Walt said:
I'm new to 'the groups' so umm Hi. As the subject says I am trying to
build my first PC, but have come across something that puzzles me.
I will be purchasing a EVGA Motherboard with 1000Mhz FSB, along with a
1000Mhz AMD Processor. However I've noticed that there is a
limitation that I can only use up to 800Mhz RAM? That sounds to me
like it would bottleneck a bit. I'm also a little confused with the
concept of DDR2 RAM. If I am using 800Mhz FSB DDR2, with a 1000Mhz
Processor, will I still benefit from having the 1000Mhz processor?
(and why exactly?)
My knowledge of clock cycles, is very limited, I'm going on the theory
of "You have 1000 people in a bus, travelling to where ever they are
going, so you need to make sure everyone has a seat where ever they
arrive also." LOL
Also the processor I'm going to get;
Processor Speed: 64 5000+ / 2.60GHz <--- WHAT'S
THE 5000+ MEAN?
Processor Interface: Socket AM2
Processor Class: Athlon 64 X2
Processor Core: Windsor
Cache Size: 1Mb
Bus Speed: 1000MHz (2000 MT/s)
Additional Technologies: HyperTransport
Architecture: 90 nm
Wattage: 89W
Fan: Included
Please give me any comments on the equipment I'm planning on using, as
I am more a software junkie, than hardware. The only reason I'm not
going for a larger processor - is I'm on a budget, and I figure this
will give me enough power to do my everyday computing.
In this Photoshop benchmark, the X2 4800+ is faster than a P4 extreme edition
running at 3.73GHz. The "4800+" is a Performance Rating, a number that
attempts to compare the processor to a P4. Performance is proportional
to clock_rate times Instructions_per_clock, so the 2.6GHz number doesn't
tell the whole story about the processor. Neither does the 4800+,
but it does a better job than the clock rate does, of indicating
performance. Since Photoshop can use two cores for some of the filters,
you get to see the full effect of the 4800+ number - many applications
can only use one core, and then it'll feel a bit slower.
http://www.xbitlabs.com/images/cpu/athlon64-x2/photoshop.png
I couldn't find a nice block diagram of Athlon64 X2, so drew this
as an example. This is not a picture of your motherboard choice, but
just a generic example. For example, VIA chipsets don't use HyperTransport
between Northbridge and Southbridge, and they use a proprietary bus
instead. But the concepts remain similar.
+----------------+ Memory Busses - dual channel
| Athlon64 X2 | 64 bits
| +------/--------DDR2-800 64*800/8=6.4GB/sec
| |
| | [ Total 12.8GB/sec ]
| | 64 bits
| +------/--------DDR2-800 64*800/8=6.4GB/sec
| |
+----+------+----+
^ | HyperTransport
| |
16 diff. pairs / / 16 diff. pairs at 2GHz (1000MHz dual data rate)
| | Giving max 4GB/sec transfer rate per direction.
| | Separate "up" and "down" bus. 64 wires total.
| v
+----+------+----+
| | 16 diff pairs
| |<-----/-------- A PCI Express x16 slot
| Northbridge | has 16 * 250MB/sec in
| | each direction. Same B.W.
| | 16 diff pairs as the HyperTransport.
| |------/-------> Separate "left" and "right"
| | busses.
+----+------+----+
^ | The bus connection here can be HyperTransport also.
| | Bus width and speed can vary (like only 8 bits or 800MHz etc).
| | The 4GB/sec shown here, would be for a high end SLI chipset.
| |
16 diff. pairs / / 16 diff. pairs at 2GHz (1000MHz dual data rate)
| | Giving max 4GB/sec transfer rate per direction.
| | Separate "up" and "down" bus. 64 wires total.
| v
+----+------+----+ 16
SATA <-> | |----/----> Could connection another PCI Express
| |<--------- X16 slot here...
IDE <-> | Southbridge |
| | 1
USB <-> | |----/----> Or some PCI Express x1 slots...
| |<---------
NIC <-> | |
| +--------+-------+
PHY ^
| | Or a PCI bus
Ethernet | here...
etc v
Note that, in the diagram, bandwidths don't have to add up. You can artificially
create bottlenecks in the diagram, and the computer still works.
The first thing to note, is the processor only has enough bandwidth on the
HyperTransport, to handle one PCI Express x16 slot. Systems which have two
video card slots, average only x8 bandwidth per video card slot.
If you had just one video card, and it was running flat out, and at the same
time you opened a hard drive in Explorer, that would be more than the bandwidth
rating of the HyperTransport bus coming from the processor. Yet all devices take
turns (bus arbitration, various fairness schemes) and the computer still works.
If you look at the memory, using two sticks of RAM in dual channel could
give a theoretical max of 12.8GB/sec, which is more than the 8GB/sec total
for the HyperTransport bus. The memory bus is not 100% efficient, so there
are always "gaps" in time, where nothing is happening. The usable transfer
rate will be less than the theoretical figure.
Since modern architectures have some decoupling of subsystems, the box still
works, even if all the numbers don't add up. For example, if you go into the
BIOS, there is a multiplier setting for the HyperTransport bus. 200 x 5 =
1000MHz for the HyperTransport clock. Two data bits are transferred per
clock cycle (double data rate interface). You have the option of turning
down that multiplier to 1 if you want. The bandwidth to the Northbridge
would drop by a factor of 5. Yet, if you benchmarked some games, you might
notice only a 10-15% drop in frame rate (or even less of a drop). So while
the bandwidths are incredibly high numbers, there is a lot of leeway
in how low they can go and still work in a useful way.
The processor L1 and L2 caches, help to reduce the number of times the
processor has to make (relatively slow) accesses to main memory. When you
get your new system, you can test how important main memory bandwidth is,
by running the system with one stick of RAM (single channel mode) and then
adding the second stick of RAM (dual channel mode). If the RAM is a pair
of 1GB sticks (2x1GB) of DDR2-800 CAS5 memory (about $75 total), even one
stick is enough RAM to run WinXP, so you can compare what happens when the
processor only has half the memory bandwidth available to it.
Paul- Hide quoted text -
- Show quoted text -
