question about AMD's different CPU's
i am kinda new too this forum and trying to follow, i hear alot of names thrown out like TBird which i think is the newest, and also i know K6 is older, but can someone tell me a little about the others inbetween like a brief bacground of the AMD processors out right now. thanks alot
I hope this is all right... someone correct me if i'm wrong
AMD cpu's, starting with the oldest...
AMD Athlon (1st true 7th generation x86 chip, L2 cache on motherboard, runs at 1/3, 1/2 or 2/5 chip speed)
AMD Thunderbird chip, slot A , on die L2 cache (L2 CACHE ON CPU CHIP SO IT RUNS at chip speed)
AMD Thunderbird and Duron, socket A, (on die L2 cache as well) -> most recent
hope this helps
grizely can u elaborate on that for me ... what do u mean by socket? and slot a . they are probably short for somethign cause u said thunderbird came out same time as socket. i know socket isn't a seperate kind of CPU like that athlon and tbird so i was wondering what those A things were and sockets thanks
AMD commercial CPUs started back in the 486 33 days. AMD started selling their own flavor of x86 CPUs cheaper than Intel's CPUs. Back then however, AMD was one step behind. When they came out with their 486 66MHz CPUs, Intel had the pentiums and so on. The K6 was AMD's first step into the pentium class market, but by the time they started selling in decent quantity, the pentium2 showed up. Basically they were gaining ground but still lagged behind (don't get me wrong, they were great CPUs for the money, but weren't top of the line). Once the Athlons came out though, they were the next step in consumer CPU evolution. The Athlon is based on the Dec Alpha EV6 bus which was new to the consumer market. Today a $400 AMD TBird 1.2gig still beats the $1000 P4 1.5 gig. K6-2+ and K6-3+ are revised older CPUs which are good for upgrading socket 7 boards.
"Water-Cooled CPU Runner"
The duron is a cheaper alternative to the thunderbird, like the celeron is to the PIII's
The duron only has 64 kilobytes of L2 cache whereas the thunderbird has 256 Kb's (more = better )
Cache, what is it? good question...
Again someone correct me
AMD and Intel have 2 caches, L1 and L2 (AMD having the bigger L2 cache, yay!)
Cache is where the instructions by the cpu are held until they are performed, and since the cache is on die or on chip, it runs at processor speed which is nice and fast.
man you are going to be absorbing info like a sponge. During the time I wrote my last post, 4 more replies showed up here. Hope all this stuff helps you out. Private messages might be better medium for this. I can try to explain it all, but I might end up writing a book.
"Water-Cooled CPU Runner"
<AMD and Intel have 2 caches, L1 and L2 (AMD having the bigger L2 cache, yay!)>
not quite. here are the cache sizes: (the most current processors)
celeron --> 32k L1 + 128k L2 (total 160k full speed)
duron --> 128k L1 + 64k L2 (total 192k full speed)
t-bird --> 128k L1 + 256k L2 (total 384k full speed)
P-3 --> 32k L1 + 256k L2 (total 288k full speed)
true true.... (athlons having most L2 cache, thanks for the correction ledzepp)
As to athlon t-bird question....
AMD makes the Athlon processor
The original athlon was replaced by the newer thunderbird processor.
Sometimes, the thunderbird is just refered to as the athlon processor for simplicity now as you can no longer buy any brand new "original athlon"
Speed Speed Speed!!!!
That is what this is all about.
Hard Drives are slow so->
Ram holds info for fast access of loaded or previosly loaded progs (when filled old progs get booted from ram)
Ram not the fastest though ->
The Cache - extremely fast memory (but expensive so not used for RAM)
Older CPUs (not too old, last generation) off die Cache (not on CPU / processor) - Athlon (Slot A) PIII (Slot 1) - aprox. 512 MB - 2/3, 1/2 & 1/3 speed of CPU
Newest CPUs on die cache so therefore able to acces faster for being on die and run faster at speed of CPU (socket CPUs)
Slot - big rectangular cpu package that stands upright and plugs in like a board (hell it is a board)
Socket - PGA pin grid array - small square cpu with lots of pins on bottom which slips on board flat
yeah yeah - PIV newest wiht littlest cache but enchanced even more (and runs faster right???)
so much to learn it never ends ... hop on board
<b> Fragg at will!!! </b>
Socket is what most motherboards used up until (k7)Athlon Slot A and the Pentium 2 (Slot 1) because there was a problem(mostly time) having level 2 cache on the actual silicon they put it on a board that the chip was attached to now to deliver this chip to us is how they came up with the slot instead of the socket since then they have resolved the issues we are now back at the socket name socket A for athlon and alot of names for intels chips but it was the socket 370 that celerons first went too
L1 vs L2 cache
L1 is nearer the processor core physically allowing for lower latency (less time to access cache memory).
L2 used to be separate from processor but has been integrated on the proc since P2 (I think), further out, higher latency.
Latency is bad, longer you wait for information from memory/cache, longer processor is doing nothing or acting off assumption about what memory will hold.
Going off personal memory (no idea where I got this it's that old), L1 latency is on order of a few clockcycles while L2 is up in tens of clock cycles.
"The answer is not in your hair."
"I'd rather jump in the lava than be fragged by you."
P4 has smaller cache yes but lower latency. It's a gamble of sorts really. If you have the info in the cache you can get it really quickly (2 clockcycles). But the likelihood of it being in the cache is lower (cache is smaller). In past, latency for cache wasn't as huge an issue as size. Xeon is prime example, tons of cache and can run better at lower clock than P3 on lots of apps.
Just like Tom said, the smaller yet faster cache could be either a blessing or a problem...
"The answer is not in your hair."
"I'd rather jump in the lava than be fragged by you."
I'm afraid AMD has been making functionally equivalent CPUs a lot longer than you think.
My first AMD processor (and my first overclocked one) was an AMD 80286-10 overclocked to 12 Mhz (without my knowledge as I believed I was buying a true 12 Mhz Zeos Computer in 1987).
I believe AMD was making Intel clone chips as early as 1977 before the existence of PCs. In those days the Intel processor was an 8088. This was followed by the 8086 & 8087 (a separate floating-point chip, the 80(1)88 (a faster 8088), the 80286, 80386, 80486 etc. AMD produced true clones for all of these. Somewhere along the way AMD started adding enhancements like the clock multiplier in the DX2, and DX4. Eventually AMD processors only maintained the same basic instruction set as the Intel equivalent. Internally the CPUs are no longer clones.
hi jojo., u got it wrong abt the duron cache. it has 64k of onchip L1 cache, and 128k of L2, total 192k onchip.
older athlons had 512k of cache but they werent onchip, they were outside of the chip on the processor card and ran at 1/2 or 1/3 of the processor speed. now amd has made athlon thunderbird (comes in both slot as well as socket flavours) which has 128k of L1 and 256k of L2 cache ruiing at full cpu speed.
CeleronA is the first L2 hard wired within the die.
PPro, P2 and Katmai P3 also come with "integrated" L2, but these L2 just being "inserted" within the same PCB of the proc.(you can see the L2 chips if you open up a P2 and Katmai P3)
you may consider these things as "a raiser card that contain L2 chips and the cpu core".
<font color=orange>What do you think? :wink: </font color=orange>
the ram thats in megabytes you use is the main memory and its too slow for the cpu. cache is a very small amount of ram, of a different type thats very fast, almost as fast as the cpu that sits between the cpu and main memory, the PC100/133 SDRAM what you call and keeps some bytes that the cpu just read from the main memery at hand, anticipating that the cpu will need them, and if it really does, it can deliver them to it instantly. the cpu will not have to go all the way to the slower main memory, so that speeds up the process. this is called a cache hit. amny a times it also happens that the byte cpu needs is not with the cache, so it has to go to the main memory and fetch it, while cache takes this chance and collects some more bytes from that area in the main memory for the cpu. this is a cache miss and the cpu then has to read the slow main memory. next time the cpu needs a byte from that area the cache has it now.
well, how many bytes does the cache read ahead for the cpu? the cache, depending on its organisation can read from 16 to 64 bytes in advance for the cpu. the cache is only about 64k (remember the 386/486 boards?) to 512k or even 2M on some p-IIIs. compared to the megabytes of main memory teh systems have this is really a tiny amount, so why bother about it?
cache memory is of static type, its a digital circuit and so its larger in size and a bit difficult to make, so its costly. whereas the main memory is of dynamic type, made of capacitors that holds the bits hence it is easier to fabricate as well as its very small in size. but it is much slower than the static ram and has to be refreshed periodically since the capacitor maybe discharged and it will forget about the bit it was storing.
you could well make your system with 256M of static main memory, but you could make its cost ten times!
well, L1 cache is the first level, first place the cpu looks for data, it always runs at full cpu speed and is only about 8k to 128k. if it misses then it could look for it in the main memory, but there is one more box of cache to look in, the second level or L2 cache and if this misses too, it will have to go to the main memory. some older boards (when the L2 started coming with the chip itself) had their own L2, so it had to be content being the L3 cache which actually degraded the performance since a lot of cycles were wasted to fill all the three caches! at best it had to be disabled. so too much cache does not help either. improvement in performance from 256k and 512k cache may not be the same as the amount of improvement from 64k to 512k. it of course depends on the application you are using.
hope you get it
ok i get all that, just one question more about the L2 and L1 cache. concerning being "cache on die" as apposed to no on the CPU, and all that about running at 1:2 or 1:3 ratio as appose to having cache run at full, where can i see this in my own computer? i think mine says something about having 256k of l2 cache? does that mean since i have 256 mb or ram i am running at full speed of processor or do i have this all wrong? i don't get that concept. thanks
um.... i checked out an article on tom's site and this is what i found...
"Duron is the close brother of 'Thunderbird', the new 'Socket-Athlon'. As a matter of fact the architectures of Thunderbird and Duron are pretty much identical. The only two differences are the size of the second level cache and the core voltage. Thunderbird comes with 256 kB on-die and full speed L2 cache, while Duron has only got 64 kB of that stuff."
I don't think it's a typo or anything, but i'm sure the duron has 64k of L2 cache
thanks anyway though
cache and main memory are different, main memory can exist without a cache, cache is processors' accessory which it needs to suffice its need for fast supply of bytes, just as you have your glass with you while all the wine is in the bottle. you fill it as and when you need it and you have it in your hands, no need to stretch your hands to the bottle! :-) is this analogy is ok....? :-)))
anyway, the 256k of cache is the ampount of super fast static RAM that is IN your CPU (is it a P-III Coppermine or a Athlon?) which is running with your CPU, and your RAM might be 128 MB or 256 MB, running at 100 or 133 MHz. btw pls tell me the specs of your system.
the distance between the cache and the processor is important, since the inductances and capacitances over the lines (on the PCB card or the tiny wires connecting to the other die in a hybrid chip)connecting them have their effect distorting the signals, so external caches are always run at a lower speed. on the die, the cache is fabricated on the same substrate as the cpu and the distances between them is very small, so it can be run at cpu speed without any problems.
this is right from amd's web site
"Q12: Does the AMD Duron processor feature on-chip cache memory? How large are the L1 and L2 caches of the AMD Duron processor?
A12: Yes. The AMD Duron processor features 128kB of on-chip L1 cache memory and 64kB of full-speed, on-chip L2 cache memory. "
check it out at http://www.amd.com/products/cpg/duron/faq.html
oh! well i just saw that page. and reread the manual (23865.pdf)! and downloaded its revision. and saw its appnote TN13.pdf that has a workaround for the Duron's misreporting of L2 cache size!
i had just started assumed that L1 has to be smaller than L2 (for no logical reasons!) and swapped the figures. it seems i have to read that appote TN13 4 times a day to clear the bug! :-) i misreported it too! what a goof-up!!
sorry abt that guys!