• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

CPU "IPC"??? Low end CPU's can't match high end CPU's at the same speed??

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

Kunaak

Member
Joined
Sep 28, 2002
Location
Juneau Alaska
CPU "IPC"??? Low end CPU's can't match high end CPU's at the same speed??

I have been trying to learn more about the "IPC" of CPU's.

this is only something I have read about here, and on some other forums, where people have stated.

that a low end CPU, say a XP 1700, has a lower "IPC", then say a XP 2100, so when the 1700 is at 2ghz, and the 2100 is at 2 ghz.
the 2100 will still be faster, cause it has a higher IPC rating the the lower end XP 1700....

atleast thats what I understand of this.

this next part is pure speculation, so please bear with me.

we all know the AMD's do 9 calcuations per cycle or something like that.
but say, the xp 1700 does 10 instructions per calculation, that would be 10x9=90.
then say 12 instructions per calculation, that would be 12x9=108.

thats just pure speculation for me personally to explain what the "IPC" may be, and how it may affect the CPU.

of course that part is simply speculation, cause I don't really know...
but it helps me understand, why people say, a lower end CPU will not beat a higher end CPU, even when at the same speed.

can anyone, shed some light on this?

I really want to learn about this in more detail.
just, after a few hours on the net, didn't really help to shed any light on the subject, and although I emaild AMD, they were of little help.
 
Re: CPU "IPC"??? Low end CPU's can't match high end CPU's at the same speed??

Kunaak said:
we all know the AMD's do 9 calcuations per cycle or something like that.

Case closed.

calculations = instructions
per = per
cycle = clock cycle

IPC

All AMD K7 CPUs have the same IPC, varies depending on what your chip is doing, but it'll be the same.

Now, it does not mean that a Barton at 600Mhz (6x100) performs the same as a Duron at 600Mhz, as the Barton would support SSE and also would have 8x the amount of cache. But it does mean that they perform the same amount of instructions per clock cycle. What makes a Barton 3000+ faster than a Duron 600 is because it can successfully complete more cycles in a given amount of time, but each of those cycles it gets the same amount of work done (excluding SSE optimizations...SSE optimizations basically combine SPECIFIC multiple instructions into one, so it may seem that the CPU gets more done, when in reality it is doing the same amount of work, just that it supports more efficient instructions.).

The bottom line is that anyone who tells you "Oh my 2100+ at 2.4Ghz outperforms your 1700+ at 2.4Ghz" is tragically misinformed. 1700, 2100, 2400, 1800, 2600, all of those chips will perform exactly the same, provided that the clock speed, Mhz, and voltage are all constant.

A 2600+ at 220x11, or something, on the same board, in the same system, with the same amount of voltage, will perform exactly the same as a 1700+ at 220x11, with the same voltage, on the same board, in the same system. (Please don't get technical with me about the amount of current that, say, 1700B DUT3C's draw compared to 1700B DLT3Cs. Technically, I believe that with more voltage, a chip will take a slight impact on performance, but it is nothing that would reflect in any benchmark.)

AMD has one architecture for Thoroughbred chips, it is an 0.13 die process, 256K cache manufacturing process. AMD creates processors on wafers, and one wafer contains many different Thoroughbred chips, all rated at many different speeds. There is no "XP2800 wafer", "XP2700 wafer" etc. Usually chips in the center are rated higher and chips towards the outer edges are rated lower; however the design process remains exactly the same.

T-Bred As, T-Bred Bs, both kinds, all perform the same at the same Mhz/voltage in the same system. Hell...even Palominos probably would perform the same, being that it has the same external instruction support, same amount/speed of cache, and a T-Bred is essentially a Palomino with an IHS and a smaller die process to allow higher scaling. The only difference between chips is their ability to scale to certain clock speeds, and even that is proving to be false. Some 1700s are doing just as good, if not better, than 2100s, and 2100s are doing just as good if not better than 2400s.

A number on a chip, is just a number. It really doesn't mean anything if you're a hardcore overclocker.
 
Last edited:
Mark Larson said:
can overclock in peace knowing your CPU has more IPC than a P4 ever will. :D

ROFL...thx...just trying to clear up the BS that people are flinging around these forums. (I am not referring to you Kunaak...you were just getting a question answered)

I have been on a lot of hardware forums and this is the first one where people think that their 2100s have a different architecture than someone else's 1700+ :eek:
 
Re: CPU "IPC"??? Low end CPU's can't match high end CPU's at the same speed??

Kunaak said:
we all know the AMD's do 9 calcuations per cycle or something like that.

Don't confuse that number with IPC. AMD says the Athlon can do 9 "operations" per cycle, though they never really define their operations. I suspect it means the chip has 9 execution units.

However, both an Athlon and a P4 can do between 1 and 3 IPC depending on the kind of work being done. Actual IPC is highly dependent on the software you're running.
 
If a 1700 & 2100 both hit the same limit say the same as a 2400, how will that affect the multiplier on both CPUs? I think the multiplier has something to do with fsb/pci ratios to keep agp/pci in spec 66/33.

What I'm asking is will the 1700 need higher multipliers than the 2100 if both run at the same speed?
 
Audioaficionado said:
If a 1700 & 2100 both hit the same limit say the same as a 2400, how will that affect the multiplier on both CPUs? I think the multiplier has something to do with fsb/pci ratios to keep agp/pci in spec 66/33.

What I'm asking is will the 1700 need higher multipliers than the 2100 if both run at the same speed?

You would either need to raise the FSB or the Mult (or both). When you are talking about ocing though there are more factors involved than just the cpu so I am not sure you question can be exactly answered. 1 person may achieve their overclock by raising their FSB while another raises their mult and the third may raise FSB alot and actually lower their mult.
 
Let me restate:

If you have both the 1700 ond 2100 at the same GHz, will the 1700 need to run a higher fsb?

The multipliers are to keep you in 33/66 pci/agp specs with your chosen fsb, right?

Or do they affect the ratio between the CPU GHz/fsb?
 
Last edited:
Audioaficionado said:
Let me restate:

If you have both the 1700 ond 2100 at the same GHz, will the 1700 need to run a higher fsb?

The multipliers are to keep you in 33/66 pci/agp specs with your chosen fsb, right?

Or do they affect the ratio between the CPU GHz/fsb?

The multipliers are there to keep your CPU running higher than your FSB :p obviously we can't run 2000Mhz FSB, nor do we want 133Mhz CPUs, so a multiplier is built into the subsystem to make the CPU run at XX times whatever the FSB is.

The 1700+ will only need to run a higher FSB if the multiplier is locked, but no 1700+ T-Bred has locked multipliers, not that I know of anyway...
 
So a 1700 running a 16 multiplier at say 2.4 GHz will be essentualy the same as a 2100 doing the same?
 
Then that 1700 just moved into first place on my buy list.

If guys run 133 fsb out to 200, they either have a way out of spec pci bus or they have a mobo that locks the pci @ 33, right?

What does RAM run off of?
 
Audioaficionado said:
Then that 1700 just moved into first place on my buy list.

If guys run 133 fsb out to 200, they either have a way out of spec pci bus or they have a mobo that locks the pci @ 33, right?

What does RAM run off of?

In mobos without PCI locks, the PCI bus will run at some fraction of the FSB speed. To keep the PCI bus in spec at 200 FSB, you'll need a 1/6 divisor ( 200/6=33.3) as a 1/5 divisor would seriously overclock the PCI bus (200/5=40.0).

Depending on the motherboard, RAM speed can be locked at a given frequency, or it can scale with the FSB, usually at a 1:1 ratio. So when the FSB is at 200MHz, the RAM runs at 200MHz also. Some motherboards have other RAM ratios available, such as 4:5 (133Fsb:166Memory) or 4:6 (133FSB:200Memory).
 
Audioaficionado said:
Then that 1700 just moved into first place on my buy list.

If guys run 133 fsb out to 200, they either have a way out of spec pci bus or they have a mobo that locks the pci @ 33, right?

What does RAM run off of?

Exactly, but most likely it's the latter.

Two chips at the same voltage/FSB/multiplier will perform the same, regardless of their rating. It doesn't mean that all chips will even RUN at a certain setting, but if they do, they will perform the same.

The only difference between a 1700+ and a 2100+ is that AMD has certified the chip for operation at 1733Mhz. Now the 1700s can probably also be certified as 2100s, or even 2400 or 2600s, but AMD has to fill a large demand for 1700s.

More people want cheaper chips than want expensive ones, so if they have 5000 chips, and 4900 of them COULD be rated at 2400+, they aren't going to mark 4900 of them 2400+ and above because then the 1700+ supply would dry up. It costs them the same amount to produce a 2400+ as it does a 1700+, so they aren't taking a loss, they are just making less of a profit, which happens very often in business.
 
The miltiplier sets the ratio between the CPU and Front Side Bus.

If you want faster raw memory speeds (discounting latency and timings for now), you have to have a faster fsb.

It's the divisor that keeps the ratio between the FSB and PCI bus so that the PCI is ~33MHz. So you need 5 or 6+ divisor mobos to have the most options.

I wonder how high this MSI mobo's divisor goes?
 
That board probably has a 1/4 PCI divider and a 1/2 AGP divider, which won't get you very far most likely.

Go for a board based on the nForce2 chipset. I like the Abit NF7-S, Epox 8RDA+ or 8RGA+, and Epox 8RDA3+ when it comes out...
 
The only easy way to reduce IPC would be to disable parts of the L2 cache. I doubt AMD would have done this -- and if they did it must have been isolated to a few instances. By chance someone could obtain a 1700+ @ 2.3GHz with partially failed L2 units and it would perform fewer instructions per clock than a fully functional 2100+ at the same speed because of more branch misprediction. But isolated incidents are far different from consistent results, and rumors need to be quelled instantly.

It would be highly unlikely that an athlon chip would work at all with partially failed L2 cache or would be somewhat dodgy whenever the chip would attempt to write/read in the failed area.

I highly doubt AMD would pull a disabling thing on the sneak -- it would be false advertising. All Palomino and tbred chips are advertised as having 256L2 - no less, no more. In the future AMD might sell bartons with half-failed L2 as tbreds, but these would be fully functional as the chip would not try to read/write in the failed area.

In any case, 1700+ and 2100+ CPUs are relatively cheap, and can easily be replaced under warranty. It's not all the time that one can buy $68 chips that can be brought to performance levels of chips six times more expensive and beyond.
 
Back