how well the new Fetch and Decode part performes as they will work with 2 cores at the same time.

[[XC] hipno650]

Well, point out what's incorrect. And I know quite much about BD's design. I would say most of what's know publically.

I don't accept avoiding answers like "I'll let you figure that out yourself." Point something out.



One of the best things with BD is that not only is the frequency increased with fewer threads, but IPC per thread is increased as well.
If you have one hexacore i7 and one hypothetical hexacore BD with equal performance at 6 threads, I believe that BD will have quite a performance advantage if both are running only one thread. All the i7 gets is more cache and bandwidth to that thread and higher frequency, while BD also gets higher IPC.

I understand that they will ahve turbo core which IMO is total garbage for OCer's cuz it gets turned off anyways (just like turbo on i7's) and of course they will be increasing the IPC per thread but I have a hard time seeing believing that they will increase it enough to match SB especially if they only OC to 4-4.2ghz (a rough guess it really could be anything). it's allot of ground to make up.... only time will tell if they can do it.

Hipno, I hope you realize that BD doesn't come with integrated graphics, not untul 2012 at least.

well now I know thats a major bonus IMO for the high end desktop market

[Kristers Bensin]

I understand that they will ahve turbo core which IMO is total garbage for OCer's cuz it gets turned off anyways (just like turbo on i7's) and of course they will be increasing the IPC per thread but I have a hard time seeing believing that they will increase it enough to match SB especially if they only OC to 4-4.2ghz (a rough guess it really could be anything). it's allot of ground to make up.... only time will tell if they can do it

Bulldozer will have a major upside in single-threaded workload, the shared L2-cache. 1MB per core and hopefully 2MB in lowthread workloads wich is 4 threads and lower for the FX8.
Core i7 2600K has 256KB L2 per core and 8MB shared L3 vs Zambezi's 8MB (module) shared L2 and 8MB (module/all cores) shared L3.

[[XC] hipno650]

Bulldozer will have a major upside in single-threaded workload, the shared L2-cache. 1MB per core and hopefully 2MB in lowthread workloads wich is 4 threads and lower for the FX8.
Core i7 2600K has 256KB L2 per core and 8MB shared L3 vs Zambezi's 8MB (module) shared L2 and 8MB (module/all cores) shared L3.

core 2 quads had up to 12MB of L2 Cache.... are they faster then a SB CPU single threaded? not even close.... more cache is always nice but won't make or break a CPU in any performance. also the 1100T has 512kb of L2 cache that double the i7's... and is it faster single threaded..... nope.

also we have no idea as to the cache speeds of bulldozer as well. the caches on Core i7 especially SB are SMOKING fast compared to anything else.

cache size can tell us next to nothing about any type of performance without seeing any other numbers. if we knew what bulldozer was capable of with a certain amount of cache we could make some guesses as to the increase in performance adding more cache would have but until then....

I do really like the whole "module" thinking as opposed to cores though as it seems that the sharing of resources more efficient. but how much more efficient?

[madcho]

we just can be sure of that ;

-cache size is larger than Phenom II, so it mean more perf
-L2 cache system is new, so more performance, AMD is not stupid to do R&D to a new system decreasing performance.
-L1D and L1I are far very different from old Phenom II. And i think it's an amazing good idea. We have just to wait to see it.

BD can't fail, or AMD will definetly fail. They don't have fabs anymore.

[Kristers Bensin]

core 2 quads had up to 12MB of L2 Cache.... are they faster then a SB CPU single threaded? not even close.... more cache is always nice but won't make or break a CPU in any performance. also the 1100T has 512kb of L2 cache that double the i7's... and is it faster single threaded..... nope.

also we have no idea as to the cache speeds of bulldozer as well. the caches on Core i7 especially SB are SMOKING fast compared to anything else.

cache size can tell us next to nothing about any type of performance without seeing any other numbers. if we knew what bulldozer was capable of with a certain amount of cache we could make some guesses as to the increase in performance adding more cache would have but until then....

I do really like the whole "module" thinking as opposed to cores though as it seems that the sharing of resources more efficient. but how much more efficient?

wow u must be a genius, looks like no1 thought of that before... of course the 12MB of l2 cache for the core2quad Q9550 is slower than todays memory, and then u have the architecture. That lecturing was so not necessary

What we do know is the memory in Zambezi will be something special, not just that its very high density stuff but also very fast. As of what i read earlier the L2 cache will be working in the same speed as each module (3.5GHz?) and also induvidualy together with its designated module. Each module will have its own L3 cache but also be shared between modules, wich inceases the speed and lower latency. The 1.1V L3 cache will be working at 2.4+GHZ with induvidual frequency depending on load. I dont think Zambezi will have no trouble competing in pure memory speed with SB. But as u wrote,

The power gating will very much interesting to hear about with all these states.

[Hornet331]

L2 cache always worked @ core frequency on pretty much every cpu since the Pentium 3, zacate is a unicum in that regarde, but i bet next iteration sees again L2 @core clock.

The L2 cache in C2 was actualy very fast for its size, the lantency was only 12 cycles with a size of 6mb... compare that to 10 cycles for nehalem and 12 cycles for sandy bridge with 256kb. K10 wasn't that much slower it had 12 cycles l2 latency. Amds architecture only had a weakness when it had to go outside the L2 cache, there L3 was always slower (cycle wise and also frequency wise) and a TLB misses hit them far harder then intel.

Comparing frequencies or size means practically nothing when you compare the cache sub system.

Cause you can compensate with one for the other, but as hipno said the cache sub system in SB is very fast, since the L3 runs @ core speed its already faster then that in nehalem + they reduced the latency even further (from 47-57 cycles form nehalem to 36-47 cycles in sandybridge).

I rely on the 7-zip benchmekr for the latency numbers.

Then againthe whole discussion about the cache system is useless without factoring in the prediction mechanisms. Cause if you have a higher hitrate you don't need to go outside the cpu and fetch the data, since this is all unknown for the moment you can basically say nothing at all. But only belive in the marketing slides.

[SMTB1963]

Lol wth your talking, I think you forget the P4 vs Athlon time (2004-early 2006). The slowest dualcore was 360$ (x2-3800) and they charged what it was worth, and suprisingly they made there best profits ever.

IIRC, the Athlon 64 X2 3800 was introduced in late 2005, at the peak of AMD's profits during that timeframe. Almost immediately after that, AMD's profits began to decline year over year. But my memory isn't what it used to be, so maybe I'm mistaken.

No read again what I wrote, if the product has the same performance it should price it a bit lower, but if it has better performance it should be price at least equal or higher depending on how much better performance it offers.

I like how many still think amd is a charity and offers cpu so cheap because they are nice....

OK, I re-read your posts. Here's some of what you wrote: "Price is based on performance" and "I can see the 8core BD somewhere around the 500€ mark, cause thats where intels hexacores start." You seem to be arguing that since Bulldozer is speculated to perform similarly to an intel hex, Bulldozer's launch price should be close to today's intel hex prices. You further claim that if Bulldozer doesn't launch at those prices, AMD will be "artificially" cutting margins (whatever that means), missing a huge profit opportunity, and screwing it's shareholders. Did I get that right?

Well, I guess one person's speculation about Bulldozer's initial pricing is as good as another's these days. But there's a helluva lot more to pricing than just performance, especially at a product's introduction. You can try to dismiss "Economics 101" all you like, but pricing a new CPU - or any new product for that matter - depends on many different factors and this case is no exception.

By your logic (AMD intros @ 500€), intel may as well say "...oh look...AMD priced their new chip about the same as our comparably performing hexas...that means we can keep our prices the same, because production volumes/market share don't really mean anything for our profitability!"

As far as opinions go, mine is that if this CPU is introduced at 500€ (~$700!), it won't be good for AMD's business - especially in the face of their declining market share. Another consideration is that intel very likely has room to drop prices on their hexes and still sell them at a profit when Bulldozer launches, something I'm sure is not lost on the folks at AMD. If intel sees Bulldozer as a threat, they will almost certainly lower the prices of their competing chips, and it won't be because they're feeling charitable towards their CPU buyers.

Hey, I would love to see a return to the glory days of AMD64 when AMD could command higher prices. Benefits of increased competition and whatnot. But info on Bulldozer's capabilities will require some time to percolate through the marketplace before buyer's perceptions change. If it turns out that Bulldozer's performance is in fact roughly equivalent to intel's hexes, then yes, AMD could eventually raise prices to match intel...but that ain't gonna happen at launch.

I'm starting to think that Particle's rules are coming into play here

[-Boris-]

"40% better performance is not the same thing as 40% better bench results"... WTH do u mean by that?

"So if 980X has 40% better results than Phenom X6, it's probably has much higher performance advantage." Huh? Maybe my english is bad, but i dont understand what u mean by that.

If you have 2 processors, one scores 30% higher than the other in over all benches. Now turn up the frequency on the other one 30%. Will the 30% higher performance from 30% higher clocks make these two processors equal? I'd say most likely not. It might still be so much as 10-20% difference.
That's why a 4GHz chip wont get twice the scores of a 2GHz one.

The most interresting part is how well the new Fetch and Decode part performes as they will work with 2 cores at the same time.

I can certainly agree on that.

[[XC] hipno650]

wow u must be a genius, looks like no1 thought of that before... of course the 12MB of l2 cache for the core2quad Q9550 is slower than todays memory, and then u have the architecture. That lecturing was so not necessary

What we do know is the memory in Zambezi will be something special, not just that its very high density stuff but also very fast. As of what i read earlier the L2 cache will be working in the same speed as each module (3.5GHz?) and also induvidualy together with its designated module. Each module will have its own L3 cache but also be shared between modules, wich inceases the speed and lower latency. The 1.1V L3 cache will be working at 2.4+GHZ with induvidual frequency depending on load. I dont think Zambezi will have no trouble competing in pure memory speed with SB. But as u wrote,

The power gating will very much interesting to hear about with all these states.

you post was structured in a way that used the thinking that More Cache=more performance.... which is totally true when comparing the same architecture and CPU line up. but means almost nothing when comparing different architectures.....

like hornet point out the L3 cache system in SB operates at the SB core speed meaning it is significantly fast then any other L3 cache system... which helps to negate the lack of L2 cache size. also the idea of shared cache is nothing new. Core 2 Duo forward have had it... (im pretty sure there was some CPU's that it before that as well)

make no mistake I know AMD will be making improvements all the way around in terms of performance. but I'm not convinced that they will be able to match or beat SB in everything on the first try.... we have not seen that significant performance jump since Intel released Conroe and demolished every P4 and Athlon around.

I would love to see that happen but I'm not getting my hopes up....

[XRL8]

...
That's why a 4GHz chip wont get twice the scores of a 2GHz one...

Um, 4ghz chip is always almost better than 2x2ghz ones.
And yes, it can be more than 2x as fast even.Depends on the work hes doing.FSB speeds, IMC speed,memory management, threading etc etc.
All this talk however is too mystical.40% more raw horsepower may or may not get us 40% higher bench score.That depends.
However people should understand that statement is most probably some huge simplification, as on say "on average".Im pretty sure BD will get often MORE than 40% ,and often much less.
You could say that 2500k has 40% more performance than a core2quad.But again, that depends on the work cpu is doing.
Simply taking into account that it has 2 more cores, gives you around 30% more computational power.Most probably higher speed, better turbo etc.And better IPC.And new instructions.
AMD would have to actually TRY to NOT get a sizably faster processor with all these changes.
Lets remember that Phenom II is OLD.They had lotsa time architecture wise.

[Hornet331]

*snip*

Again your just pulling out fragments... I never said AMD should introduce all BD @ 500€ (an jfyi, 500€ here means basically 500$, because retailers go 1:1 for $:€ and including vat).

You even quoted it that I mentioned 8c bulldozer... there are still 6c core bd and 4c bd...

This is a nice portfolio that leaves enough SKUs open to be filled. If the 6c BD are a good match for the 4core SB the can saturate the 400-250€ price segemtn with it and the 4 cores thake the lower 250-1xx€ segment.

But yeah I guess it makes much more sense to sell your top of the line chip that offers better performance then the product of your competitor for the same price and just squeeze all the other SKUs into a tiny price range...

[Dresdenboy]

...
That's why a 4GHz chip wont get twice the scores of a 2GHz one...

Um, 4ghz chip is always almost better than 2x2ghz ones.
And yes, it can be more than 2x as fast even.Depends on the work hes doing.FSB speeds, IMC speed,memory management, threading etc etc.

This is even more difficult to answer since some components, which affect performance, scale even less than the ones you list. How about memory latency/bandwidth?

[SMTB1963]

...if the product has the same performance it should price it a bit lower, but if it has better performance it should be price at least equal or higher...

then

But yeah I guess it makes much more sense to sell your top of the line chip that offers better performance then the product of your competitor for the same price and just squeeze all the other SKUs into a tiny price range...

First you say it's OK for AMD to sell BD at the same price if it has better performance....then you say it doesn't make sense for AMD to sell BD at the same price if it has better performance?



Nice. So much for the possibility of engaging in a rational debate with you on Bulldozer launch pricing.

[XRL8]

He was being sarcastic.
On this one, im in between, with this line of thinking, 2600K should be priced higher, but intel changed its tactics on these ones.
Maybe in anticipation of AMD`s move.
But im pretty sure, we wont get highest bin model at or below 2600K price.

[Hornet331]

wow more selective quoting.. plus do even know what at least means?

If you try to troll at least make your selective quots a bit better... you forgot to remove the "or higher" (depending on how much better performance it offers)...

Anyway to only thing i can agree is that its not possible to have a rational debate about launch prices, since someone is so fixated on a idea (for whatever reason) that AMD can't charge more then 300€ for its topend cpu line...

[geo]

wait Zambezi has integrated gpu? google points to this thread.

[DarthShader]

Nope. Next generation will.

[SMTB1963]

Here's another selective quote for ya:

Anyway to only thing i can agree is that its not possible to have a rational debate about launch prices, since someone is so fixated on a idea (for whatever reason) that AMD can't charge more then 300€ for its topend cpu line...

If you can show where I said anything like that, I'll send you one of these:



I promise!

[ubuntu83]

Here's another selective quote for ya:



If you can show where I said anything like that, I'll send you one of these:



I promise!

Everyone with internet access can send him a scan of a 500 Euro note.

[duploxxx]

Nope. Next generation will.

nope the successor of LIano in the Sabine platform will be based on buldozer cores (Piledriver) in the Comal platform, and probably also on the desktop part, Zambezi successor will still only be CPU