[HH] cores: AMD vs Intel

[Kurt Keville]

I mirror Mark's sentiments... HW purchase decisions are becoming progressively determined by your selection of workload with very little flexibility built around that decision...  the Calxeda description from Anandtech is mostly on target except that he forgot a few caveats...

in many deployments CAPEX is paramount and OPEX matters little if at all... in these scenarios, acquisition cost is deterministic, and most other criteria fade into insignificance... we have many machines at my DC that don't do what we want but they were quite affordable...

For deployments that you actually have to justify the TCO or ROI, I would bet the Calxeda gear pays for itself in 1 or 2 Moore's cycles (3 years), at about which time you are looking to expense your capital assets off the books anyhow... everything is a ratio if you have to make those purchasing decisions; FLOPs per watt or per square foot or per amortized dollar... and then you have to decide if they are "quality" FLOPs or OPs... with ECC or double precision caveats... 

one thing I wasn't sure of in Anand's article... he said the ARM Cortex-A15 was capable of addressing 16 GB but the A9 was only capable of addressing 4GB... they are both ARMv7 so I don't think that is correct...

On Mar 13, 2013, at 1:47 PM, markw at mohawksoft.com wrote:

> A lot of guys are either in Intel or AMD camp and many benchmarks will
> show that intel seems to do better. For my money, however, I have an AMD
> bulldozzer and I think the benchmarks are incorrect.
> 
> If you are a gamer, then an intel chip is probably your best bet. For me,
> I and quite pleased with the AMD bulldozer over the intel offerings. In
> the measurements important to me, hard core threading and processing, AMD
> stands up quite well. For the price, an AMD chip is quite a bargain for
> servers and processing.
> 
> The 8 core AMD chip has 8 real processing cores and 4 floating point
> processors. It isn't really really 8 cores, but it is pretty damn close.
> The intel chips are 4 cores with four hyper-threads. Hyper-threads are
> extremely limited pseudo cores, they are basically a parasitic add on a
> single processing core that shares the instruction processor and
> instruction cache. Unless the OS does a good job scheduling,
> hyper-threading will reduce processing performance.
> 
> I am skeptical of benchmarks because they are of questionable value. Some
> of the benchmarks are things like:
> 
> "media creation" and "office productivity"
> 
> In a database, workstation, or server environment, threads rule. The
> processing is marginally different, and if you choose your I/O carefully,
> cost vs performance ratio is better with AMD.
> 
> 
>> Hello HH,
>> Any nice comparison of per-core performance between AMD and Intel?
>> Because, looking at the FX processors six and eight core prices, they are
>> nowhere near what a 6 core Intel chip costs… so i'd like to educate myself
>> on the why.
>> 
>> I have seen this before with the Phenom chips… but it seems to be going
>> even farther with the FX.
>> 
>> Thanks -F
>> 
>> 
>> _________________________________________
>> -- "'Problem' is a bleak word for challenge" - Richard Fish
>> (Federico L. Lucifredi) - flucifredi at acm.org - GnuPG 0x4A73884C
>> 
>> 
>> 
>>