L1 Cache is all you need!!!

[Win_98]

so who care about level 2, level 3 etc right?
since they are so slow that do nothing but hiner the performance of level 1 cache.
know what i mean, level 1 cache is slowed down to the same speed as that of level 2
let say you have 256K level 1 cache
this would blow away athlon 128 L1 and 256K level2 right?!
so the truth in this built in level 2 cache on motherboard is ****
because it like 1/10 speed of level 1 cache
this is why I see no speed difference when disabled.
a slight increase in memory performance but nothing else.

[CMonster]

You see no difference in speed because you are not testing with an application that has data or instructions that are being cached, or that exceed the capacity of the L1 cache:

http://webopedia.internet.com/TERM/c/cache.html

As for the current Athlon processors: They have 128K L1 running at full CPU speed and 512K L2 running at 1/2 or 1/3 CPU speed (depends on programming)- which on a 600mhz CPU would still make the L2 cache 3x faster than using the slower system memory for caching programs or data - right?



[This message has been edited by CMonster (edited 05-21-2000).]

[RobRich]

CMonster and Killerbug are right on this one. If you want to see how L2 becomes extremely important, do some memory throughput or disk i/o benchmarks with it turned off. As for L3 config's, check out ther server performance ratings of a k6-3. or a DEC Alpha. Their use of a three layer cahing architecture revolutionized that server market industry. Even the old k6-3 450 can still outperform the p3 500 in most server orientated tasks. Never disable your L2, unless you really want a slow performing system.

Also the reason for multiple caches lies within memory buffering. The L1 cache buffers only used cpu instructions. The L2 cache buffers not only the used cpu instructions, but also certain memory addressing data.

Robert Richmond

[Konan555]

I turned off my L2 cache out of interest, just to do a q2 timedemo.

I gave up in the end - after 20 mins NT 5 had only got half way throught booting.

[wtp]

Win, u are missing the point, they wouldn't have added the L2 cache if it wasn't important. The L2 is extremely important as it makes your computer faster. Compare a CPU without a L2 with one that doesn't have one, and uses onboard L2. There is a major difference. Just remember, though the L1 is more important, the L2 is just as important, as it helps your CPU perform to it's maximum capabilities. http://sysopt.earthweb.com/forum/smile.gif

wtp http://sysopt.earthweb.com/forum/smile.gif

[Win_98]

so you are saying that a slow 1meg onboard cache pipelined is better
then celeron's 128K level 1 cache if this is
true so many have made mistake buying celeron! http://sysopt.earthweb.com/forum/smile.gif http://sysopt.earthweb.com/forum/smile.gif http://sysopt.earthweb.com/forum/smile.gif
funny someone told me a puny 128K like celeron is way better then a huge 1meg cache
they also said dont matter how much 1meg, 2meg
and that it is ****
then again, having more cache actually improve memory performance by at least 50 percentage?

Im using a 350mhz BTW
Celeron maybe the most overclockable but
no 2nd level cache.
i dont know why anyone bought this at all
which we all know is something Intel
try to get a share of a market and succeed in doing so cutting K6-2 sale by at least
1/2

[Dash]

Win98, do some research or something.. man!

Celerons have 128k level 2! cache, not Level 1, Celerons have 32k Level 1.

The 1meg caches on motherboards run at BUS Speed, Celerons runs at Processor speed
So lets see, your 350 has a bus speed of what... 100mhz?, my Celeron... hmm, my cache is running at 600, yours is running at 100.

Only the original Celerons didnt have second level cache, 266 and 300, 300A beyond have it, thats all thats around now.
http://www.intel.com/design/celeron/index.htm

[RobRich]

The Celeron only has 32kb of L1 cache. It is true that the early Celeron's had no L2 cache, but all Celeron from 300a and up have a 128kb on-die L2 cache. The reason for the good performance of the L2 on a Celeron is that it runs at the chip's actual speed. So if you have a Celery at 550mhz, you L1 and L2 cache are both running at 550 as well. This also applies to Intel's Coppermine and Xeon cpu's.

If I remember, your 350 is a socket7 Cyrix cpu. The socket7 platform places the L2 cache on the motherboard, not the cpu (expect for k6-3). This means that the L2 is limited to running at the Front Side Bus speed, which is commonly 100mhz. This means that overall caching is severely limted in throughput. The k6-3 was the only socket7 cpu to overcome this issue. It has a 256kb on-die full speed L2 cache, and uses the motherboard's cache as L3. The lack of a high speed cache architecture is one of the larget reasons for the abandonment of the super7 platform.

Robert Richmond

[This message has been edited by RobRich (edited 05-21-2000).]

[wtp]

uhm.. u got it wrong win, i'm saying if the L2 cache is integrated with the CPU, it performs faster than depending on ONboard cache. And L2 and L3 would help, though minimize the use of RAM. But that's why the latency is changable http://sysopt.earthweb.com/forum/smile.gif (in some mobos)

wtp http://sysopt.earthweb.com/forum/smile.gif
RESEARCH IS ALWAYS KEY

[KillerBug]

L2 cache is extremely important, kill the L2 and you make the system go like it is 150mhz slower. L2 forever! Even if it is only 128k full speed, it is better than nothing. One concept, though...if on core L2 runs at the same speed as the chip, why not just make a huge L1 cache? and forget the L2 cache that is on the core.

[Andy_L]

Since you feel so strongly, Win_98, I encourage you to do 2 things, first, disable your L2 cache and leave it that way, second, never buy an Intel product. : )

[Win_98]

I will say this again.
having 128K cache that run at full speed
is better then 1meg that run at a slower speed?
so regardless of how much cache you got
say 2meg it is no good?!?
let see memory
you say 32meg run at full speed
is better then someone with 128meg that run at turtle speed hmm Im confused.

[wtp]

it's all about math, all if have to say is that running at full speed always wins. That was the problem with the Athlon. Yes high speed 64MB would be better than a slow speed 128MB stick of ram. Example, if you have a 64 people that are strong and smart against 128 people that are weak and stupid, who's more likely to win? http://sysopt.earthweb.com/forum/wink.gif well, you get it. http://sysopt.earthweb.com/forum/smile.gif

wtp http://sysopt.earthweb.com/forum/smile.gif

[AuraEdge]

..its not quite like that, but interesting comparason http://sysopt.earthweb.com/forum/smile.gif
On a Celeron 500 and 64MB of RAM for instance, u will have
32k of L1 (16k Data cache, 16k instruction cache)
128k of 500Mhz L2
65536k of 66Mhz RAM

Now if they app your using has TONS of stuff to cache..around 512k, it can stuff it into the RAM but not the L2, and can pull it out of the RAM, so it doesnt have to go searchin in the depths of the Harddrive.

However, if you have only 64k worth of stuff to cache, it can go into the L2, which would be retrieved theoretically 7.5x faster.

I dont really get the L1 as well as I should, so I dont want to comment on that...Its just...so confusing...

[wtp]

here, let me explain, again, the 128 people and 64 others. Though the 64 might be fast and smart and the 128 is slow and dumb, the 128 can hold more weapons than 64 people. If it comes to having a race on track (go to finish, and come back and get some more of those little sticks, and bring them back to the finish, just like a cycle) , the 64 people would win, and would be heading back while the 128 would still be reaching the finish line, but of course, when the 128 gets to the finish line, they'd have more stuff processed, the only difference is that the 64 people gets the job done faster. it's hard to calculate which is better, it's all about math, and this is where it gets complicated. It's just all about speed and distance, and that's what confuses me.

wtp http://sysopt.earthweb.com/forum/smile.gif

[FrozenLiquidity]

No L2 chache? Take a look at the first Celerons...

See what I mean?

FrozenLiquidity

[This message has been edited by FrozenLiquidity (edited 05-25-2000).]

[Win_98]

so this explain why the celeron have better memory performance then k6-2 or cyrix mii
I think you guy are mistaken about
K6-3 level 2 cache being run at full speed
it more like half speed.
the benchmark for it say 146 M/sec
at 450mhz.
The name of game is speed here and I doubt that most people really know much about this anyway or give a **** about it.
all they care is the Name Intel and all they know is that it better but the technical side
was never explained!
I have no ideas what full speed or half speed meant in the first place.
I guess I did learn a little so dont think
I am that hard headed.



[This message has been edited by Win_98 (edited 05-22-2000).]

[jad1097]

L2 cache support Augments L1 cache, making data and instruction access even faster 256 KB (full speed) Referring to the K6 3 processor.

The above quote was copied from http://www.amd.com/products/cpg/k6iii/compchart.html ANd a better explaition is here. http://www.amd.com/products/cpg/k6iii/trilevel.html


BTW; I did not intend for you to take what I said personal so please don't. I am just a bit demented and have a sick sense of humor.

[jad1097]

Cache Size Comparison: AMD-K6®-III vs. Pentium® III

K6 3 L3/ total k6 3/ P3/ /Advan
512KB /832KB /544KB /153%
1,024KB /1,344KB /544KB /247%
2,048KB /2,368KB /544KB /435%

http://www.amd.com/products/cpg/k6iii/trilevel.html

TriLevel Cache Design
Enhanced Cache Architecture for AMD-K6®-III Processor Enables Premium Performance with Industry's Largest Total System Cache for Windows® Compatible Desktop PCs

Overview
Largest Combined System Cache
Efficient Multiport Cache Design
Full-Speed Cache
Overview

New and enhanced software applications for the home, business, and the Internet continue to require higher levels of system performance. Photo- and video-editing software, Web design tools, soft DVD, voice recognition, and 3D games are particularly computation intensive. Multitasking-launching and simultaneously running multiple applications in a Microsoft® Windows® environment-places an added burden on the CPU and system memory.

For the most part, x86 processors have been able to keep pace with the performance needs of "power users" and PC enthusiasts through evolutionary enhancements of CPU architecture and continual increases in clock speed. Another way for the CPU to work more efficiently and deliver more performance at the same clock speed is to provide an intelligently engineered and flexible cache architecture. Cache is a form of high-speed memory that is located close to the CPU and thus can be accessed much more quickly than main memory. Traditionally, PC processors have relied on two levels of cache-Level 1 (L1) cache, which is usually located internally on the silicon die, and Level 2 (L2) cache, which can reside either externally on a motherboard or in a slot module, or internally in the form of an "on-chip" backside L2 cache.

In designing a cache subsystem, there is a general rule of thumb: the larger the cache and the faster the cache, the better the performance (the more quickly the CPU core can access instructions and data). Recognizing the benefits of a large and fast cache design in feeding power-hungry PC applications, AMD has developed an innovative cache architecture that enhances the performance of PCs based on the Super7™ platform. Debuting in the new AMD-K6®-III processor with 3DNow!™ technology, AMD's new TriLevel Cache design significantly improves overall PC performance by providing the desktop PC industry's largest total system cache-more than four times larger than systems based on the Pentium III processor.1

TriLevel Cache design architectural innovations:

An internal 256KB L2 write-back cache operating at the full speed of the AMD-K6-III processor and complementing the 64KB L1 cache, which is standard in all AMD-K6 family processors.
A multiport internal cache design, enabling simultaneous 64-bit reads and writes to both the L1 cache and the L2 cache.
A 4-way set associative L2 cache design enabling optimal data management and efficiency.
A 100-MHz frontside bus to an external cache [a scalable Level 3 (L3) cache on the motherboard], enabling a maximum combined system cache that is unsurpassed in the desktop PC industry.
By combining AMD's TriLevel Cache design with 3DNow! technology, the AMD-K6-III processor enables superior PC performance in the Windows 98 and Windows NT® operating systems.

Largest Combined System Cache for the Windows® Compatible Desktop PC Industry

When it comes to cache, bigger is better. In fact, the large full-speed cache provided by the TriLevel Cache design can boost overall system performance by up to 10 percent over designs that use an external L2 cache. The addition of an external L3 cache can further enhance delivered performance.

The frontside bus of the AMD-K6-III processor also enables access to an external cache on the Super7 motherboard that functions as an L3 cache, scaling from 512KB to 2,048KB and communicating with the processor core at 100 MHz. No other x86 processor design supports an external L3 cache on the motherboard. This powerful combination of both internal and external caches results in the largest total system cache available today.

Note: The AMD-K6-III processor's total internal cache size is 320KB (64KB internal L1 cache plus 256KB internal L2 cache), while Pentium III processor's total internal cache is only 32KB (the size of its L1 cache). The L2 cache of Pentium III is external, runs at half the processor speed, and is limited to 512KB. In addition, the AMD-K6-III supports an external L3 cache capability, while Pentium III does not.

Efficient Multiport Cache Design

The AMD-K6-III processor's multiport internal cache design enables both the 64KB L1 cache and the 256KB L2 cache to perform simultaneous 64-bit read and write operations in a clock cycle. This multiport capability allows data to be processed faster and more efficiently than non-ported designs.

In addition to this multiport cache design, the AMD-K6-III processor core can access both L1 and L2 caches simultaneously, which further enhances overall CPU throughput.

Full-Speed Cache for Faster Data Access

The TriLevel Cache design is not only the desktop PC industry's largest cache implementation, it is exceptionally fast, with the backside 256KB L2 cache operating at full processor speed. The internal L2 cache of an AMD-K6-III/450 processor, for example, operates at 450 MHz and provides nine times the peak bandwidth of an external L2 cache operating at 100 MHz. The maximum peak bandwidth of an external L2 cache operating at 100 MHz is 800 Mbytes/sec, while an internal L2 cache operating at 450 MHz delivers a maximum peak bandwidth of 3,600 Mbytes/sec per port. Because the internal L2 cache of the AMD-K6-III processor is dual-ported for simultaneous reads and writes, the total peak bandwidth is doubled to 7,200 Mbytes/sec, resulting in a maximum peak bandwidth nine times as large as a 100-MHz cache implementation.

About the AMD-K6-III Processor

1This estimate is based on a PC powered by the AMD-K6-III processor using a publicly available Super7™ motherboard with 2,048KB of external cache (total system cache of 2,368KB) and a Pentium® III based system configured with a maximum possible total system cache of 544KB (32KB internal, 512KB external).



[This message has been edited by jad1097 (edited 05-22-2000).]

[wing7788]

hey jad...

did you ask permission from amd to copy this document here? :-)

[Brydon]

Leave it alone http://sysopt.earthweb.com/forum/smile.gif.

[mnx]

konan--> what cpu / mobo were you running?? I run my celery 366 @ 616 with the l2 off... (just to get a few more kkeys/sec in rc5) and in UT the difference isn't noticable (only a few fps... with the l2 on it is just more consistant performance...) and windows boots up a little slower but not that much slower...

mnx

[JohnRobins]

Lets end this load of bollocks, my celeron 400 runs a third faster at 450Mhz than at 500Mhz with no Level 2 cache, If it weren't needed they wouldn't of put it there!

[Spiderdrake]

But offence to anyone but seriously, if the L2 is sooo unimportant, why does AMD pay to have a 512kb L2 Cache on the Athlon ? Why is it so crucial to performance that the Intel chips have a CPU speed L2 cache ?
Those L2 Caches on the Athlons are adding a lot of cost to their chips. If they are needed, they wouldn't be there ?
So Win 98, are you saying you're smarter than the AMD engineering team ? I've disabled my L2 Cache for overclocking before and although games run fine, Win98 is itself is much slower.

[oxOGradiusOxo]

Win_98, just change your name to Win_3.11, b/c I think you're just a little off base with this L2 cache thing.

From own personal experiences, I had the chance to benchmark two proessors before I put them in a friends computer. A Celeron 300@450 and a P2 450. Although the p2's cache runs at 1/2 bus speed, I still got better results in Q3, UT, 3Dmark2000, Sandra (same system same components just dif. cpu)

Try turning off your cache, and give us a post IF you ever get into Windows.

[jad1097]

Yes as a matter of fact! http://www.amd.com/legal/copyright.html

[Savant]

All of this explains why the Pentium Pro (L2 cache at core speed) outperformes its contemporarys by so much (their L2 is at bus speed) incedentally I think there is something else up with the celery, (other than cache size, and bus rate) as my Pentium Pro 200 can still outperform by a longshot my freinds celron 333.

[CMonster]

sometimes it sounds like people are smoking a little cachish around here..


sorry, couldn't resist http://sysopt.earthweb.com/forum/wink.gif

[This message has been edited by CMonster (edited 05-25-2000).]