Replies to post #4789 on AMD strictly moderated (AMD)

[mas]

Yeah and they all add up to 5% tops. IMC adds the other 30%. How about finding a real link rather than quoting an ingredient list !

[mas]

Here's an AMD link for you, out of many out there I hasten to add, many of which I have read. Now I have told you before and I am telling you again, stop lying and stop making stuff up like fictitious discussions and conclusions to further your agenda !!!

http://www.people.virginia.edu/~avw6s/images/opteron/CPUarticles/Computer_Power_User_Article-FredWeb...

Fred Weber: Absolutely. There are quite a number of things that are important for 32-bit performance. It's absolutely crucial for everyone to remember and realize that this is our flagship
32-bit processor. The 64-bit capability is for some people immediately valuable and for others is valuable later, but first and foremost this is going to be our flagship 32-bit new processor. We've brought a lot of new technology to the machine as part of that.

On-Chip Memory Controller

The most important is our on-chip memory controller and HyperTransport bus. By bringing [the]
memory controller onto the processor, we've greatly reduced the latency to main memory. In reality these very fast processors spend a lot of time sitting at red lights, burning gas and not really accomplishing anything, because they're waiting for data to come in from the DRAM, from memory. In a typical, traditional PC, whether an Athlon or an Intel Pentium 4 system, DRAM is on the order of 100 nanoseconds away from the processor. When you're running at multiple GHz,that means that every time you have to wait to get something out of memory, you're waiting for 200 to 300 cycles for that memory to come in and accomplishing nothing in that time period while you're waiting for data to come back from memory. With [AMD] Athlon 64 and [AMD] Opteron designs, we focused on reducing that incredibly important time period. By putting [the] memory controller directly on the processor, we've reduced that number by on the order of 30% so a lot less time is spent with a very fast processor not having something to do while it waits for memory. There are a lot of other "tricks" that are used in
processors to avoid this problem—out-of-order execution being the most obvious (which is used to expose memory requests much earlier so that you can actually do work while the memory request is outstanding), also prefetching and things like that. But even [with] all of those tricks, in the end, sometimes you just have to wait for memory, and it's well known that the latency to first memory access is a first-order effect on the performance of processors. So we've attacked that problem
very directly by putting the memory controller directly on the processor.

HyperTransport Bus

Another major innovation is the HyperTransport bus, which is our new very high performance interconnect for connecting I/O to the processor. What the HyperTransport bus does by using a
point-to-point serial interconnect rather than a bus structure [is] it allows us to run much higher performance I/O so that we achieve on our main interconnect out of the processor 6.4GBps of bandwidth. Since the memory controller is not separate from the CPU, we don't have to drag memory data back and forth across this bus, but instead all of the bandwidth of that interconnect can be used for I/O operations, getting your graphics data out to your graphics card and moving your data back and forth to your disk and your network. So the HyperTransport bus increases the total amount of bandwidth available to your I/O subsystem, and the memory controller on-board reduces the load on that I/O subsystem, which again gives much higher performance in I/O intensive environments.
So with both of these things and with a never-ending increase in clock frequency and in architectural innovations such as branch prediction and things like that, the [AMD] Athlon 64 and
[AMD] Opteron represent our eighth-generation processors. Just as the [AMD] Athlon, our seventh-generation processor, was a major step forward over our K6, the Athlon 64/Opteron is a
major step forward and will be our flagship product and a tremendously fast 32-bit processor.

CPU: Does moving the memory controller onto the chip have any negative effect on existing apps, which were designed to work with existing system latencies?

Fred Weber: No, it really doesn't. As far as correctness of execution, applications cannot be designed to depend in any way for their correctness on the latency to memory because things like caches mean that the latency to memory in any given situation can vary greatly already. So there's absolutely no issue of compatibility or correct operation. AMD has a long history of our 32-bit processors being absolutely compatible. In fact, we are very careful with all of our testing and
internally refer to our processors as the most compatible processors because we implement essentially all instruction-set extensions, whether they be ones that we've come up with or the
Intel ones, so that we can execute all code compatibly.
As far as performance tuning goes, code that is specifically tuned to an [AMD] Athlon 64 or an [AMD] Opteron will always run faster on those systems than code that is tuned to some other
processor and then run on the [AMD] Athlon 64 or an [AMD] Opteron. However, AMD has had a long history of knowing that most code is tuned for other processors, and so we are very
determined to make sure that our processors are not what I call finicky, that they are very good at executing code no matter what it looks like. So we find that we get better performance on code no matter where it's optimized. That said, that last few percent can always be better if you specifically
optimize, but in the end, all programs are optimized around doing the best to get rid of memory latency dependency wherever they can and then they just have to live with it. So when we come in and reduce memory latency, it always does give a boost—it's just a question of how much that boost is. It's very significant in a lot of cases, though; you can see a 20% performance boost for having the memory controller on-chip on very many crucial applications.

[mmoy]

If you look at that site that was posted last night that allowed
you to compare performance between a wide number of Intel and
AMD processors, you'd see that the 4400 beats the 4600 about
as many times as the 4600 beats the 4400. And on a lot of
benchmarks, the results are very, very close.

This would indicate that the extra cache outweighs the advantage
of 200 Mhz in quite a few benchmarks and shows that access time
to memory is a very important factor. Which is why I feel very
comfortable with my purchase of the X2 4400 as the best chip in
the lineup for the money with dual core. And which is probably
why there is no 4000 chip as I suspect that benchmarks would
show that it is overall comparable in performance to the 4200.

Of course you can get an Opteron at the X2 4000 level but
finding one at retail is near impossible.