Quote:Original post by v0dKA
Are AMD's processors generally better than Intel's? I'm a bit lost on AMD's system of processor rating. Here's a (short) article I googled:

http://www.ocfaq.com/article.php/overclocking/AMD/36

So, as far as I understand, that means AMD's 3700+ performs at about ~3.7GHz, even though the actual GHz is lower.

First of all, before I make a decision based on a misjudgement, did I misunderstand anything? Second of all, would anyone actually recommend AMD over Intel?

Thanks for any help in advance.

Seems most of the replies have just been suggestions to go with AMD, not really explanations of the rating system. I agree, an AMD cpu is by far the best choice at the moment, but that wasn't really your original question.

There's a lot more to processor performance than pure GHz. Without getting too technical, it should be pretty obvious that there are two factors. How many cycles can you run per second (Clockspeed), and how many instructions can you run during each cycle. (IPC)

The latter can vary a lot, depending on countless factors. In pipelined cpu's (whic is everything since the 486), you will occasionally waste cycles because the result from the previous instruction has not been finished yet. The entire pipeline has to stall while waiting for one instruction to finish. There are various workarounds to minimize this, but you will always take a performance hit from this. On the other hand, pipelining allows far higher clock speeds.
Further, you waste a lot of time reading data from cache, ram or harddrive. A faster memory controller helps with the ram latency, and means less time is wasted there. Similarly, bigger cache also means fewer ram accesses, and again, less time wasted. But it can also means slower cache, and then time is wasted there instead. :P

And finally, most modern processors are superscalar, able to process 2 or 3 new instructions in parallel.

As you can probably see, there's a lot of factors influencing performance. If you have an IPC of 2, then a 1.5GHz machine performs just as well as a 3GHz with IPC = 1.

That's basically the difference between Pentium 4 and Athlon:
The P4 has a high clock speed, offering a (theoretically) huge top speed, assuming everything goes well, the code is in the cache already, there are few branches in the code, you have some fast ram and so on and so on.
Unfortunately, this *never* happens. There's always something to slow you down, and the P4 takes a much bigger performance hit when this happens, than an Athlon does, mainly due to its longer pipeline and smaller level 1 cache.

In practice, the Athlon's (and particularly Athlon 64's) perform as good or better than equivalent Pentium 4's. They don't have to access slow l2 cache or even slower ram as often, and it doesn't have to stall as long when it encounters a branch, because the pipeline is shorter.

However, keep in mind that the rating system is made by AMD, and so it's not guaranteed to give an accurate impression of performance. It is generally pretty accurate, but you have no guarantee that it'll continue to be so, or that it's accurate for every single one of their chips. It's a good rule of thumb, but no more than that. The 3700+ rating doesn't neccesarily mean that it performs equal to a 3.7GHz. It only means that AMD believes it does. Still, in truth it does perform really well, and the ratings usually match the performance pretty accurately.

And hplus0603: Thats not entirely true. All Athlon 64's use a Hypertransport link instead of the traditional FSB, yielding 800-1000 Mhz, equal or above what you get with the P4. And apart from this, the Athlon's just aren't as memory-starved as the P4. Even the singlechannel memory on Socket 754 chips is able to compete (and beat) P4's performance-wise. It's just a much more efficient architecture. It's no secret that the P4 was designed with clockspeed in mind, even at the expense of performance.

the dual channel 939 3000+ and 3200+ are extremely good buys for those who do workstation tasks ...

for those who do not do workstation tasks (such as us gameplayers), the single channel 754 2800+ can be had for $140 (and the 3000+ for $165).

all three current platforms (athlon xp, athlon 64, pentium 4) are really good, and should be more than adequate for nearly any task. The newest pentium 4 chipsets with PCIe are very nice - but beware of any motherboards using anything but DDR 400, because the DDR-II memory is still too expensive.

The althon xp line seems to be the best inexpensive line right now, with the 2800+ being worth well over its $85 price.

The athlon 64 2800+ - 3200+ give great top end performance for $140-$210

and The pentium 4 2800-3200 perform well for $180-$220.

I personally buy the AMD products because their XP line is much more affordable, and their 64 line is quite competitve performance wise, while giving the you the 64 bit future path pretty much for free. But I have friends and coworkers buying P4's and do not think they are making poor choices.

BTW, back to the original post, the AMD rating system had completely gone to hell in recent months ... the best idea - IGNORE the rating. Simply go look at the "sweet spot" priced CPUs, then go to somewhere like AnandTech and see how they perform on the types of things you do ... like if you game, pay attention to those benchmarks more, if you do 3D rendering, pay attention to those ... because the architechtures differ enough, even between the Athlon XP and Athlon 64, that each earns it's rating or better in some tests, while falling short in others ...

Not sure where Xai got the idea that the ratings has gone to hell.
The highest rated Athlon XP's were a bit exaggerated. The Athlon XP 3000+ barely beat a P4 2800mhz, for example. But with all Athlon 64's, the rating is, if anything, lower than what it should be. A Athlon 64 3000+ can pretty much compete with a 3.2GHz Pentium.

But of course, don't rely on the ratings. Use them as a guideline, and read up on some reviews and benchmarks to get an accurate picture.

Quote:Original post by BiGF00T
i've never experienced problems with my intel cpus... but i had some problems when i had an amd. this could have many reasons, not only the cpu and the 166mhz amd is long since forgotten so maybe they improved their cpus.

They did. Their early cpu's were pretty crappy. Athlon changed that though. And to be fair, Intel has made some crappy cpu's too. The P3 that couldn't divide comes to mind... :D

Quote:
i am an intel user since then and i never regretted it. my friends had heat problems with their amd cpu and a big heatsink, my intel p4 1.6 runs at 2.6 with the boxed hsf. i cant say anything about the latest generation cpus but i heard that intels cpus are now equally hot as amd cpus.

Not equally hot. Rather they're something like 50% hotter than AMD's.

Here's an interesting link:
http://www.silentpcreview.com/article169-page2.html
http://www.silentpcreview.com/article169-page3.html

I agree though, the only reliable way to choose is to read some benchmarks, plenty of benchmarks. Dont trust any one site, because a benchmark can be made to show a lot of conflicting results. Use one codec, and the P4 is unbeatable in media encoding. use another, and the Athlon 64 wins. Use one motherboard, and the P4 can (almost) compete with an Athlon 64. Use another, and it lags clearly behind. Read reviews and benchmarks on all the different sites you can find, and on average, you'll be able to get an accurate picture.

At the moment, I'd say AMD is by far the best choice, having the most features, the best performance, and the lowest heat dissipation, but in one year? Two years? Who knows?

AMD

The A64 and Axp are not as bandwidth hungry as the p4's. This is due to architechtural reasons. Since the p4 has a longer pipeline it will require it to have a much more larger cache subsystem and a faster memory interface cause the pipeline keeps stalling at times and it then has to do some memory lookup which slows it down.

The Hypertransport Bus is a Bi-directional serial Bus unlike traditional bus systems which the p4 is using now. Note however that this means that the AGP bus and all other buses will have to share the total system bandwith that is coming from the main memory in such Bus systems. There are two type of HTB. One is the one that connects to the system, the other is to other processor(cohorent links)...but that with come shortly.

Hypertransport however, is serial...it can support more than one read/write operation on the Bus meaning that these Buses don't have to wait/share the Bus, cause with traditional Buses you can only do one read/write operation on the Bus system which is a downside on current p4 systems :).

Because the A64 has a hypertransport Bus and a memory controller(memory Bus) on-die APART...the Bus systems do not interrupt each other and do not depend on each other to function. For example if you were to add PCI-Express on a p4 system with single channel memory like DDR-400..it will not be enough to feed the PCI-E Bus. PCI-E requires at least 4.0 Gb/sec of bandwith up/down to rip it's benefits(well enjoyable :)). Therefore the Bus will saturate :). But on an A64 system, i can even run it with DDR200 and still measure significant performance..why? Because the System bus(AGP,PCI...) is connected to the Hypertransport Bus not the memory Bus for they are APART unlike the p4 sys where everything runs from a northbridge(tranditional Bus). A HTB of 1000Mhz is enough to feed a PCI-E bus cause it can deliver up to 4.0 Gb/sec up/down(Bi-directinal). But think again...PCI-E is only one device on the Serial Bus...

Hehehe, Hypertransport bus can vary in speeds from 200Mhz up to 1,4Ghz with HTB 2.0 delivering over 22,4 Gb/sec to the system while still using your DDR200 memory :). And you know what's even more fun? Because the Memory Controller(Northbridge) is on-die it runs at CPU core speed :). But when it access memory it does it with a divider depending on the CPU clock speed to match the DRAM's speed. So if the CPU clock speed should increase on an A64 system, the memory latency(access time) in nanosecond(ns) goes down. At 2Ghz the memory latency is 45 ns. At 2,2 Ghz it's 40ns and so on. Unlike a p4 system that has to go over a northbridge to memory resulting up to at least 130ns of mem latency.

Also since the Hypertransport Bus and memory controller are on-die it can be easily cooled. So an A64 system will never lack bandwith unlike traditional Bus systems.


SCALABILITY OF AMD64 SYSTEMS

The AMD64 architechture is very efficient with SMP. An AMD64 processor can have more than one HTB link on die :). A desktop AMD64 has 1. A two way system has 2. One for the system the other to it's brother processor. A 4 -way system has 3. One to the system. Two to the other two brothers :).

Now, because each A64 has it's memory controller on-die...if each controller is capable of delivering up to 6,4 GB/sec then on a 2-CPU system you will see your system bandwidth climb up to 12,8 GB/sec. On a 4-way system it can climb up to over 20 Gb/sec. This is due to NUMA awareness. Each CPU is connected with a HTB cohorent link. if CPU0 needs data from CPU1's memory bank it simply 'streams' it from CPU1 through the HT link. This link currently @ 800Mhz and can transmit up to 6,4GB/sec through AMD64 CPU's. So it scale very well...unlike xeon's.

Let me explain. An Xeon system not only uses the older bus system bus it is shared. if the northbridge is only able to deliver 3,2 Gb/sec from memory then if i was to add another xeon CPU they will have to 'Share' this bandwith. With Four CPU each Xeon will have only 800Mb/sec left :(. That's why they always come with large caches...so that it doesn't have to fetch too much from the memory.... so this is the end of my long rant.....:=)

[Edited by - Ohmy on October 17, 2004 4:02:44 PM]

Rating or no rating system, AMD would kick intel's ass any day of the year :-p