You are not logged in.

#1 2005-04-11 18:46:09

phrakture
Arch Overlord
From: behind you
Registered: 2003-10-29
Posts: 7,879
Website

MHz vs Cache - CPUs and Geforces galore!

Ok, my dilemma right now:
Pentium 4: 3.0GHz, 2MB Cache - $225
Pentium 4: 3.2GHz, 1MB Cache - $195

Is an extra MB of Cache worth 200MHz? Anyone know?
Right now I'm leaning towards the $195 one, due to price mainly... but I don't know how much of a performance boost the cache will cause...

Any pointers would be helpful

Offline

#2 2005-04-11 19:02:02

phrakture
Arch Overlord
From: behind you
Registered: 2003-10-29
Posts: 7,879
Website

Re: MHz vs Cache - CPUs and Geforces galore!

And while I'm at it - seeing as these are the last 2 pieces remaining for my uber-box... some opinions on the following choices would be nice:
Geforce 6800 Ultra ~$450+
Geforce 6800 GT ~$350
Geforce 6800 ~$230
Do the minor differences make these things worth the extra (to put it lightly) cost?

differences can be seen here: http://www.guru3d.com/article/content/151/

Offline

#3 2005-04-11 20:43:10

rose
Member
Registered: 2005-02-09
Posts: 64

Re: MHz vs Cache - CPUs and Geforces galore!

Pentium 4: 3.0GHz, 2MB Cache - $225
Pentium 4: 3.2GHz, 1MB Cache - $195

I'd go for $225 one. I cannot give you exact numbers, it depends on what you are going to do with the CPU. However, cache is always precious. Take into account the following numbers:

L1 cache adressing: 2 clocks
L2 cache adressing: 20 clocks
RAM cache adressing: ~500 clocks
(this means that random access to ram is not faster than on 386! Intel is cheating here...)

So, if your program is going to access 1-2 MB of data randomly, then the difference will be huge. However, usually you don't access so much memory randomly.

Btw., I remember times when RAM was as fast as L2 (L3?)  cache is now.

Offline

#4 2005-04-11 20:46:27

phrakture
Arch Overlord
From: behind you
Registered: 2003-10-29
Posts: 7,879
Website

Re: MHz vs Cache - CPUs and Geforces galore!

rose wrote:

L1 cache adressing: 2 clocks
L2 cache adressing: 20 clocks
RAM cache adressing: ~500 clocks
(this means that random access to ram is not faster than on 386! Intel is cheating here...)

yeah well, you never directly access RAM... that takes into account page swapping.. but that's a whole different ballgame.

I know there's a difference, but it is going to be enough to warrant the extra cost compared to the other one?

rose wrote:

Btw., I remember times when RAM was as fast as L2 (L3?)  cache is now.

yeah but that's measured in # of clock cycles... which is alot faster now-a-days 8) Still, you have a good point... cache is just evolving faster than RAM

Offline

#5 2005-04-11 21:11:15

rose
Member
Registered: 2005-02-09
Posts: 64

Re: MHz vs Cache - CPUs and Geforces galore!

Sorry! I exagerated! Just run this program:

#include <stdio.h>
#include <stdlib.h>
#include <time.h>

char *buf;
int size;
volatile int vi;

double MAX = 0;

void test(int v) {
    int t = time(0);
    printf("Testing with blocksize=%dKB: ", 1<<(v-10));
    fflush(stdout);
    buf = (char*)malloc(1<<v);
    while (time(0) == t);
    t = time(0);
    
    int pos = 0;
    int step = 37*11*4095*4095;
    int n = 0;
    int j;
    while (time(0) != t + 1) {
        for (j = 0; j < 1000; ++j) {
            vi = buf[pos];
            pos += step;
            pos &= (1<<v)-1;
        }
        pos += rand();
        pos &= (1<<v)-1;
        n+=1000;
    }
    if (MAX < n) MAX = n;
    printf("%d iterations/second (%lfx slowdown)n", n, MAX/n);
    free(buf);
}

int main() {
    int i;
    for (i = 10; i < 26; ++i)
        test(i);
    return 0;
}

(fixed so that it compiles with gcc, not olny with g++)

It just reads bytes in more or less random fashion in a large array. I get the following results:

Testing with blocksize=1KB: 499005000 iterations/second (1.000000x slowdown)
Testing with blocksize=2KB: 500133000 iterations/second (1.000000x slowdown)
Testing with blocksize=4KB: 499317000 iterations/second (1.001634x slowdown)
Testing with blocksize=8KB: 496881000 iterations/second (1.006545x slowdown)
Testing with blocksize=16KB: 438288000 iterations/second (1.141106x slowdown)
Testing with blocksize=32KB: 431539000 iterations/second (1.158952x slowdown)
Testing with blocksize=64KB: 424596000 iterations/second (1.177903x slowdown)
Testing with blocksize=128KB: 420950000 iterations/second (1.188105x slowdown)
Testing with blocksize=256KB: 149170000 iterations/second (3.352772x slowdown)
Testing with blocksize=512KB: 43777000 iterations/second (11.424561x slowdown)
Testing with blocksize=1024KB: 39650000 iterations/second (12.613695x slowdown)
Testing with blocksize=2048KB: 40285000 iterations/second (12.414869x slowdown)
Testing with blocksize=4096KB: 41931000 iterations/second (11.927524x slowdown)
Testing with blocksize=8192KB: 47252000 iterations/second (10.584377x slowdown)
Testing with blocksize=16384KB: 49756000 iterations/second (10.051712x slowdown)
Testing with blocksize=32768KB: 48074000 iterations/second (10.403399x slowdown)

The situation is even worse: when access to the memory is fast (<512KB on my CPU), the main cost of the loop is the execution of adding/anding opcodes.

When blocksize=1024KB one iteration takes 61 = 2400/39 clocks (I have P4 2.4 GHz).
This clocks is mostly waiting for memory access.
Well, I didn't pay for a 39 MHz CPU!
(sorry about those 500x numbers; it's at most 50-100x).

Offline

#6 2005-04-11 21:15:28

rose
Member
Registered: 2005-02-09
Posts: 64

Re: MHz vs Cache - CPUs and Geforces galore!

Compile the program with "gcc -O3 test.c -o test" and tell me your results.

Offline

#7 2005-04-11 21:17:54

phrakture
Arch Overlord
From: behind you
Registered: 2003-10-29
Posts: 7,879
Website

Re: MHz vs Cache - CPUs and Geforces galore!

when I get home - not at my arch box now...

but another thing is that it's normal for IO to take a long time... it's the IO scheduler that will switch things around so other operations are done while waiting for ram.... try doing the same with with disc access, it won't be pretty....

the kernel does a fairly good job of optimizing this stuff while waiting for IO to return...

Offline

#8 2005-04-11 21:21:02

rose
Member
Registered: 2005-02-09
Posts: 64

Re: MHz vs Cache - CPUs and Geforces galore!

phrakture wrote:

when I get home - not at my arch box now...

but another thing is that it's normal for IO to take a long time... it's the IO scheduler that will switch things around so other operations are done while waiting for ram.... try doing the same with with disc access, it won't be pretty....

the kernel does a fairly good job of optimizing this stuff while waiting for IO to return...

No! We are talking about a delay which occurs every 61 clocks. Kernel doesn't switch tasks so fast. The CPU just waits doing nothing (or does something if you have HyperThreading (tm), but as someone pointed out this sollution effectively halves your cache). It isn't normal IO (like IDE IO). Tasks can be switched only every .001s, not every 61 ticks.

Again, this is not the delay to access hard disk. It is the delay to access core RAM, which is MUCH lower, but also MUCH higher than some people expect.

Offline

#9 2005-04-11 23:10:19

MNKyDeth
Member
From: MI
Registered: 2003-09-13
Posts: 89

Re: MHz vs Cache - CPUs and Geforces galore!

If I were you on the vid card choice, I would opt for the 6800GT. basically it is just a slower clocked ultra but if you get a good card should be able to OC to an ultra if so desired. The GT and Ultra use 16pipelines I beleive and the standard 6800 only uses 12. I could be wrong on this but I think that is what i read over on anandtech.com a while back. Might want to check out anandtech.com and find a review of the cards to see wich will best fit your setup.


I thought it interesting the test rose put up. I ran just to see what i would get on my A64 3400+ socket 754 1MB cache cpu. Here are my results.

Btw, I used the default script Rose posted with this settings, no fancy opts for my cpu just the default setup as posted.
gcc -O3 test.c -o test

Testing with blocksize=1KB: 664345000 iterations/second (1.000000x slowdown)
Testing with blocksize=2KB: 664789000 iterations/second (1.000000x slowdown)
Testing with blocksize=4KB: 664412000 iterations/second (1.000567x slowdown)
Testing with blocksize=8KB: 664577000 iterations/second (1.000319x slowdown)
Testing with blocksize=16KB: 664499000 iterations/second (1.000436x slowdown)
Testing with blocksize=32KB: 156821000 iterations/second (4.239158x slowdown)
Testing with blocksize=64KB: 106395000 iterations/second (6.248311x slowdown)
Testing with blocksize=128KB: 93442000 iterations/second (7.114456x slowdown)
Testing with blocksize=256KB: 84870000 iterations/second (7.833027x slowdown)
Testing with blocksize=512KB: 84481000 iterations/second (7.869095x slowdown)
Testing with blocksize=1024KB: 84314000 iterations/second (7.884681x slowdown)
Testing with blocksize=2048KB: 83436000 iterations/second (7.967652x slowdown)
Testing with blocksize=4096KB: 60302000 iterations/second (11.024328x slowdown)
Testing with blocksize=8192KB: 56091000 iterations/second (11.851973x slowdown)
Testing with blocksize=16384KB: 50867000 iterations/second (13.069161x slowdown)
Testing with blocksize=32768KB: 50135000 iterations/second (13.259978x slowdown

Offline

#10 2005-04-12 01:35:26

aCoder
Member
From: Medina, OH
Registered: 2004-03-07
Posts: 359
Website

Re: MHz vs Cache - CPUs and Geforces galore!

I agree that the 6800GT is the best buy for a high end video card right now.  Although I don't know much about CPUs, I do know, from experience, that there tends to be a much more noticable performance increase when cache size is increased than when clock speed is increased.


If you develop an ear for sounds that are musical it is like developing an ego. You begin to refuse sounds that are not musical and that way cut yourself off from a good deal of experience.
  - John Cage

Offline

#11 2005-04-12 15:12:04

phrakture
Arch Overlord
From: behind you
Registered: 2003-10-29
Posts: 7,879
Website

Re: MHz vs Cache - CPUs and Geforces galore!

aCoder wrote:

I agree that the 6800GT is the best buy for a high end video card right now.  Although I don't know much about CPUs, I do know, from experience, that there tends to be a much more noticable performance increase when cache size is increased than when clock speed is increased.

Yeah, I was thinking about the GT vs Ultra... according to the specs there's like a 50MHz ram clock difference... for $100 less, that's fine by me (and I can always OC it

About the mhz/cache thing, I think I'm going to go larger cache - 100% increase in cache size vs a ~7% increase in CPU clock speed... seems obvious when you think of it that way...

Offline

#12 2005-04-13 15:59:15

RedShift
Member
From: Belgium
Registered: 2004-07-16
Posts: 230

Re: MHz vs Cache - CPUs and Geforces galore!

My results:

build with the following:
[glenn@polaris build]$ gcc -march=k8 -mtune=athlon64 -O2 test.c -o test

[glenn@polaris build]$ ./test
Testing with blocksize=1KB: 172670000 iterations/second (1.000000x slowdown)
Testing with blocksize=2KB: 170533000 iterations/second (1.012531x slowdown)
Testing with blocksize=4KB: 172971000 iterations/second (1.000000x slowdown)
Testing with blocksize=8KB: 173047000 iterations/second (1.000000x slowdown)
Testing with blocksize=16KB: 173141000 iterations/second (1.000000x slowdown)
Testing with blocksize=32KB: 93707000 iterations/second (1.847685x slowdown)
Testing with blocksize=64KB: 72986000 iterations/second (2.372249x slowdown)
Testing with blocksize=128KB: 66548000 iterations/second (2.601746x slowdown)
Testing with blocksize=256KB: 61851000 iterations/second (2.799324x slowdown)
Testing with blocksize=512KB: 61665000 iterations/second (2.807768x slowdown)
Testing with blocksize=1024KB: 61548000 iterations/second (2.813105x slowdown)
Testing with blocksize=2048KB: 61111000 iterations/second (2.833222x slowdown)
Testing with blocksize=4096KB: 47696000 iterations/second (3.630095x slowdown)
Testing with blocksize=8192KB: 44928000 iterations/second (3.853744x slowdown)
Testing with blocksize=16384KB: 41402000 iterations/second (4.181948x slowdown)
Testing with blocksize=32768KB: 40753000 iterations/second (4.248546x slowdown)
[glenn@polaris build]$

AMD Athlon64 3500+ with 512 kB of L1 cache. Do note I was running alot of programs in the background...

Build with the following:
[glenn@polaris build]$ gcc -march=k8 -mtune=athlon64 -O3 test.c -o test

[glenn@polaris build]$ ./test
Testing with blocksize=1KB: 173987000 iterations/second (1.000000x slowdown)
Testing with blocksize=2KB: 174025000 iterations/second (1.000000x slowdown)
Testing with blocksize=4KB: 174032000 iterations/second (1.000000x slowdown)
Testing with blocksize=8KB: 173973000 iterations/second (1.000339x slowdown)
Testing with blocksize=16KB: 173764000 iterations/second (1.001542x slowdown)
Testing with blocksize=32KB: 94104000 iterations/second (1.849358x slowdown)
Testing with blocksize=64KB: 73390000 iterations/second (2.371331x slowdown)
Testing with blocksize=128KB: 66942000 iterations/second (2.599743x slowdown)
Testing with blocksize=256KB: 62306000 iterations/second (2.793182x slowdown)
Testing with blocksize=512KB: 62087000 iterations/second (2.803034x slowdown)
Testing with blocksize=1024KB: 61977000 iterations/second (2.808009x slowdown)
Testing with blocksize=2048KB: 61552000 iterations/second (2.827398x slowdown)
Testing with blocksize=4096KB: 47961000 iterations/second (3.628615x slowdown)
Testing with blocksize=8192KB: 45253000 iterations/second (3.845756x slowdown)
Testing with blocksize=16384KB: 41262000 iterations/second (4.217731x slowdown)
Testing with blocksize=32768KB: 41253000 iterations/second (4.218651x slowdown)

Optimalization level 3 seems to be a tad faster.

As a comparison, the following on a dual pentium III 1 Ghz, each 256 kB L1 cache:
[glenn@galaxy build]$ gcc -march=i686 -mtune=pentium3 -O2 test.c -o test

[glenn@galaxy build]$ ./test
Testing with blocksize=1KB: 176959000 iterations/second (1.000000x slowdown)
Testing with blocksize=2KB: 177018000 iterations/second (1.000000x slowdown)
Testing with blocksize=4KB: 176988000 iterations/second (1.000170x slowdown)
Testing with blocksize=8KB: 177018000 iterations/second (1.000000x slowdown)
Testing with blocksize=16KB: 175503000 iterations/second (1.008632x slowdown)
Testing with blocksize=32KB: 175669000 iterations/second (1.007679x slowdown)
Testing with blocksize=64KB: 176321000 iterations/second (1.003953x slowdown)
Testing with blocksize=128KB: 176464000 iterations/second (1.003139x slowdown)
Testing with blocksize=256KB: 176079000 iterations/second (1.005333x slowdown)
Testing with blocksize=512KB: 130435000 iterations/second (1.357136x slowdown)
Testing with blocksize=1024KB: 109448000 iterations/second (1.617371x slowdown)
Testing with blocksize=2048KB: 109881000 iterations/second (1.610997x slowdown)
Testing with blocksize=4096KB: 109716000 iterations/second (1.613420x slowdown)
Testing with blocksize=8192KB: 100267000 iterations/second (1.765466x slowdown)
Testing with blocksize=16384KB: 73404000 iterations/second (2.411558x slowdown)
Testing with blocksize=32768KB: 57420000 iterations/second (3.082863x slowdown)

I'm getting disturbing results here, the dual Pentium III 1 Ghz seems to perform more iterations than the Athlon64 3500+. Conclusion: this test is b0rken.


:?

Offline

#13 2005-04-13 17:06:18

phrakture
Arch Overlord
From: behind you
Registered: 2003-10-29
Posts: 7,879
Website

Re: MHz vs Cache - CPUs and Geforces galore!

RedShift wrote:

I'm getting disturbing results here, the dual Pentium III 1 Ghz seems to perform more iterations than the Athlon64 3500+. Conclusion: this test is b0rken.

you can't say that... it's kinda like saying a basketball player is taller therefore he should get more rebounds.... it's speculation based on numbers.

It's possible that the dual cache/mem bus/proc combo accesses memory much faster than a single processor... and it kinda makes sense...

This test has nothing to do with clock speed, it's memory access...

Offline

#14 2005-04-13 17:12:27

RedShift
Member
From: Belgium
Registered: 2004-07-16
Posts: 230

Re: MHz vs Cache - CPUs and Geforces galore!

phrakture wrote:
RedShift wrote:

I'm getting disturbing results here, the dual Pentium III 1 Ghz seems to perform more iterations than the Athlon64 3500+. Conclusion: this test is b0rken.

you can't say that... it's kinda like saying a basketball player is taller therefore he should get more rebounds.... it's speculation based on numbers.

It's possible that the dual cache/mem bus/proc combo accesses memory much faster than a single processor... and it kinda makes sense...

This test has nothing to do with clock speed, it's memory access...

Yes but consider this:
The Athlon64 runs at 2200 Mhz, with integrated memory controller, DDR400 memory with 2-2-2-5 timings.
The dual Pentium III runs at 1000 Mhz, with an appollo pro 133 chipset, PC133 memory.

Both theoretical and in practice, the Athon64 is able to push far more data in less time than the dual P-III.


:?

Offline

#15 2005-04-13 17:23:42

MNKyDeth
Member
From: MI
Registered: 2003-09-13
Posts: 89

Re: MHz vs Cache - CPUs and Geforces galore!

I don't think the test is broken, What you are seeing is the impact of the IPC of the cpu in relation to the cache, I think. The P3 in theory was the only cpu Intel had that could match the Athlon or even A64 cpu's IPC wise. The P3 was a nice low latency cpu like the A64's and Athlons. The problem the P4's had was that technically there IPC went from 10 in the P3 to 20 in the P4. This made the instructions per second take longer from beggining to end. Roughly a P3 1ghz could keep pace just fine with the original P4 @ 1.4ghz without falling behind.

When Intel moved to the Prescott they they increased the IPC again to 30, so now it takes even longer than the Northwood P4's. That's why you see lots of benches on various sites were a 2.8 - 3.0ghz P4 Northwood can keep pace with a 3.2 -3.4ghz P4 Prescott. Intels thinking was to increase mhz speed and not to execute the data efficiently wich has bit them in the butt a couple times now. The flaw, is, Intel at the stage of Northwood, the Netburst architechure reached it's limits. The prescott imo was an attempt to try and push it further wich failed miserably. Also, Intel is only using half of an FPU unit on there P4's. There P3 had a full FPU like Althlon and A64's. What Intel did to make up for this reduction in transistors was to make new optimizations for only a half FPU, like SSE2 and SSE3 an so on and so forth. This is why in any benchmark that relies on FPU and is not optimized for any specific instructions like SSE2 or SSE3 the AMD chips now pummel Intel chips.

So, in theory, since Intel cpu's have longer gates (IPC) and are starved from there bandwidth perspective because they use hyperthreading to imitate a duall cpu setup, (kinda), they need mass amounts of bandwith from the front side buss and very large cahce's to perform imo "efficiently". A fast mhz cpu that waits for information is going to be slower (possibly) than a cpu that is slower but does not have to wait for it's information. This is accurate to a point though, each cpu is made different so each architechure has it's advantages.

Like, some programs rely on mass mhz to perform well or good. Some rely on optimizations. Some rely on how fast it gets through the cpu. Each program and how it was programed will perform different on a different cpu so it's very hard to tell wich cpu is actually the best overall in this sense. Basically, you need to do your research and find out how each cpu performs for the tasks you are most going to ask it to do.

Sorry for my rant, this is not an AMD vs Intel flameware, I just thought it would be good to help inform some of those that don't exactly get what is going on inside there cpu's and I hope I gave a good enough explanation for the above posters results.

Offline

#16 2005-04-13 17:42:48

phrakture
Arch Overlord
From: behind you
Registered: 2003-10-29
Posts: 7,879
Website

Re: MHz vs Cache - CPUs and Geforces galore!

RedShift wrote:

Yes but consider this:
The Athlon64 runs at 2200 Mhz, with integrated memory controller, DDR400 memory with 2-2-2-5 timings.
The dual Pentium III runs at 1000 Mhz, with an appollo pro 133 chipset, PC133 memory.

There's so many more variables than that:
pipeline size, bus speed, cache levels, and a hundred more in the memory control unit... it's too much to say "setup X's memory should perform better than setup Y's" - yeah the numbers *should* tell you something... but you're also talking dual procs vs a single proc (I've seen dual 800MHz procs outperform a 3GHz).

It's just too much to positively prove the test is broken... it's not proof... just because these theoretical numbers tell you what "should" happen, doesn't mean it will, and it doesn't mean something is broken if it doesn't happen.

Offline

#17 2005-04-13 17:52:57

MNKyDeth
Member
From: MI
Registered: 2003-09-13
Posts: 89

Re: MHz vs Cache - CPUs and Geforces galore!

I think my last post pretty much summed it up, but in the case of the dual cpu vs A64 I beleive this is still the IPC that you are seeing, not the dual core exactly.

The P3 with IPC of 10 vs and Athlon with IPC of 11 I beleive or even vs a A64 with an IPC of 12 (again I beleive, or think), should be able to transfer from it's cache faster than the Althon or the A64. The point I think is, is that this is a cache becnmark no?

It's not really using the system buss or physical memory unit it goes over the actual cache size. In this respect, again, the P3 was a very low latency designed architechture. There really havn't been any architectures with it's "refinedness" in this sense. The BX chipset was a very low latnecy chipset aswell. Now think for one second, since you have 2 P3's with 256k L2 cache on them, in theory this would total 512k like on your A64 so you don't see the penalty like you normally would I think. Only way to be exactly for sure would be to diasble the 2nd cpu and run the test again and see if the reults vary greatly.

There are many more factors why an A64 can cream a P3 in the bandwidth department, but the reults do speak mounds for the P3 architechure and it's incredible low latency.

Offline

#18 2005-04-13 20:29:55

rose
Member
Registered: 2005-02-09
Posts: 64

Re: MHz vs Cache - CPUs and Geforces galore!

RedShift, the test isn't broken. Don't run too much background processes. Don't compile with -O2 (it doesn't make sense here - -O3 is the best suited level here).
Is everything alright with your Athlon 3400+? It seems to be nearly 3 times slower than my P4 2400 MHz in the 1KB test, when the only factor is the cpu speed (everything fits in cache). I can't believe this.


One thing I find good about my P4 is that the L2 cache is quite efficient. The test showed only negligible performance degradation until blocksize >= cachesize. On AMD procs the performance degradation comes at much lower blocksize, which is very very very bad.

Offline

#19 2005-04-13 20:34:52

rose
Member
Registered: 2005-02-09
Posts: 64

Re: MHz vs Cache - CPUs and Geforces galore!

RedShift wrote:

AMD Athlon64 3500+ with 512 kB of L1 cache.

REALLY??? SO BIG? Original Pentium had 8KB L1 data cache, Pentium MMX had 16KB L1 data cache, Pentium 4 has 8KB L1 data cache.
I can't believe AMD engineers stuffed 512 kB of L1 cache. I suspect you ment L2 (or L3?) cache?

Offline

#20 2005-04-13 20:44:47

cactus
Taco Eater
From: t͈̫̹ͨa͖͕͎̱͈ͨ͆ć̥̖̝o̫̫̼s͈̭̱̞͍̃!̰
Registered: 2004-05-25
Posts: 4,615
Website

Re: MHz vs Cache - CPUs and Geforces galore!

yup. I imagine. Intel is especially fond of l2 lately..
http://www.intel.com/products/processor_number/info.htm


"Be conservative in what you send; be liberal in what you accept." -- Postel's Law
"tacos" -- Cactus' Law
"t̥͍͎̪̪͗a̴̻̩͈͚ͨc̠o̩̙͈ͫͅs͙͎̙͊ ͔͇̫̜t͎̳̀a̜̞̗ͩc̗͍͚o̲̯̿s̖̣̤̙͌ ̖̜̈ț̰̫͓ạ̪͖̳c̲͎͕̰̯̃̈o͉ͅs̪ͪ ̜̻̖̜͕" -- -̖͚̫̙̓-̺̠͇ͤ̃ ̜̪̜ͯZ͔̗̭̞ͪA̝͈̙͖̩L͉̠̺͓G̙̞̦͖O̳̗͍

Offline

#21 2005-04-13 21:08:37

RedShift
Member
From: Belgium
Registered: 2004-07-16
Posts: 230

Re: MHz vs Cache - CPUs and Geforces galore!

rose wrote:
RedShift wrote:

AMD Athlon64 3500+ with 512 kB of L1 cache.

REALLY??? SO BIG? Original Pentium had 8KB L1 data cache, Pentium MMX had 16KB L1 data cache, Pentium 4 has 8KB L1 data cache.
I can't believe AMD engineers stuffed 512 kB of L1 cache. I suspect you ment L2 (or L3?) cache?

My mistake, 512 kB of L2, ofcourse sad


:?

Offline

#22 2005-04-14 20:24:13

fuse
Member
From: california
Registered: 2004-04-11
Posts: 38

Re: MHz vs Cache - CPUs and Geforces galore!

I believe the main difference of the Ultra vs GT is the ram 1.6ns vs 2.0ns and the ultra has two molex connectors.

Offline

Board footer

Powered by FluxBB