Topic: CPU mining and Bulldozer ? (Read 2353 times)

As I indicated above BD may be inferior to Phenom II but it has 8 integer cores and scrypt based hashing seems to really only care about having enough fast cache and clockspeed.   If BD smaller L1 cache isn't a bottleneck the fact that it has 25% more cores means on a per CPU basis (not per core basis) I would expect it could outperform Phenom II.

As far as ASICS.  Meh.  By the time ASIC miners hit the market (if ever) your BD will be the equivelent of a 486 anyways.  I mean it would be like saying I am not going to buy a Bulldozer today because a 16 integer core Bulldozer 4 in released in 2016 will destroy it

I expected to build new computer based on bulldozer CPU, but the Bulldozer sometimes is inferior to Phenom II x6. And the CPU-Based mining will be destroyed by ASIC chips just like GPU mining made CPU mining almost useless.

I thought of getting a BD but
a) it tends to suck for most applications
b) there is a very good chance cpu-based chains go nowhere.

If BD had better general performance I would buy one but currently it seems like a dud unless AMD significantly cuts the price.

I hope we get some results soon
Can't afford a Bulldozer right now but maybe next year

The scrypt code itself has two separate loops that use 4k memory each, from what I can tell, so >8k is good to have.  Beyond that, it also uses salsa20/8 and SHA256, and I don't know how much memory those use up, if they ever even drop out of the registers.

FWIW my Athlon XP sees a bit more than 1/2 the performance per GHz per core( vs Phenom II) when hashing scrypt coins, but don't know how much is due to cache size, special instruction architecture(using amdfam10 for hashing on the phenom), or something else.  Of course that's per core, so the Phenom II can do a whole hell of a lot more than just 2x overall.    In other words:  Athlon hashes at 1.15k per sec at 2.4 GHz, Phenom hashes at 3.2k per sec at 3.8GHz per core.  Also, the phenom did almost exactly 3k per sec when it was stock clocked at 3.4.

The Bulldozer is a failure!

these video parodies never get old- lol

the video is F*&^ing hilarious

did anyone buy a bulldozer chip and test it out on tenebrix?

^Yeah,I read something about that in the upcoming AVX2 instructions as it pertains to Ivy bridge about increasing integer performance.

Really only two things affect scrypt based "CPU alt coins".  

1) Cache size
2) Number of Cores

Pros:
The Bullldozer combines 2x integer cores (only thing hashing uses) w/ a single floating point core.  On paper this is beneficial because more of the die is devoted to integer performance.  Thus you are getting 8 integer cores (vs 6 on Phenom II) for roughly the same die space.

Cons:
It only has 16K of L1 cache (vs 64K on Phenom II chips).  That might not be enough to avoid a bottleneck when processing scrypt algorithm.  If it isn't then you have the longer latency trip to L2 cache.

Irrelevant:
Turbo Boost.  all 8 cores are going to be maxed you won't have any TDP left for any boost.
lower idle wattage - LOLZ
Enhanced L3 cache - useless for hashing
higher speed DDR3 - nothing
A bunch of new instructions (XOP, FMA4 and CVT16, SSE4, AVX, etc).  I looked over them briefly none look promising for improving code efficiency.

So without having the chip in my hand my SWAG would be:

If 16K of L1 cache doesn't cripple performance then you are likely getting roughly same performance per core as Phenom II chips.  8 cores vs 6 cores is your real improvement.  It comes down then to cost.  Price Phenom II vs Bulldozer on per core basis and take cheaper one.

On the other hand if the reduced L1 cache cripples performance then they are junk.  They will never run scrypt based algorithms well and no amount of tweaking is going to change that.


The future:
Keep an eye on improvements to AVX instruction set in future versions of chips (both AMD & Intel).  Currently AVX only works w/ floating point data types.  There are some whitepapers on extending AVX instructions to integer datatypes.  This would allow using the 256bit SIMD registers to process 8x 32bit integers at the same time. Making CPU more "parallel" more GPU like.  Too early to say but it is "interesting".

Considering the latest Alt-currencies have opted more for CPU mining and Bulldozer is basically paper launch,how do you think it would fair for CPU mining the latest forks ?

To me the benchmarks are the following in a nutshell :

http://www.youtube.com/watch?v=SArxcnpXStE&feature=youtu.be