Opteron 1212 vs Core 2 Duo

[Atxa]

Hello,

I'd want to get an Opteron possibly the 1212, I want to know if it's
better than an Intel E6400 and E6600.

Does anybody know that, I didn't find benchmark yet.

Thank you !

 

[General Schvantzkoph]

Hello,

I'd want to get an Opteron possibly the 1212, I want to know if it's
better than an Intel E6400 and E6600.

Does anybody know that, I didn't find benchmark yet.

Thank you !

The Opteron is a server processor, the E6xxx (Conroe) is a desktop
processor. The Intel Woodcrest is the part to compare an Opteron against.
The Athlon 64 X2 is AMD's dual core desktop part.

For system's with two or fewer CPU chips the Intel parts are faster. On
the desktop you want the Intel part, it's faster and in that
configuration AMD doesn't have any advantages. For two CPU chip servers,
i.e. 2 Opteron 2xx or 2 Woodcrests, it's a closer call. Woodcrest is
faster but it also uses a new type of memory called Fully Buffered RAM
which has horrible latency and is much harder to find (I don't see any
listed on Newegg or Pricewatch). For large servers, i.e. more that two
chips, the Opteron 8xx is the part of choice. AMD parts have on chip
memory controllers and a very efficient processor to processor
interconnect which Intel processors lack. In a single CPU chip system this
doesn't matter at all. The Intel chips suffer from longer memory latency
because of the external memory controller but they more than make up for
it with bigger caches. The lack of hypertransport doesn't matter at all in
single CPU chip systems because there is nothing to talk to. In dual chip
systems the supporting Intel bridge chips have dual frontside buses so the
lack of on chip memory controllers doesn't hurt them their either. The
bandwidth is pretty much the same as the Opterons and there is an
advantage to having equal access to the same memory from both CPU chips.
In an Opteron system a processor has faster access to it's own memory
and degraded (but not terrible) access to the memory on the other chip.
Operating systems are aware of this, so they are smart about how the
allocate physical memory, as a result the Opteron systems don't suffer
much because of the Non Uniform Memory Access (NUMA), however it's still
better to have uniform memory access if you can. Beyond two chips AMD's on
chip memory controllers and hypertransport give them a big advantage.
AMD's memory bandwidth scales because to the on chip memory controllers,
and the hypertransport gives them a glueless interconnect between the
processors. Intel systems at that level are starved for memory bandwidth
and they also have to use the same scarce resource of the FSB to do the
interprocessor communication.

 

[Atxa]

The Opteron is a server processor, the E6xxx (Conroe) is a desktop
processor. The Intel Woodcrest is the part to compare an Opteron against.
The Athlon 64 X2 is AMD's dual core desktop part.

For system's with two or fewer CPU chips the Intel parts are faster.

Thank you !

 

[wizzywiz]

The Opteron is a server processor, the E6xxx (Conroe) is a desktop
processor. The Intel Woodcrest is the part to compare an Opteron against.
The Athlon 64 X2 is AMD's dual core desktop part.

For system's with two or fewer CPU chips the Intel parts are faster. On
the desktop you want the Intel part, it's faster and in that
configuration AMD doesn't have any advantages. For two CPU chip servers,
i.e. 2 Opteron 2xx or 2 Woodcrests, it's a closer call. Woodcrest is
faster but it also uses a new type of memory called Fully Buffered RAM
which has horrible latency and is much harder to find (I don't see any
listed on Newegg or Pricewatch). For large servers, i.e. more that two
chips, the Opteron 8xx is the part of choice. AMD parts have on chip
memory controllers and a very efficient processor to processor
interconnect which Intel processors lack. In a single CPU chip system this
doesn't matter at all. The Intel chips suffer from longer memory latency
because of the external memory controller but they more than make up for
it with bigger caches. The lack of hypertransport doesn't matter at all in
single CPU chip systems because there is nothing to talk to. In dual chip
systems the supporting Intel bridge chips have dual frontside buses so the
lack of on chip memory controllers doesn't hurt them their either. The
bandwidth is pretty much the same as the Opterons and there is an
advantage to having equal access to the same memory from both CPU chips.
In an Opteron system a processor has faster access to it's own memory
and degraded (but not terrible) access to the memory on the other chip.
Operating systems are aware of this, so they are smart about how the
allocate physical memory, as a result the Opteron systems don't suffer
much because of the Non Uniform Memory Access (NUMA), however it's still
better to have uniform memory access if you can. Beyond two chips AMD's on
chip memory controllers and hypertransport give them a big advantage.
AMD's memory bandwidth scales because to the on chip memory controllers,
and the hypertransport gives them a glueless interconnect between the
processors. Intel systems at that level are starved for memory bandwidth
and they also have to use the same scarce resource of the FSB to do the
interprocessor communication.

It seems like the bean counters at Intel and AMD must be calculating
the proportion of CPU sales that go to 2-core systems, 4-core systems,
and so on. If for the next couple years, most chips sold will be in
2-core systems, Intel will make the most money. If the most chips sold
will be in 4+ -core systems, AMD will make more money. This must have
been calculated before Intel and AMD designed their chips (??).

I wonder if AMD and Intel sell more server class chips or more desktop
chips? And I wonder what the typical number of CPUs per server is.

wiz

 

[General Schvantzkoph]

It seems like the bean counters at Intel and AMD must be calculating
the proportion of CPU sales that go to 2-core systems, 4-core systems,
and so on. If for the next couple years, most chips sold will be in
2-core systems, Intel will make the most money. If the most chips sold
will be in 4+ -core systems, AMD will make more money. This must have
been calculated before Intel and AMD designed their chips (??).

I wonder if AMD and Intel sell more server class chips or more desktop
chips? And I wonder what the typical number of CPUs per server is.

wiz

They make more money on the server chips. The manufacturing cost of an
Opteron and Athlon 64 is nearly identical but the price of a 2xx Opteron
is significantly higher than the cost of a A64 at the same clock rate. The
8xx parts are four or five time the price of an A64. The chips
are basically the same, the only difference is if bus snooping is enabled
on the hypertransport buses. My guess is that it's just a metal option
that enables the cache coherency on a hypertransport interface. I could be
wrong, they might be able to use a simpler cache on a A64 but even if they
do that doesn't account for a 2-4x price difference.

 

[NilEinne]

On Sat, 30 Sep 2006 11:21:22 -0700, wizzywiz wrote:
They make more money on the server chips. The manufacturing cost of an
Opteron and Athlon 64 is nearly identical but the price of a 2xx Opteron
is significantly higher than the cost of a A64 at the same clock rate. The
8xx parts are four or five time the price of an A64. The chips
are basically the same, the only difference is if bus snooping is enabled
on the hypertransport buses. My guess is that it's just a metal option
that enables the cache coherency on a hypertransport interface. I could be
wrong, they might be able to use a simpler cache on a A64 but even if they
do that doesn't account for a 2-4x price difference.

I know the 1xxx are just the same as normal A64s but are you sure the
2xxx and 8xxx are? I believe AMD plans to support FB-DIMM in the future
so I would assume in the future at least they won't be the same as I
would assume AMD won't build the FB-DIMM capable memory controller into
the A64 and just disable it.

In any case, I think it's worth pointing out that the Opterons tend to
be the cream of the crop so to speak (as demonstrated by their
overclocking potential)

 

[NilEinne]

On Mon, 11 Sep 2006 14:56:51 -0400, Atxa wrote:
The Opteron is a server processor, the E6xxx (Conroe) is a desktop
processor. The Intel Woodcrest is the part to compare an Opteron against.
The Athlon 64 X2 is AMD's dual core desktop part.

For system's with two or fewer CPU chips the Intel parts are faster. On
the desktop you want the Intel part, it's faster and in that
configuration AMD doesn't have any advantages. For two CPU chip servers,
i.e. 2 Opteron 2xx or 2 Woodcrests, it's a closer call. Woodcrest is
faster but it also uses a new type of memory called Fully Buffered RAM
which has horrible latency and is much harder to find (I don't see any
listed on Newegg or Pricewatch). For large servers, i.e. more that two
chips, the Opteron 8xx is the part of choice. AMD parts have on chip
memory controllers and a very efficient processor to processor
interconnect which Intel processors lack. In a single CPU chip system this
doesn't matter at all. The Intel chips suffer from longer memory latency
because of the external memory controller but they more than make up for
it with bigger caches. The lack of hypertransport doesn't matter at all in
single CPU chip systems because there is nothing to talk to. In dual chip
systems the supporting Intel bridge chips have dual frontside buses so the
lack of on chip memory controllers doesn't hurt them their either. The
bandwidth is pretty much the same as the Opterons and there is an
advantage to having equal access to the same memory from both CPU chips.
In an Opteron system a processor has faster access to it's own memory
and degraded (but not terrible) access to the memory on the other chip.
Operating systems are aware of this, so they are smart about how the
allocate physical memory, as a result the Opteron systems don't suffer
much because of the Non Uniform Memory Access (NUMA), however it's still
better to have uniform memory access if you can. Beyond two chips AMD's on
chip memory controllers and hypertransport give them a big advantage.
AMD's memory bandwidth scales because to the on chip memory controllers,
and the hypertransport gives them a glueless interconnect between the
processors. Intel systems at that level are starved for memory bandwidth
and they also have to use the same scarce resource of the FSB to do the
interprocessor communication.

I wonder if your info might be a bit misleading. I haven't looked into
Opteron's at all but since there is limited difference between Opterons
and A64s and Woodcrests and Conroes what I do know about the A64 vs
Conroe should help. AFAIK, you need about 20% more clockspeed to make
up the average difference between Conroe and A64s. Obviously this
depends a lot on app. Apps for a server/workstation will vary
significantly from typical desktops so obviously you should get info on
how your specific app config compares between the two. However my point
is what you really need to consider is the price/performance. I've seen
this a lot with the Conroe in particular where people say the Conroe
destroys the A64 but this is simply not true. You need to consider the
price of an equivalently priced and speced AMD vs Intel system. In the
Conroe case for example AFAIK if your comparing a similarly speced A64
X2 4200+ (2.2ghz) to a E6300 there is no clear performance advantage
(without overclocking).

Obviously when comparing a Opty 1212 (2ghz) the Woodcrest 5110 for
example I suspect the Opty 1212 would win (although it'll depend a lot
on your apps).

Clearly it depends a lot on the price for the system. My point is it's
silly IMHO to state Intel is the clear winner simply because the
they're faster for an equivalent clock speed. This is irrelevant. What
matters is the price for an equivalently specced system.

N.B. AFAIK various tests have also shown A64 systems may have lower
power consumption for the average user as CnQ appears to work better
then whatever Intel calls their thing. Obviously this is even more
pronounced for EE models.

 

[General Schvantzkoph]

I know the 1xxx are just the same as normal A64s but are you sure the
2xxx and 8xxx are? I believe AMD plans to support FB-DIMM in the future
so I would assume in the future at least they won't be the same as I
would assume AMD won't build the FB-DIMM capable memory controller into
the A64 and just disable it.

In any case, I think it's worth pointing out that the Opterons tend to
be the cream of the crop so to speak (as demonstrated by their
overclocking potential)

AMD has backed away from supporting FB DIMMs in the future, for that
matter Intel is planning on dropping it in future chipsets. With the 940
pin Opterons it was pretty clear that there were no real differences
between the Opteron and the Athlon64. The Opteron uses registered DIMMs
and the A64 uses unbuffered DIMMs but there is no real difference between
a memory controller that does registered DIMMs and one that does
unbuffered, it's just a difference in timings. It would be crazy for AMD
to have two different controllers when one controller would do so I'm dead
certain that the only difference is a mode signal that's either set in the
mask on in the package. With a hypertransport interface most of logic
between a coherent interface and an incoherent interface is identical.
There might be a cost difference between a cache that can handle bus
snooping and one that can't, however my guess is that the cost difference
is small enough that AMD is willing to eat the slightly higher cost of a
A64 that contains disabled coherency logic in return for the flexibility
of being able to produce A64s and Opterons at the same time.

As for the 1212 pin Opterons, can someone tell me what they added to
justify the additional pins? I've looked over their product briefs and it
doesn't jump out at me. With that many extra pins I would have thought
they would have made the memory interfaces wider or added an additional
hypertransport bus, but that doesn't seem to be the case.