AMD Zambezi news, info, fans !

[flyck]

That would a nice score if it was only 4 cores

I wouldn't be suprised if BD FX8 doesn't run far away from X6 in fpu intensive applications. Given the BD shares it fpu they will have lower core scaling then x6. If you compare it to a deneb at 3,7GHz which gets 4,35points. There is a 50-60% speedup for the same number of fpu ex-units.

not saying those numbers are correct, but i wouldn't expect magic for the part in which they made compromises with BD.

[freeloader]

Well if it doesn't have decent integer performance and lousy FP performance, then what does it have?

[flyck]

Well if it doesn't have decent integer performance and lousy FP performance, then what does it have?

"Better to remain silent and be thought a fool than to speak out and remove all doubt."

[mongoled]

"Better to remain silent and be thought a fool than to speak out and remove all doubt."

LMAO

[freeloader]

"Better to remain silent and be thought a fool than to speak out and remove all doubt."

BD top chip at launch is going to be 10% slower on average than Intel's 2600K. I've said it and I stand by it. There has not been enough IPC improvements to overcome Intel's SB. It will do fantastic in
multi thread programs, but not so good in programs that utilize one or two threads.

[Manicdan]

BD top chip at launch is going to be 10% slower on average than Intel's 2600K. I've said it and I stand by it. There has not been enough IPC improvements to overcome Intel's SB. It will do fantastic in
multi thread programs, but not so good in programs that utilize one or two threads.

in what benchmark? single thread both maxing out turbo, 4 threads? 8 threads? overclocked or stock?

[flyck]

BD top chip at launch is going to be 10% slower on average than Intel's 2600K. I've said it and I stand by it. There has not been enough IPC improvements to overcome Intel's SB. It will do fantastic in
multi thread programs, but not so good in programs that utilize one or two threads.

What kind of logic is that?
-10% on average it cannot win any multithreaded application!
Because if it would win by 20% in multithreaded you won't reach 10% slower on average.

So you probably mean -10% on low threaded applications? Well guess what, AMD has 10-20% clockspeed advantage on anything using 2-4 threads. (which is the majority of the desktop applications). Even phenom2 with l3 at 2.6GHz would be competitive in those applications with that clockspeed advantage. So i see no reason why FX8 8150 wouldn't be competitive in those applications. It will be a different matter for the FX4 and FX6 which do not have such a clockspeed advantage between 2-4 threads. FX4 (~2%) (FX6 has 2-10% for 2-4threads)

[informal]

BD top chip at launch is going to be 10% slower on average than Intel's 2600K. I've said it and I stand by it. There has not been enough IPC improvements to overcome Intel's SB. It will do fantastic in
multi thread programs, but not so good in programs that utilize one or two threads.

Let's see:
2600K on average gets 213.4pts. 1100T gets 165.8pts. 10% slower than 2600K for 3.6Ghz Zambezi would mean 213.4/1.1=194pts. This in turn is around 16% faster than 3.3Ghz X6 Thuban according to the chart. Note that Zambezi has 9% clock advantage.
So basically you expect that Bulldozer design,which is a:
-whole new uarchitecture,with many bottlenecks from K8/K10 removed and better utilization of execution units,new redesigned int and fp pipelines,
-33% more cores,
-13% higher Turbo core clock : 4.2Ghz vs 3.7Ghz (which now runs across 4 cores and not only 3),
-higher default clock of 9% (3.6Ghz vs 3.3Ghz),
-somewhat higher IPC in integer/fp(as per AMD)
-better power management
to run slower than Thuban 1100T. Do you see the problem in your logic? Zambezi will not be only 16% faster than Thuban. Turbo core ALONE will get you there,disregarding the effect of 33% more cores or improvements in the cores.

[repman]

Let's see:
2600K on average gets 213.4pts. 1100T gets 165.8pts. 10% slower than 2600K for 3.6Ghz Zambezi would mean 213.4/1.1=194pts. This in turn is around 16% faster than 3.3Ghz X6 Thuban according to the chart. Note that Zambezi has 9% clock advantage.
So basically you expect that Bulldozer design,which is a:
-whole new uarchitecture,with many bottlenecks from K8/K10 removed and better utilization of execution units,new redesigned int and fp pipelines,
-33% more cores,
-13% higher Turbo core clock : 4.2Ghz vs 3.7Ghz (which now runs across 4 cores and not only 3),
-higher default clock of 9% (3.6Ghz vs 3.3Ghz),
-somewhat higher IPC in integer/fp(as per AMD)
-better power management
to run slower than Thuban 1100T. Do you see the problem in your logic? Zambezi will not be only 16% faster than Thuban. Turbo core ALONE will get you there,disregarding the effect of 33% more cores or improvements in the cores.

Correct me if I'm wrong but isn't the BD "core" smaller than K10 core? How much of a performance impact could that have (single threaded loads)?

[demonkevy666]

Keep in mind Cinebench is very FPU dependent as with all 3d apps. That being said we know BD only has 4 FPU's.

that's only in 256 bit mode is there 4.
in 128 bit There are still 8 of them.
it was also said that a single core can use both 128 bit FPU if needed.

[CrazyNutz]

that's only in 256 bit mode is there 4.
in 128 bit There are still 8 of them.
it was also said that a single core can use both 128 bit FPU if needed.

I see, and since cinebench does not support AVX yet I't would be running the 128bit x 8 configuration.

[cal_guy]

Each floating point unit has 4 pipes. The first two (pipe 0 and pipe 1) contain the units that mainly perform FP arithmetic such as ADD, MUL and FMA, while the other two pipes (pipe 2 and pipe 3) perform the other AVX, SSE, and MMX instructions. When compared to 10h processors bulldozers FPU is more specialized and balanced.