AMD Zambezi news, info, fans !

[xVeinx]

http://news.morningstar.com/articlen...aspx?id=396010

the old story ended and new one's just start

I believe that, grammatically, the author meant to say that once these manufacturing problems are resolved. The "As manufacturing problems are resolved" doesn't denote completion but an ongoing effort; having said "are now resolved" would be less ambiguous. There was nothing in AMD's past statement that indicated that things had completely cleared up, so I think this to be the most likely scenario, especially if Interlagos is being pushed further back...

[Apokalipse]

It's an 4 core, 8 integer clusters simple and plain.
More like a HT thing.
Intel Core quad aproach( to put 2 dual cores together) was more "true" core design than BD is now.
8 cores, o yeah...

BD modules are completely different to HyperThreading.

They achieve the same basic goal (high increase in MT performance with tiny increase in die area), but they do it very differently.

a BD module has extra hardware to run two independent threads simultaneously. Hyperthreading uses spare execution resources of one core.

[TESKATLIPOKA]

Oliverda

Can you tell me that a non 8 core processor how can do 6.72x speedup in a multithreaded task?

Can you tell me why SB2600 is getting 4.5x speedup, isn't it just a 4 core?
For a stupid question a stupid answer.
By the way Deneb has 3.91x so almost the theoretical maximum, SB2500 has 3.67x speedup. AMD 8 core should have 7.82x and not 6.72x, do you see the difference?

SB2600 and BD are not 8 cores and never will be, one is using SMT(simultaneous multi threading) the other CMT(cluster-based multi threading).
Here is a nice picture, also in patent AMD calls it as a core with 2 integer clusters(module).
http://www.blog.de/srv/media/media_p...tem_ID=3663732
Phenom x4 980 is 4 CMP cores
SB2500 is 4 CMP cores
SB2600 is 4CMP cores +HT
BD8150 is 4CMP cores +CMT and AMD has 2 names for it 4 modules what is correct and for BFU they call it as 8 core because they buy what has bigger frequency, more cores more Adidas bands.

And here is a comparison of power, size and performance, that won't ever change and please use the same architecture basis, lets say SB.
4 cores CMP < 4cores CMP +HT < 4cores CMP +CMT < 8cores CMP

[tifosi]

...A true core doesn't share L1 cache, FPU, prefetch with other core...
A core could have 2 clusters which share between them all this and work in 2 threads....
Anyway AMD 4C/8treads cpu beats INTEL 2500K quad would have sound better than...
AMD 8 core beats Intel 2500K quad...

Psst... if you read carefully. BD chippery has dedicated L1. However, L2 is shared between cores of a module. Try and read a little before you start actually beating down on something.

Also, why would Intel 2500K sound better if 8150 beats it hands down and in most multithreaded scenarios will sit between 2600K/ Gulftown and SB-E? If you talk about gaming, Canard is already stating that BD can keep up with SB's (and that's not fully optimized yet).

Comparisons are inevitable, but trolling is not! So if people are coming over to AMD forum and a BD thread and talk about SB-E, they should also note the difference in costs of platforms, with SB-E coming almost at thrice the cost of BD. Also, why'd you want to pay more for something (competing Intel products) with little difference if any at all? Also, there are no SATA problems what so ever(chew also said that... but you omitted that from your post mate :P )

[TESKATLIPOKA]

tifosi

Psst... if you read carefully. BD chippery has dedicated L1. However, L2 is shared between cores of a module. Try and read a little before you start actually beating down on something.

The same for you, try reading something about BD before you start posting.
There are 2 different L1 caches, data and instruction.
L1D is dedicated 16KB for each integer cluster as you said but there is another shared L1I 64KB (instruction cache).
Try reading it once more

Anyway AMD 4C/8treads cpu beats INTEL 2500K quad would have sound better than...
AMD 8 core beats Intel 2500K quad...

What xdan said in this quote is actually correct, people will look down on BD because an 8core can't compete against a 6 core SB-E, but if its 4 core against 6 cores then its understandable or why BD competes against SB2600 because 4 vs 4 cores capable of executing 8 threads.

Actually it's a dilemma for AMD. For BFU its better if you call it as 8 core, but for people like us it's better to call it as 4 core or 4 module because we look at tests and also how it looks inside the chip.

[Olivon]

If you talk about gaming, Canard is already stating that BD can keep up with SB's (and that's not fully optimized yet).

No. In Canard PC Hardware N°10 (oct-nov), Doc TB said that FX lags behinds SB regarding game :

http://library.madeinpresse.fr/samples/MPqY2Vg2v45Z-f

Page 8/9 for those who understand french

edit : Translation

"In the more upscale, regarding FX processors based on the Bulldozer architecture we are also disappointed : while they are still generally more performant than their predecessors and allow AMD to approach much of the last Core i5 and i7 but their performance remains below expectations. Besides, as we announced already in our previous issue, if they can sometimes compete with Sandy Bridge in the applications of rough calculations, the results are in video games very far behind. Only overclockers (and fanboys) will find them a great interest given their predisposition in this area."

"The AMD FX-8120 is probably the model of the new series "Bulldozer" that offers the best price/performance ratio. It is able to compete with the i5-2500K in most computing applications even if lags behind in video games. But, don't forget substantial overclocking capabilities are available with no additional cost. Faced with the old Phenom X4 980, this is a very good alternative."

"Offered at a price slightly lower than the Core i7-2600K Intel, the FX-8150 is currently the most powerful model from the new architecture "Bulldozer" for AMD. Unfortunately, he fails, at best, it cans match its direct competitor in some media processing applications but is always behind in games."

[Oliverda]

I'm sure that there will be some Asrock mb's at 200$. - Extreme6- Extreme 4

Calm down. There won't be any mobo on that price in the near future.

Oliverda

Can you tell me why SB2600 is getting 4.5x speedup, isn't it just a 4 core?
For a stupid question a stupid answer.
By the way Deneb has 3.91x so almost the theoretical maximum, SB2500 has 3.67x speedup. AMD 8 core should have 7.82x and not 6.72x, do you see the difference?

Don't you feel any difference between 4.5x and 6.72x?



Anyway the six-core Phenom II X6 speedup is only around 5x.

[FlanK3r]

Its exactly 5.2x

If FX8150 is slower in single and a bit better in multi, than single can be about 0.8 and multiscore about 5.9x, efectivity 7.4x in R11.5?! But example in R10 is efectivity of CPUs lower

[drfedja]

Anyway the six-core Phenom II X6 speedup is only around 5x.

Including turbo for single thread.

[liberato87]

some "benchmarks" spotted by italian friends (General Blue , TheBestFix) on thw hwupgrade.it forum thread about bulldozer (http://www.hwupgrade.it/forum/showth...7299&page=1246)

http://www.cpubenchmark.net/high_end_cpus.html







and http://forums.anandtech.com/showpost...postcount=2046

[gosh]

No. In Canard PC Hardware N°10 (oct-nov), Doc TB said that FX lags behinds SB regarding game :

Are there any graphs?
Some magazines have writers who write about things because they are good writers, they don't really need to know what they are writing about. Maybe the writer have asked someone about AMD cpu's and that person talked about phenoms

[Olivon]

Are there any graphs?
Some magazines have writers who write about things because they are good writers, they don't really need to know what they are writing about. Maybe the writer have asked someone about AMD cpu's and that person talked about phenoms

The writer is Doc TB aka xsecret aka Samuel D. who's appeared in this thread ...

[informal]

Yes, xsecret is a reliable guy. I'd trust what he says. But as always,we should wait for NDA to be lifted and rest of the reviews.

As for the passmark results, just look at the 2500K vs 260K,the difference is a massive 33+%. Do you see this in real world benchmarks? Yes,but only sometimes. Usually they are very close and only in some workloads that are well MT 2600K can have such a lead. This tells us that passmark suite is geared towards the MT aspect and says nothing about single core performance . 8150 ,if it ran at stock clocks, is 37% faster than 1100T.

[FlanK3r]

Passmark looks good, special in performance/price value (because first is x6 1090T-whorse eprformance than FX, no much lower price)

[TESKATLIPOKA]

Oliverda Its actually 5.26x Thuban 1100t

Don't you feel any difference between 4.5x and 6.72x?

Yeah I feel it and? one is HT(SMT) the other is module+SMT
Don't you feel any difference between 3.67x and 4.5x? The same 4 core chip.

Anyway the six-core Phenom II X6 speedup is only around 5.26x.

In the same link you have other AMD cpu's or in this http://techreport.com/articles.x/20188/14
x4 910e 3.93x
x4 640 3.92x
x3 455 2.92x
x2 565 1.96x
Its evident the efficiency for some reason is lower in Thuban than in Deneb or Heka or Regor.

BTW Cinebench uses FPU and that is one in BD shared between 2 integers yet it has also HT for higher utilization and that's the reason why the speed up is so high.

P.S. what about the rest of my post? Do you agree or not.

[muziqaz]

Efficiency tends do drop with higher thread count. It might not be a CPU fault as multithreaded software might be loosing that efficiency as it has more overhead dealing with more threads.

[Oliverda]

Oliverda Its actually 5.26x Thuban 1100t

I said around. 1055T's speedup is 5.07x for example.

Yeah I feel it and?

Doesn't look like.


Please share us the official description of one x86 core.

[Musho]

Working with the things we know, we can calculate how fast roughly the 8150 is compared to the 1100T in single threaded tests. Looking at Cinebench screenshot comparing the 1100T singlethreaded vs multithreaded, the difference is a 5,261261261 speedup, which is explained by the fact in the single threaded load the 1100T is running at 3,7ghz, while in the multithreaded test it's running at 3,3ghz. Those CPU Passmark benchmarks posted above are very well multithreaded so we'll assume it will show roughly the same scaling going from multithreaded to singlethreaded:

6314/5,261261261 = 1200 for the 1100T @ 3,7ghz

Now we know from movie posted about Bulldozer running Fritz Chess benchmark, that turbo in single threaded workloads can keep Bulldozer running at 4,2ghz with 1-2 cores loaded, while it runs at 3,6ghz with 8 cores loaded. Also, the cores are running at 80% efficiency when both cores in a single module are loaded, so instead of dividing by 8, we'll be dividing by:
8*0,8 = 6,4.
We also shouldn't forget the speed difference in multi and single threaded tests:
4,2/3,6 = 1,166666667

This brings Bulldozers calculated single thread score to:

8681/6,4*1,166666667 = 1582 for 8150 @ 4,2ghz

This gives Bulldozer a 1582/1200 = 1,319 or 31,9% speedup compared to thuban in single threaded workloads at stock speeds with turbo enabled for both.

OR

(1582/4,2)/(1200/3,7) = 1,162 or 16,2% IPC boost

Disclaimer: This is all just some theorycrafting, but it feels right about what to expect. Take these calculations with a grain of salt however

[Warwian]

Can anyone help me understand how the PassMark scores are calculated? For example: the i7 2600K looks to have a 1021 point advantage over the i7 2600. (http://www.cpubenchmark.net/high_end_cpus.html)
Isn't the only difference between the i7 2600K and the i7 2600 an unlocked multiplier and beefier IGP on the K-model? Does this affect the score somehow? Or do they include overclocked samples when calculating the score?

[undone]

hmmmm looks like the bench result looks better than before, and getting better and better, and Oct.12 announcement become more unrealistic, what's going on?

[slaveondope]

Can anyone help me understand how the PassMark scores are calculated? For example: the i7 2600K looks to have a 1021 point advantage over the i7 2600. (http://www.cpubenchmark.net/high_end_cpus.html)
Isn't the only difference between the i7 2600K and the i7 2600 an unlocked multiplier and beefier IGP on the K-model? Does this affect the score somehow? Or do they include overclocked samples when calculating the score?

Likely many/most of the 2600K submissions are overclocked hints the spread from 2600.

[Olivon]

Likely many/most of the 2600K submissions are overclocked hints the spread from 2600.

Overclocked CPU's got another page :

http://www.cpubenchmark.net/overclocked_cpus.html

[slaveondope]

Overclocked CPU's got another page :

http://www.cpubenchmark.net/overclocked_cpus.html

OK so they've separated them since my last look at Passmark. Any explanation why 2600K would perform better than a 2600 when both at stock?

[Olivon]

Maybe more weak platforms, H67, H61 with 1333 RAM, honestly, I don't know slaveondope

[tbone8ty]

http://www.pcper.com/news/General-Te...ober-12th-Shhh