Dual Core Athlon 64 Preview
- Thread starter swaaye
- Start date
Indeed, im just awed at its benchmark results. Hope we can get something more detailed soon. :drool:swaaye said:
epic
I'm as excited about dual core as the next guy (provided that the next guy is a hardware geek, that is), but I am not sure why you find this to be so impressive. It's doing everything it is expected to do. It is matching two Opteron 250s (so one dual core at 2.4GHz v two CPUs at 2.4GHz each...we should expect a tie). Did you expect dual core to be less performant here than two separate CPUs working in tandem?
Killer-Kris
Regular
ANova said:
I might be mistaken, but what I think it's doing is simply running in single threaded mode. So the extra core is not actually disabled, it's just not being used.
wireframe said:
Yes I sorta did. The dual Opteron setups have independent memory setups for each CPU. This dual core CPU has to share one dual channel RAM bank I believe. Actually in one of those tests the dual core was ahead of the dual Opterons, perhaps because of core improvements in this stepping.
Killer-Kris
Regular
swaaye said:
But you forget to factor in the performance boost that you get from not having to do cache coherency over an external bus.
you both hit it on the nail. (imo). There were many questions that needed to be answered, this has at least shown that its going to be good news for amd. Intel went a different route in their dual core, and its going to be interesting to see who did it better.Killer-Kris said:
epic
swaaye said:
*sigh* I am sighing because I wrote this huge piece on interepreting the data and some additional discussion on what could be done, but I pressed the delete key because I think it's safer to just say this for now:
Look at the scores between Opteron 250 and dual Athlon 64 (2.4GHz) and ask yourself this question:
Can the slight improvement in performance be explained by Opteron 250 using registered/ECC memory (this must be true) and the Athlon 64 dual core using unregistered/non-ECC memory (probably true)?
My answer is: yes.
As for the other points I had in mind, inluding SSE3, shared memory controller, and FSB bottlnecking, let me leave that for later in the thread when we will no doubt talk about "what could be done with a dual core CPU to really impress."
Hahah ok. Sorry to hear about the grand post . If it was using unbuffered that could explain it I suppose. However, the dual-core chip is undoubtedly putting more stress on its single memory subsystem, whereas the dual Opterons have more overall bandwidth per chip than the dual core, per core.
Is there any way this could really be better?.... The dual core A64 basically performs as well as dual Opterons, which I believe is the whole point of the dual core chip. K8 again demonstrates that AMD developed a very balanced, scalable chip. Now we get to look forward to dual dual-core chips, lol.
. If it was using unbuffered that could explain it I suppose. However, the dual-core chip is undoubtedly putting more stress on its single memory subsystem, whereas the dual Opterons have more overall bandwidth per chip than the dual core, per core.Is there any way this could really be better?.... The dual core A64 basically performs as well as dual Opterons, which I believe is the whole point of the dual core chip. K8 again demonstrates that AMD developed a very balanced, scalable chip. Now we get to look forward to dual dual-core chips, lol.
swaaye said:
We know, from the article, that the dual core Athlon 64 is running on a socket 939 board. Unless something else has changed I think it is very safe to assume the memory type is unregistered. We know for a fact that Opteron 250 is using registered RAM without the article stating this because this is always the case.
I think you'll find that many 2P opteron boards use a shared memory controller where one CPU acts as master. I would also believe this to be the case in this benchmark, but I cannot prove that. Read on to find out why this is moot.
However, look at the Xeon (NetBurst) performance scaling and you can get a very strong hint that this test is not FSB dependent. NetBurst is very demanding on FSB/memory bandwidth. Now compare the single NetBurst 3.6GHz to the dual Xeon 3.6s that share an FSB. The 2P Xeon system scales very well. So well, in fact, that it suggests there is no FSB contention.
So this test doesn't seem like it would reveal the subtle nuances you may see going from a 2P Opteron system with independent memory busses to a shared bus dual core. However, for most intents and purposes the shared memory bus even on Opteron 2xx setups is not a major problem. The memory independence really only becomes important for very large scaling 4->8->beyond and certain applications. It would not make much sense to think and design in this direction for a desktop part meant to operate only as a 2P (dual core Athlons still cannot be coupled with more CPUs, you need an Opteron for that).
This was exactly the gist of my initial post. The Athlon 64 dual core is performing exactly as it should be expected without any real enhancement beyond the packaging and some benefits that may bring in terms of power consumption. So, it's great that dual core make the mainboard cheaper, in comparison to dual socket boards, and eveything simpler (chassis, PSU, etc, unregistered memory, etc), but it really has no "wow" beyond what is to be expected. That was my point.
My soon-to-be legendary missing post
P) addressed this aspect a bit more. By having two cores on one piece of silicon and in one package, you can do some magic that would be impossible with two separated cores such as in a 2P Opteron 250. AMD is not doing this (for now), but this would have been a real "wow! look at that!" as opposed to what we are seeing, which is completely predictable.Also of note is that this engineering sample is running dual cores at 2.4Ghz. It was thought AMD would clock the dual core chips lower due to heat difficulties, but it seems that if an engineering sample is running 2.4Ghz that things are not nearly so bad. Perhaps they did implement some enhanced thermal mechanisms.
Similar threads
