Monday, October 22nd 2012
AMD Opteron "Piledriver" Processors Arrive Mid-November
While AMD FX "Piledriver" client processors in the AM3+ package are just around the corner, slated for a little later this month, the company's first enterprise processors for servers and workstations, based on the new micro-architecture, are slated for mid-November, according to a report. AMD could begin with an overhaul of its multi-socket enabled Opteron 6200 series and single/dual-socket enabled Opteron 4200 series with the new Opteron 6300/4300 series, featuring the 8-core "Piledriver" silicon. The multi-socket enabled Opteron 6300 series will consist of nine models, tabled below.
Source:
MyDrivers
47 Comments on AMD Opteron "Piledriver" Processors Arrive Mid-November
anyhow I think it's kinda odd the amount of L3 cache doesn't scale with the number of cores.
Intel cannot compete against these chips in server. They're much cheaper and as fast or faster for the same solution. The only exception is if you're using some draconian software licensing per core.
But, nothing will change, just as we saw with the XP and 64. Marketing, brainwashing, and the money to coerce will win out.
i know of no reason why intel's treason
should make AMDs performance drop ;p
But, nothing will change, just as we saw with the XP and 64. Marketing, brainwashing, and the money to coerce will win out.
You mention power consumption, and you're right that it's a big deal. But when you can buy one 130W Xeon to do the work of two 140W Opterons, it just falls as an argument in Intel's favour.
All AMD have going for them in this space is that the initial purchase cost of Opterons is far lower.
I mean isnt that what you bought your machine initially for too?
btw TDP is measured differently between both companies so it still is a moot point, there is no standard of measuring it
/thread
Also,
max(bang/buck)
does not equal
min(buck) for which (bang=x)
And in addition, when we talk about bang/buck in consumer terms, we normally totally ignore "buck" contributors other than initial purchase price.
The sad thing about this is that I actually partially agree with you. Opterons are the superior value proposition for most users, providing that no floating point capability is required. But your irrational argumentation and unwillingness to acknowledge rational arguments that oppose you totally discredits you.
www.anandtech.com/bench/Product/287?vs=434
See also FX41** vs. i3, etc.
www.anandtech.com/show/5553/the-xeon-e52600-dual-sandybridge-for-servers/2
So I expect these Xeons to perform around the level of existing SB 8-cores.
So lets consider the things that each does best.
Xeon:
Good IPC
High Clocks
8 real cores, plus 8 non-linear scaling threads (consider that HT performance is not consistent and can suffer when doing similar tasks at once).
Opteron:
Lower price
16 "almost real cores", (Integer ops here are the best, floating point ops aren't as powerful but you know how much every module will give you as far as performance is concerned.)
So with all of this said, yeah Intel might be a little bit faster per thread because of the improved IPC but AMD is getting really good at cramming a lot of cores on to a CPU. HyperThreading is great but if you're using the same parts of the CPU for both threads that are running you're performance could suffer on your HT threads. At least with AMD all 16 of those threads have dedicated resources with the exception of the floating point unit, but even that can be efficiently used with the proper instructions.
So all in all, as a System Admin, I would rather have an Opteron for a database or a computing cluster where for an application server, a Xeon might be better. For storage, I would opt whichever used the least power in combination with the price tag.
So how about everyone who is uninformed doing this bulls**t move of "AMD Opteron is better!" and "Intel Xeon is better!" just drop it. Both the Xeons and Opterons are both very good processors and do certain things better than the other.
Might want to ask BuckNasty about his one and about to be two 4P Opteron servers using the older magny cours 12 core procs, that's 48 cores. I'm willing to be he paid no more for that entire machine than for one middle model 8-core Xeon. The Xeon might be faster, but the big question is, would it be worth it and is it cost effective. Most of that performance mostly isn't off the hyper-threading. It's also highly task dependent. Consider for a moment if your "core" is using the ALU to add and your HT thread wants to add too, the HT thread has to wait until the core "thread" is done adding. Then the HT thread goes to add, and if the core has to do it again, it will have to wait for the HT thread. (This is more true of instructions that take more than a couple cycles, so ADD isn't an accurate one because I'm pretty sure both Intel and AMD ALUs do integer ADDs in one cycle.) The general point being, is that AMD CPUs scale almost linearly as more workload is applied to the server, where Intel servers with HT enabled do not.
Example: for Virtualization you need as many physical cores you can get (in this case 4x16). Hyper-threading won't help you here, and I remember reading a document (provided by HP for their servers) that you should actually disable HT.
Simply saying Intel does everything better is not true. Depends....sometimes you only look at the performance demands which need to be met.
And also a thing to remember is that the initial cost of hardware is lower than software (licensing etc. sometimes a lot lower). You also need to plan the power consumption over a longer period of time (my ML350G4 worked 24/7 from the day it was turned on) and the cooling costs.
AMD has winner here.