what will amd do if hyperthreading catches on?

[nortexoid]

i'm still freaking waiting for AMD to implement SSE2 instructions into their CPUs....that should help

is barton expected to have it?

 

[sharkeeper]

AMD will leapfrog intel with four way hyper threading in the hammer. 🙂 Imagine a dual hammer showing 8 "cpu's" in the task manager! :Q

Cheers!

 

[Duvie]

I don't think AMD has much to worry about at current time...Like SSE2 that has been slow to catch on the HT will also be and some of the things it can help in now wont help the average user see much in performance lift. AMD will do fine except for this being a great marketing gimmick for a company that already has the advantage in marketing with higher overall mhz and brand name recognition.

Were it will mean something is about a year or so from now when HT matures and even more things are SSE2 and the gap will grow between older p4 chips and the xp and tbird counterparts. This will be seen in not HT chips that just have the advantage of SSE2...For many of us here that is likely not a factor, and we wont even think about it...We are about now and not what is promised in gimmicks that are hedged on a lot of other ppl to get it to work properly.

 

[Cerb]

Currently the XP 2800+ and P4 3.06GHz can be bought easily.
If my understanding is correct, Intel will use hyperthreading as a marketing tool, which will aid them, and it will keep them fron needing an extra 200MHz. By marketing tool I mean like the P4 1.6GHz with a "400MHz" FSB using PC133 for memory (c'mon, check your sunday paper every now and then). Also, AMD will be using hypertransport in the Hammer boards...hopefully the efficiency and bandwidth offered will allow them to compete with the higher-speed P4s. For one thing, I like how both Intel and AMD's prices are nice and low.

AMD's XP 2800+ coupled with the nForce2 (there are a few places you can get the Asus nf2 board, so likely we will be flooded with choices before christmas) manages to compete with the P4 3.06, even with HT enabled, on several benchmarks. While HT is a great thing and does help out, the nForce2's nice drivers (That's one thing nVidia manages most of the time quite well) help give it better loading times...I'll be checking it out myself as soon as I get to make someone a new machine again 🙂.

Overall, if AMD can pump out the Barton and Hammer on their current schedule, all should be well. While Intel has the R&D and fab power, AMD manages to one-up or at least meet them each time since the first slot Athlons. Let's hope it stays that way, for the sake of all our wallets--the Intel fanboys, AMD fanboys, and all those who could care less and just go buy a Dell.

 

[KF]

>Actually everyone uses stuff that HT will be helpful, like simultaneously burning
> a CDR on an IDE-based CDRW drive and playing a Dvix at the same time.

>Its at sub-maximum performance where this will matter. You don't have to
>run 100% to fubar the stream and create a coaster.

Is there some kind of test that shows HT will reduce the likelyhood of a coaster? I can't see how it would. The CPU is loafing while you burn CDs. It has vitually nothing to do but wait. If another task contends for memory or the system bus, like another thread would, it is more likely these resources will not be available at the time the burner program needs them, increasing the possibility of a coaster. There is nothing hyperthreading can do to help that. The only thing provided by HT is possible use of unused CPU resources, which is a different thing.

As for burning CDs while you view DiVX, do people actual do this? It only takes 3 minutes to burn a CD. You'd barely get focused on the video before you'd be interrupted in order take the written CD out. If you are going to burn several CDs, it would be even worse. If these DivXs are porn, you would get frustrated very fast.

When you run two programs simultaneously, at least one should not require your attention, I would think. If you run two heavy duty programs simultaneous, you basicly accept that it doesn't matter how long they take.

I find it difficult to believe that anyone genuinely concerned with how long it takes to encode to DiVX would want to run some other program simultaneously which would make the encoding take longer.

Interestingly, DivX encoding is one thing that benefits quite a bit from HT according to Anandtech. (Single program with multi-thread.) One wonders why. Since the resources are totally controlled by the programmers, they could write their program to get results at least as good. All hyperthreading is doing is running instructions through the CPU, and the programmers could certainly do that and do it better. Programmers should study the HT results and figure out where they flubbed.

Games do not benefit from HT. Is that a surprise? No. Games are definately optimized to use CPU resources as effectively as possible. If you find a game that benefits from HT, you know the programmers have let you down.

If a programmer cannot beat the results of HT technology, he is not worth his salary. What HT can do that a programmer can't is use resources he has no control over, such as running two programs simultaneously.

Before people get too worked up about how AMD is going to respond, realize that in a sense AMD has already responded. The unused resources that HT is making use of are the exact same under-utilized resources that make the P4s average instructions per cycle so low compared to the Athlon. Athlons make use of these better than P4s without even using HT. HT just makes P4s not quite as bad. By the same token, if AMD were to add HT, there would be less opportunity to make use of unused resources.

Anandtech says that resource utilization is low in the average application because of "the nature of most applications." What is that nature? Most parts of an application are not optimized because they are judged "fast enough" by the manufacturer. For the average application that is ALL the parts, simply because processors are so fast. In some classes of applications there are a few slow parts. Programmers concentrate their efforts on these few parts. Then there are programs, such as 3D games, that spend nearly all their time doing things that must be as fast as possible, and programmers sweat blood optimizing them. It is exactly those applications that don't need to run any faster -in the judgement of manufacturers- that are the ones that benefit from HT.

Then we have some applications that perhaps do not conform to the theory. For media creation, either the manufacturers don't see speed as significant as some users seem to, or the programmers have areas in which they need to correct their oversights.

According to Anandtech, HT hardware is present and identical in slower chips than 3GHz that have been available for quite a while. Some say that HT hardware has been there for a couple of years. Yet Intel did not enable HT because it was not ready until now. What is it that wasn't ready? The hardware was ready. It must be the software. In other words, HT is not exactly a hardware solution, as it has been portrayed, although it does require hardware. There must be a piece of software that is doing a very difficult part of making HT work acceptably, a program which has taken years to get right. You can see that this is a difficult problem because this program is running simultaneously with the programs that are hyper-threading, and therefore taking up resources for itself.


In case people skipped over that part of the Anandtech article and didn't notice it, Windows since 95 has been multi-threaded. When you write a W95 style program you can run multiple threads. Thus you can do disk access interleaved with spell-checking, for instance, if your program will benefit from that. What Intel's HT adds is hardware-assisted threading, which should be more efficient, only it is not totally hardware. Windows XP evidently can select HT by using its built-in multi-processor support. It is interesting that XP Home DOES have multi-processor support; it just won't use it unless it detects Intels HT. Mighty neighborly of MS to allow this exception.

People are thinking AMDs 64 bit capability will compensate for HT, in the sense that programs could run faster. As one excellent FAQ points out, the advantages of 64 bits are very limited. This is simply because programs so seldom do operations on 64 bit data. Operations on 8 bits or fewer suffice for 95% of program code, I'd guess.

An important aspect of x86 instruction op-codes is that they can compact the data and address parts down to 16 or 8 bits when that is sufficent. I imagine AMDs extensions will do likewise for 64 bits, so the FAQ is wrong that accessing 64 bit instructions will slow things down by requiring more memory accesses for the same operations. Likewise there is no need to store 64 bits of data when you only need 8 bits just because the processor is a 64 bit processor. 64 bits will offer direct addressing, not segmented, beyond the 4G which 32 bits provides. The need for that is not far off. (I can remember when we wondered what we would do with 64K of memory and that amount of memory cost $500. It doesn't cost a lot more for 2G now.) And yes, over 4G will make some things run faster. Imagine a whole DVD in memory. and no disk access to slow down encoding. That's in a couple of years.

AMDs extensions provide more CPU registers. This will make a more convenient programming model. Currently the number of registers is so few that is futile to attempt to keep things around in registers. It is not really true that register operations are faster than memory operations; if the data is in the cache, it is just as fast. Of course with registers you do not have to worry if it is still in the cache or if it has been bumped out. But the x86 instruction set has lots of ops that do not treat memory the same as registers, and AMD's set is an extension. Beyond that, it seems to me, the CPU can do optimizations with registers that cannot be done with memory, if only because other things (video card, HD controller) can access memory beside the CPU. Current CPUs actully do have a lot of registers (for optimization use), while only a few are visible to the programmer. The new way of using registers should make things faster.

So what is the use of operations on 64 bits? Most of the operations will be done on addressing. Just as now, CPUs generally do more operations on addressing than they do operations that the instruction designates, and these operations will be 64 bits. In some ways, you get the 64 bit ALU operations as a tacked-on bonus just because you need to do 64 bit addressing. But SIMD and SSE2 provide almost all of the advantage of the P4 over the Athlon, when there is an advantage. What these instructions do is operate on multiple sets of data simultaneously. Instead of doing operations on two sets of 16 bit data one after another, it might put the sets side by side in a register to make 32 bit wide data sets, and do both operations simultaneously. Being able to do this with 64 bit registers instead of 32 should bring an advantage to Hammer, if in fact they do this. Being able to load 64 bits with one op instead of two is an advantage.

What being close to Intel's performance brings AMD is prestige. That prestige translates into a higher price for AMD's CPUs. In terms of sales, I guess we all know that a tiny part of the buying public buys top-of-the-line CPUs. Even the public knows that dropping down a few notches drops the price stupendously without dropping performance much. But without that prestige model, AMD looks second class, and they can't get the price they need to sustain a business. If AMD "loses contact" with Intel, like a runner in a race, they are going to have to start over, or switch to a different business stategy. You can't fake prestige -not for long. Although plenty of companies have made large, sustained profits by being a classy second, I don't see AMD changing strategy. AMD will get Hammer out pretty soon. It should be competitive with P4s. It should "maintain contact". They are going to call Clawhammers Athlons, I believe. For quite a while after that, most of the sales are going to be what we now think of as Athlons. Does Intel have something ready to go that will "pull away" from Hammer ? I wonder. They've done it before. Still, as Intel says, they have optimized x86 CPUs about as far as they can go.

 

[CraigRT]

what will amd do if hyperthreading catches on?

A. It won't catch on for the price it's going to be:

or

B. The same thing as Intel! 😀

 

[MadRat]

KF-

If HT allows for more sloppy programming with less loss of performance then how is that a bad trait?

 

[CrazySaint]

Originally posted by: Yield

what will amd do if hyperthreading catches on?

A. It won't catch on for the price it's going to be

The high price is temporary. In a few months, the 3.06GHz P4 will be a mid-range CPU and will obviously cost much less.

 

[KF]

Originally posted by: MadRat
KF-

If HT allows for more sloppy programming with less loss of performance then how is that a bad trait?

It's not. HT is good idea. Its performance can be duplicated, and exceded, with careful programming.

Ideally programs would not be sloppy and bloated, but they are easier to do and cost less in programmers' time. It is irritating to see.

I remember the original Zip program (pkzip?) That's when 16MHz was fast, so speed really did matter. Some amateur programmer ripped it off and made a program that ran in a fraction of the time. They prosecuted the guy (he violated their copyright), but ended up using his program code. The company should have apologized for their lousey code.

 

[slackware1995]

Originally posted by: draggoon01
with reviews due in days, and at least 1 review up early, it seems hyperthreading will give a nice small boost. would have been great move on intel if they had included it or allowed it to be turned on in some of the more recent processors ~2.4ghz and up, to help prevalence of ht, and thus speed up development of applications with ht optimizations.

regardless, ht seems good enough to be desirable and sought after, especially when it comes with cpu's in mid to low price range in the future. and already talk of ht2. so i got to wondering, what will amd do in the face of this? will they try to develop future cpu's with ht? do they already have cpu's in the pipeline that will support ht?

or does this just mean they adjust their performance rating again; with ht being no different than a bump in cache or ghz.

HT is one thing. A marketing point Something Intel can say that they have that AMD doesnt have. Everyone agrees that HT does help some things to an extent. Most agree that 99% of users wont ever notice it (Real World users, not power users).

So, it boils down to buy Intel we have HT and it makes your computer faster! Same as they did with MMX and beyond.