CPU Speed vs. FSB speed

Do you have a question? Post it now! No Registration Necessary.  Now with pictures!

Threaded View
Hi there

The plan being to build a new server based on the Tyan i5000XT MB, I
am stuck on the decision on what Xeon processors to buy. I am not
going for leading edge (read expensive) CPU but would be greatly
interested in an opinon on what should have priority: BUS speed or CPU

My preferences are the Intel Xeon 5060 at 3,2 GHz and a 1066MHz FSB
OR the Intel Xeon 5130 at 2.0 GHz and 1333MHz FSB.

These CPU cost just about the same. How do I decide between them?

The machine will be used as a high end Workstation with a decent
graphics card (but not gaming), and I plan two Seagate 73GB 15k SAS in
a mirrored RAID; and the same for two Western Digital 36 GB Raptor 10K
drives.  I am comfortable with my HDD decisions; believing that the
two separate controllers will be an advantage over more capactity on
only two drives on, eg the SAS controller.

But, as I said, how do I decide on the key question of BUS vs. CPU

Re: CPU Speed vs. FSB speed

PeterF wrote:
Quoted text here. Click to load it

5130 - Dual core, 4MB cache, 65nm, FSB1333, Core 2GHz, LGA771, 65W

5060 - Dual core, 4MB cache, 65nm, FSB1066, Core 3.2Ghz, LGA771, 130W

Now, if I poke around a bit, the CPUID of a Pentium D 950 is 0F64.

So that would make the 5060 the same generation as "Pressler".

A CPUID of 06F6, is the same as a Core2 Duo. That means the 5130
is similar to a Conroe.

Now, the Xeons probably have their own internal code names, that
you can see in a list here. 5060 is "Dempsey". 5130 is Woodcrest.


The difference between the processor generations, is in the IPC
or Instructions Per Clock. For example, compare the benchmark here.
The Conroe at 2133MHz core clock, is beating the Pressler at 3.2Ghz.
This is an Integer benchmark. Conroe has higher IPC.


On floating point, they're getting close together in terms of the
benchmark result, but still have that wide difference in core clock speed.
So the performance ratio is still around 3.2/2.133 = 1.5x


In conclusion -

1) To a first order approximation, performance is proportional to -
    Core_clock * IPC

    Using Core_clock alone, is not sufficient to judge them. IPC
    varies from generation to generation.

2) If you compare two processors, that have identical internal architectures
    (so their IPCs are the same), you make their cores the same frequency,
    and one has FSB1066 and the other FSB1333, you'd be hard pressed to
    tell the difference. Maybe if the processor was a quad, and was
    starving for memory bandwidth, then it would matter. But otherwise,
    FSB is not that big a deal.

3) Memory speed matters. But the difference again, is not directly
    proportional. If you could somehow boost the memory bandwidth
    by 10%, the application performance might increase by 3%. Wasting
    a lot of money on RAM speed alone, is probably not going to make
    you a hero.

You really shouldn't take my word for it. Look for the SPEC
benchmarks, and check the tables of results there. With some
luck, maybe you can find results for those two different
generations of Xeons, and figure out the conversion factor
between core clocks based on those results. I don't consider
a Sandra benchmark, to be good enough to use for making
purchasing decisions.

Also, for specialized applications, sometimes they use things
like the SSE instructions. If the thing that "pays the bills"
for you, used something like that, then you'd need an even
more obscure benchmark, to figure out the best choice.


Re: CPU Speed vs. FSB speed

Hi  Paul

What an effort: so much information in one hit. Unfortunately, I only
managed to find a SPEC result for the newer 2 GHz 5130. But even those
suggest to me that speed increases do not create more processing power
linearly. Instead, the faster versions up from the 5130 kind of lose
their momentum, so to speak, as they get faster. So the same would
have been true for the earlier 5060.

I feel that the following more tangible inputs may sway me to the
slower, but newer 5130 CPU:
*    FSB speed, obviously, is at 1333 MHz, vs. 1066 MHz
*    Thermal Design Power is half at 65W for the 5130 (130W for 5060).
*    Bus/Core Ratio is 6 for the 5130, double that for the 5060)
While I am speculating here, I do think that TDP would pretty much
translate into power consumption, apart from its primary meaning of
the heat needing to be dispersed.
Equally speculative is the thought that the lesser Ratio would be an
advantage that translates into 'output', ie speed.

thanks again for your input.

Re: CPU Speed vs. FSB speed

Quoted text here. Click to load it

The driving force behind your decision may come from which memory speed you
choose.  The 5060 is a Dempsey processor, based on the Pentium IV design
(read flawed), while the 5130 is a Woodcrest processor based on the Core
design (an enhanced Pentium III (Coppermine)).  With a front side bus of
1067, you would want to get FB-DIMM 533 MHz memory (PC2-4200).  With the
Woodcrest and a front side bus of 1333, get FB-DIMM 667 MHz memory
(PC2-5300).  These memory choices will then be matched with your processor
to give best performance.

Also, get four identical modules at a time, to load all four channels of
FB-DIMM to get the benefit of dual-channel/dual-branch mode.  The population
rules would have the black slots populated first, then the white slots
later.  The black slots are the first slot in each channel.

In the end, I would recommend the Woodcrest processor over the Dempsey,
mainly because the Pentium III design supports higher IPC than Pentium IV,
and the Core enhancements push the IPC higher.

Site Timeline