Bulldozers first screens

[FlanK3r]

I think, here is not much people, who has final revision of ES...Maybe nobody in this forum. Dont know.

[richierich]

Canihaz omgwtfbbq bulldozer?!?

[keithlm]

Canihaz omgwtfbbq bulldozer?!?

Can we get cheez on the chips?

[El Gappo]

Extra cheez is not compatible with OMGWTFBBQ BULLDOZER on AM3 sockets. That feature will be enabled on OMGWTFBBQAM3r2

[PizzaMan]

Extra cheese will cost you extra. It's not an included topping.

[AKM]

What, no beer?

[keithlm]

What, no beer?

OH... beer cooled.

[stangracin3]

I think, here is not much people, who has final revision of ES...Maybe nobody in this forum. Dont know.

i predict we have a few people lying to us here on the forum. I think chew* has one and i don't care if he says otherwise. I still think he has one if he says he doesn't.

so chew* don't try to convince me otherwise

[RussC]

If he has early silicon, I think B1 or earlier, we know those parts are really slow, ~K8 performance.

RussC

I see online translators completely suck at slovenian or whatever this is

What does this mean Nice consumption at rest, in the load it had such fame is not, but given the number of cores and capabilities, it could wait.

Nice consumption except at load but given nr. of cores and capabilities it could ?


And this?
Significantly increased memory bandwidth, the Sandy's but far from being, Bridge run is still leaps and bounds.

Significantly increased mem bandwith but Sandy Bridge is still better by leaps and bounds?

Maybe some native speaker can clear this up.

[Manicdan]

how much better can the memory get when jumping from thuban to BD, since both are still on AM3/AM3+?
is it even possible for amd to modify the memory section of the cpu to be up there with SB in terms of MB/s of read/write/copy?
i know the northbridge might not really control the memory much anymore, but does it still have an impact on the efficiency?

[[XC] Oj101]

Well for one, official memory support is up from DDR1333 to DDR1866, a theoretical increase of 40% right there. Secondly, we know how NB-F affects bandwidth, and if it's up from 2000MHz to 3200MHz... That'll be another nice jump.

[Manicdan]

1866mhz means almost nothing
my new ram over 2000mhz is hardly 30% faster than ddr2 800mhz i have from 3 years ago, due to timings.

[Particle]

That isn't really the same as in this case where we're talking about the same generation of memory. DDR3-1333 at C7 versus DDR3-1866 at C7 will yield a large improvement. C7 DDR3 is available through at least DDR3-2000 speeds, so it's certainly possible to scale bandwidth a lot more than you're suggesting. It looks like you can even get DDR3-2000 @ C6 these days.

[Manicdan]

That isn't really the same as in this case where we're talking about the same generation of memory. DDR3-1333 at C7 versus DDR3-1866 at C7 will yield a large improvement. C7 DDR3 is available through at least DDR3-2000 speeds, so it's certainly possible to scale bandwidth a lot more than you're suggesting. It looks like you can even get DDR3-2000 @ C6 these days.

im interested in efficiency
right now the same speeds between thuban and SB is no where near the same results in memory tests. can BD jump that gap while still being on the same socket?

[keithlm]

im interested in efficiency
right now the same speeds between thuban and SB is no where near the same results in memory tests. can BD jump that gap while still being on the same socket?

Please define what you mean by "efficiency".

Since the memory controller is on the chip then theoretically there is no reason that bandwidth can't be higher and the latency lower on BD than it was on Deneb/Thuban.

There is no way to predict relative performance at this time.

[Manicdan]

Please define what you mean by "efficiency".

how many MB/s you get when using a specified mhz/timing combination

just looking at the bandwidth section you see top amd chips get 15000MB/s while SB gets 26000MB/s, when both are running similar ram.

i think there is an actual number for efficiency percentage in apps like sandra which tell you how much of your theoretical performance you are getting. ill look for a screenshot

[[XC] Oj101]

And what do SB and Thuban NOT have in common? The memory controller. There's no reason for not being able to use a completely different memory controller with a given number of pins - remember that with the memory controller on the CPU the motherboard doesn't really come into play that much.

[drfedja]

There is few main reasons for increase of bandwidth efficiency with Bulldozer CPU:
1. memory controller is improved, but, memory controller on K10 isn't bad, except relativly low clocked NB/L3 plane. With 3.2 GHz bandwidth efficiency can go up by 20% with fast 1866MHz CL7 RAM. Also latency will go down with fast memory and such fast northbridge.
2. Cache architecture. That is main reason why Nehalem and SB has better memory performance than K10. Now, L1-D cache is inclusive, because BD has large L2 cache, and larger L3, There is no need for snoop trafic between L1-D and L2. Now in L2 BD has copy of L1-D, from core0 and core1.
Also cache bandwidth is much improved.
3. Better prefetcher. L2 cache has non-strided and strided data prefetcher.
Load-Store buffers, and OOO Load-Stores are vastly improved. Now Bulldozer CPU is much more bandwidth hungry, and far less memory latency sensitive, just like Intel.

Of course, we don't know how Dozzers works and we have to wait to see real difference from older AMD CPU's.

[informal]

Good summary list drfedja. We already know from analyst day slide that 20% is core level improvement in IMC while the rest of the quoted(50% memory BW improvement) comes from higher supported DRAM speeds,namely 1600Mhz Vs 1333Mhz for current opterons. So for zambezi we will have 1866Mhz vs current 1333Mhz(Thuban) and on top of this speedup we have the 20% that AMD is stating for improved IMC. In total up to 1.2x1866/1333=68% higher BW with Zambezi X8/X4/X6 running 1866MHz DualCh. memory versus Thuban X6/X4 running 1333Mhz DualCh. memory.

[drfedja]

If you put DDR3 2000 with Thuban Core, you can't get more bandwidth than DDR3 1333 if you have 2 GHz NB only. But if you clock nb@3 GHz, and DDR3 @2 GHz, memory performance will be much better. Also, if CPU cores are clocked higher, memory bandwidth (efficiency) will be higher.
However, if you have better cache architecture and better uarchitecture bandwidth efficiency will be higher.

This slides ilustrates influence of ram frequency and NB frequency to memory performance scaling and everest read/write/cpy bandwidth.

[BeepBeep2]

Guys the box Brian holds and the box the guy from PCGH holds have the same writing on it lol what did some of you expect by blowing it up and enhancing it?

[-Boris-]

1866mhz means almost nothing
my new ram over 2000mhz is hardly 30% faster than ddr2 800mhz i have from 3 years ago, due to timings.

You said it yourself. It isn't the memory speed that makes the difference so small. It's the memory controller in K10 that is to weak.

[FlanK3r]

K10 has good memory copy, but not much good write and read.

[Musho]

Some price information, not sure if this has been posted already.

llano
E2-3250 = 70$
A4-3350 = 80$
A6-3450 = 110$
A6-3450P = 130$
A6-3550 = 150$
A8-3550P = 170$

Bulldozer priced similar to sb.
FX4110 = 190$
FX6110 = 240$
FX8110 = 290$
FX8130P = 320$

PII's are expected to have 15-20% pricecut.
That's probably what they pay per 1000 units, retail will likely be more

The FX6110 is more expensive than 2500k, and the FX4110 is right below it. The FX81xx is actually _more_ expensive than the 2600k. Knowing AMD, they probably wouldn't make it more expensive if it isn't faster. Looking good so far!

[Mechanical Man]

Some price information, not sure if this has been posted already.



The FX6110 is more expensive than 2500k, and the FX4110 is right below it. The FX81xx is actually _more_ expensive than the 2600k. Knowing AMD, they probably wouldn't make it more expensive if it isn't faster. Looking good so far!

FX6 and FX8 have more cores than 2600k tough.