I didn't say a word about desktop computers, or the 4x4 platform, why are you implying I did?

i just modified my previous posting. check it out.
we are all using desktop here, even my top 1 machine is a Desktop/workstation machine.

who?

Is Clovertown desktop, workstation, or server, or several of these?

I'm considering getting a 2 x 4-core box near the end of next year (can't afford it yet), and as far as I can see it the various choices are:

* 2 x Clovertown (or Peryn? Yorkfield?) CPUs on a dual socket motherboard

* 2 x Opteron 2xxx on a dual socket motherboard once they've got a 4-core offering (Barcelona)

* 2 x FX-76 on a 4x4+ motherboard (not 4x4)

---

Is Clovertown desktop, workstation, or server, or several of these?

I'm considering getting a 2 x 4-core box near the end of next year (can't afford it yet), and as far as I can see it the various choices are:

* 2 x Clovertown (or Peryn? Yorkfield?) CPUs on a dual socket motherboard

* 2 x Opteron 2xxx on a dual socket motherboard once they've got a 4-core offering (Barcelona)

* 2 x FX-76 on a 4x4+ motherboard (not 4x4)

I don't think K8L will be able to beat Core 2 based processors on Integer performance, and barely come to 10% on float. By the time it gets out, SSE4 will be available and vectorization in compilers will do a big jump.
The 128bits execution units will hurt big time the power of K8L, and at 2.5Ghz, it will be no match for my 2.6Ghz clovertown of today.

For processing, Quad core with 2 front side bus is shown to be a better solution yet. For workload like BOINC workload, the Cache is fairly important. I wonder why AMD choosed a so small L3 cache size.
Todays Quad core can provide you the speed of what AMD will do next year, With the perfecters of Core 2, most of the latencies are hidden, and the FSB on desktop does not saturate at all.

You can already buy some Dual clowertown at lower frequency, those are not cheap, but not that expensive. Vista support for NUMA is so "dissapointing", i don't think desktop will get any acceleration due to it.
The most amazing part of marketing it that NUMA was marketed as a "good thing", while, 99% of the threaded application like uniform memory allocation:
Think about it, most of your programs, internally use the same data to process:
In Povray or 3DSmax kind, the 3D world is allocated in a single buffer, and all threads go and pick up all the 3D information out of the 3D database of your scene. The thread only use NUMA only for their local variables (very likely to fit in the cache)
Even in 3D games, 3D collision detections and 3D rendering are sharing the same memory space.
In video encoding, the frame buffer is a uniform memory space too, so, NUMA in all of those actual processing power hungry apps is useless, so does Hyperslowtransport.
I don t even speak about the thread migration ... Vista support for numa is actually only a "stop" of the migration, plus some gadgets for Malloc/free

Bottom line, if you NUMA increase the distance to memory between one bank of memory and the cross CPU, it is bad ...

NUMA is not a good think, every programmer who had to program for it will tell you, it make your code less portable and less scalable==> You have to reset your affinities as soon as you add more cores.

who?

This is my own opinion.

CSI is a much better solution than Hyperslowtransport, the only common point it is that it is on chip. I am not going to comment, on desktop, hyperslowtransport showed how bad it is with GrandFather 4x4.

I am sick and tired of hearing people using Server workload to justify Hperslowtransport on desktop. 4x4 Fx74 and their motherboard is slower with 4 core than 2 core. what more do you need.


Again, with Hyperslowtransport enable, the GrandFather 4x4 is slower than its dual core version on games ... on Rosetta, it does not improve anything.

STOP USING SERVER WORKLOAD TO JUSTIFY HYPERSLOWTRANSPORT ON DESKTOP COMPUTER, IT IS STUPID!

oupsss, I forgot, if you look for carefully, the 1st x86 processor with integrated controler was a prototype called Timna, AMD did not invent anything: Timna picture

Stop giving credit to somebody who does not have it. Timna was what AMD is planning to do next year, a CPU + mem controler + GPU on a chip ... it turned out to be a not so good idea ... when you put you memory controler, your CPU and your GPU in the same die, you have to deal with the speed path of both.
Next year is going to be fun :):) are you ready for the very impressive 3800+ plus a X300 on the same chip ... it is my guess for the best they can do ...
for next year ... a 200$ PC: bye bye marging.

who?

I am fully responsable for this posting, my employer is not aware of it, and it does not ask me to post this. it is my own opinion.

Okay, you make a lot of assumptions about future releases of architectures AND you make assumptions about how CSI is faster then HT. But on the other hand: CSI is reporterd to support between 4.8 and 6.4 GigaTransfers/sec while HT makes for *SURE* 5.2 GigaTransfers/sec. So, how on earth can you tell that CSI is faster?

Further, your statement about the "very impressive 3800+ and a x300 on a chip" implies that *NOBODY* can *EVER* work on that piece of outdated hardware. Where do you live? Does everybody there needs a Cray for normal work? You, the poweruser, represent a *VERY* small part of the total market. Most of the market are office users, who don't need a fancy quadcore which is overclocked while using huge amounts of power to do their work.

Also, you seem to believe the intel PR talk about their "agresive prefetchers" who do "a darn good job". The actual truth turned out to be that if you turned the agresive prefetchers off, you would get a fair boost in the performance. So far the dumb prefetchers.

---

[

Also, you seem to believe the intel PR talk about their "agresive prefetchers" who do "a darn good job". The actual truth turned out to be that if you turned the agresive prefetchers off, you would get a fair boost in the performance. So far the dumb prefetchers.

and you forgot, the chickens have teeth ...

the prefecters are slowing down. That's a new one!!! (where did you read this? on AMD pdf?) I guess, all the part of the Core 2 are not working, it is why it is much slower than K8?
Get online, and read the benchmark scores! Hellooooooooooooooo???
Core 2 is way faster!!! where did you spend the last 8 months?
And please... don't try to teach me about Core 2 .. if you dig, you ll figure out how ridiculous your post was...

ask yourself! who is who? (sorry guys, i love this one)

who?

who is who? (sorry guys, i love this one)

I know, i got it! ... this dude thought I was Dr Who, and Dr Who in the last episode got stock in the VOID during one of his interdimentional travel ... he propably think i was stock there, with Rose, and I did not see the Core 2 performance test for the last 8 months. In fact, I think Core 2 is the most lucky processor on this planete, the Memory force it to get the right cache line before the loads units even need it. It is probably due to Dr Who sonic screw driver pushing it into the FSB.

Some time, those fan boys are really funny, but they can do nothing against the trans-dimensional Core 2 prefecter that does not work :)

too good ...

who?

G'day DHG,
welcome to Rosetta@home.

Who? is an Intel employee, and is unmercifully ribbing Mats Petersson (he has a AuthenticAMD AMD Engineering Sample CPU, and may work for AMD)

It's all in good fun (Intel has bragging rights at the moment), and the competition is very good for Rosetta@Home.

Let AMD and Intel fight, we win whatever the outcome :)

Happy crunching

- Hugo

Edit: to make a little more sense

G'day DHG,
welcome to Rosetta@home.

Who? is an Intel employee, and is unmercifully ribbing Mats Petersson (he has a AuthenticAMD AMD Engineering Sample CPU, and may work for AMD)

It's all in good fun (Intel has bragging rights at the moment), and the competition is very good for Rosetta@Home.

Let AMD and Intel fight, we win whatever the outcome :)

Happy crunching

- Hugo

EXACTLY!!!!
hummm hummmm, my RAC will pass 3400 over night on my baby ;)

who?
PS: Nothing personal... it is all fun ;-)
notice that it is only my opinion, my employer is still not responsable for my posting, i do it on my own, using the 1st amendment. Actually, they are some time red purple that i speak my mind here, but it is all fun :)

[

Also, you seem to believe the intel PR talk about their "agresive prefetchers" who do "a darn good job". The actual truth turned out to be that if you turned the agresive prefetchers off, you would get a fair boost in the performance. So far the dumb prefetchers.

and you forgot, the chickens have teeth ...

the prefecters are slowing down. That's a new one!!! (where did you read this? on AMD pdf?) I guess, all the part of the Core 2 are not working, it is why it is much slower than K8?
Get online, and read the benchmark scores! Hellooooooooooooooo???
Core 2 is way faster!!! where did you spend the last 8 months?
And please... don't try to teach me about Core 2 .. if you dig, you ll figure out how ridiculous your post was...

ask yourself! who is who? (sorry guys, i love this one)

who?

Yes, the "agresive" prefetchers are slowing the thing down. Speaking about benchmarks, why is it that a benchmark which runs with the "better" prefetchers enabled has a lower score then one where they are disabled?

And I don't know where you get the impression that I'm saying that a Core 2 is slower then a K8? I only did make a small statement about the new "agresive" prefetchers that you claimed to be the best of the best are actually slowing the thing down ... Of course, even with them enabled, the core 2 architecture is faster then the K8. Perhaps you should see some benchmarks, shouldn't you?

And what's ridiculous about my post? My statement that it is inmposible to determine which technology (that has to be released yet) is superiour/inferiour? Or perhaps it is the fact that everybody where you lives uses so badly writen software that they *INDEED* need to use a Cray to do their office work (the 3800+ & x300 thingy you disdained). No, it must be the prefetchers which where claimed to be soooooo good but actually turned to be a slowdown. A woodcrest without them enabled sees an increase of about 14% in a SPECjbb2005 benchmark.

@hugothehermit: Of course we must them let fight, but I only state that even the best of the moment has some flaws. I'll hope I can have here as much crunching fun as on the other projects where I used to crunch (they sadly are most of the time out of work :( )

---

Yes, the "agresive" prefetchers are slowing the thing down. Speaking about benchmarks, why is it that a benchmark which runs with the "better" prefetchers enabled has a lower score then one where they are disabled?

May be you need some numbers:
Prefetcher sucess rate:
3DsMAX: 98%
Povray: 99.9%
Halflife II:99.8%
FEAR: 99.1%
DivX: 99.7%
for most of the application, the success rate of the prefecther is over 99%

I guess you are using a synthetic unpredictable memory access and then you claim that the prefetcher are not working ... very classical for a bias mind who want to make a point.

Bottom line: when you make a claim, make sure you know what you are taking about, in all natural memory pattern generated by real applications, the prefetchers do wonders, use may have find a corner case with a synthetic test, you can't generalize it.

Yes, the "agresive" prefetchers are slowing the thing down. Speaking about benchmarks, why is it that a benchmark which runs with the "better" prefetchers enabled has a lower score then one where they are disabled?

May be you need some numbers:
Prefetcher sucess rate:
3DsMAX: 98%
Povray: 99.9%
Halflife II:99.8%
FEAR: 99.1%
DivX: 99.7%
for most of the application, the success rate of the prefecther is over 99%

I guess you are using a synthetic unpredictable memory access and then you claim that the prefetcher are not working ... very classical for a bias mind who want to make a point.

Bottom line: when you make a claim, make sure you know what you are taking about, in all natural memory pattern generated by real applications, the prefetchers do wonders, use may have find a corner case with a synthetic test, you can't generalize it.

The following is in dutch, but it has a little chart somewhere on the page. I guess that even you must be able to understand some charts. Prefetchers and disabled prefetchers, the bar with a little green 'check' before it indicates where the advanced prefetchers are enabled (a little hint: it's the bar which is the shortest of all woodcrest cpu's)

Now, if you again dare to say that they are good, it proves that it is you that is biased and not me. I NEVER said that the core 2 architecture was bad, but some of the things that where said to make it fast turned out to be things that could slow it down. Now, if it is possible to slow down the whole thing with something, then that something is not good, it might be good, but not 'IS' like you claimed to be. It can be something like hyperthreading, somethimes good, but not always (=> that means that it isn't a key component for the better perfomance, like you claimed).

Little hint: try to search for the SPEC (= Standard Performance Evaluation Corporation) benchmark. The aim of that benchmark is to test "real-life" situations. That doesn't sound very synthetic to me, does it?

---

PS for Matt: I am still running without Hyperslowtransport.

Okay, that's getting annoying now. Could you give it a rest please?

why should i? The speed of the K8 never cam from it, it was just hyped this way. this myth need to be broken, Hyperslowtransport did not buy anything on the long run.

who?

Yeah yeah, we know, we know.. You don't have to mention it in *every post*, whether it's on topic or not. You're starting to sound like a broken record.

It almost looks like you were traumatized by the few years when AMD kicked Intel's ass and now you're overcompensating...

---

PS for Matt: I am still running without Hyperslowtransport.

Okay, that's getting annoying now. Could you give it a rest please?

why should i? The speed of the K8 never cam from it, it was just hyped this way. this myth need to be broken, Hyperslowtransport did not buy anything on the long run.

who?

Yeah yeah, we know, we know.. You don't have to mention it in *every post*, whether it's on topic or not. You're starting to sound like a broken record.

It almost looks like you were traumatized by the few years when AMD kicked Intel's ass and now you're overcompensating...

AMD never had a lead of 30%, dude! when they got 8% on gaming, it was a big deal, isn't it? so, 30% is a super mega big deal :)

who?

Yes, the "agresive" prefetchers are slowing the thing down. Speaking about benchmarks, why is it that a benchmark which runs with the "better" prefetchers enabled has a lower score then one where they are disabled?

May be you need some numbers:
Prefetcher sucess rate:
3DsMAX: 98%
Povray: 99.9%
Halflife II:99.8%
FEAR: 99.1%
DivX: 99.7%
for most of the application, the success rate of the prefecther is over 99%

I guess you are using a synthetic unpredictable memory access and then you claim that the prefetcher are not working ... very classical for a bias mind who want to make a point.

Bottom line: when you make a claim, make sure you know what you are taking about, in all natural memory pattern generated by real applications, the prefetchers do wonders, use may have find a corner case with a synthetic test, you can't generalize it.

The following is in dutch, but it has a little chart somewhere on the page. I guess that even you must be able to understand some charts. Prefetchers and disabled prefetchers, the bar with a little green 'check' before it indicates where the advanced prefetchers are enabled (a little hint: it's the bar which is the shortest of all woodcrest cpu's)

Now, if you again dare to say that they are good, it proves that it is you that is biased and not me. I NEVER said that the core 2 architecture was bad, but some of the things that where said to make it fast turned out to be things that could slow it down. Now, if it is possible to slow down the whole thing with something, then that something is not good, it might be good, but not 'IS' like you claimed to be. It can be something like hyperthreading, somethimes good, but not always (=> that means that it isn't a key component for the better perfomance, like you claimed).

Little hint: try to search for the SPEC (= Standard Performance Evaluation Corporation) benchmark. The aim of that benchmark is to test "real-life" situations. That doesn't sound very synthetic to me, does it?

little hit for you ... you spoke about SPECjjb ... try SPEC2006 CPU ... and your argument collapse, the prefetchers are a great help on SPEC2006 int and SPEC2006 FP. Again, you present one case to demonstrate the entier story wrong.
In most of the case, the prefetchers increase performance, do you really think we will release a Tech that slow down our CPU? you really take us for more stupid than we are ... or may be you over-estimate your understanding of the situation.
If you look for hard on the net, you 'll find a ppt explaining what you tried to describe with a big green logo on the top right corner. hummm hummm it is called FUD.

Who?
PS: Spec CPU is the undisputed CPU test.

I forgot: My personal RAC went over 7000 units, using 4 computers. I am at the 24th place for the Top list, i hope to make it to the TOP 20 list with only 4 computers :) Let's crunch some more units and find some high energy match! cure some deases! Remember, we are all here to crunch and find cures. Don't hold back, give your cycles away to rosetta!

(*cut*)

little hit for you ... you spoke about SPECjjb ... try SPEC2006 CPU ... and your argument collapse, the prefetchers are a great help on SPEC2006 int and SPEC2006 FP. Again, you present one case to demonstrate the entier story wrong.
In most of the case, the prefetchers increase performance, do you really think we will release a Tech that slow down our CPU? you really take us for more stupid than we are ... or may be you over-estimate your understanding of the situation.
If you look for hard on the net, you 'll find a ppt explaining what you tried to describe with a big green logo on the top right corner. hummm hummm it is called FUD.

Who?
PS: Spec CPU is the undisputed CPU test.

I forgot: My personal RAC went over 7000 units, using 4 computers. I am at the 24th place for the Top list, i hope to make it to the TOP 20 list with only 4 computers :) Let's crunch some more units and find some high energy match! cure some deases! Remember, we are all here to crunch and find cures. Don't hold back, give your cycles away to rosetta!

Wait wait wait, I presented 1 case to prove your story wrong (the godly prefetchers) ... perhaps you should read my last comment a bit more (I agree that when writting previous posts, I perhaps was not very clear on the matter), that they are not always good (the hyperthreading like statement).

Besides, I don't have the time to search for every bloody test that's out there, just stating that I've seen benches that says otherwise. Papers with the logo of a manufacturer on it are not readed by me (quit saying that I must search for that AMD ppt), UNLESS they are datasheets of specific components (in which case I'm out of options for independent results). I DO know that their whitepapers and such are next to useless for comparisons.

What do you actually develop at intel? Software? Hardware?

---

AMD never had a lead of 30%, dude! when they got 8% on gaming, it was a big deal, isn't it? so, 30% is a super mega big deal :)

who?

You can prove anything with statistics. Fact is, for a long while AMD had a better arch (first with x86-64 too), cooler chips that had better performance and were competitively priced. Few people are arguing that Netburst is better than K8 (which is 2003 tech, btw - of course it's not competing with Conroe).

But anyway, I don't have a conflict of interest here, I'm just happy that there's competition that is pushing both companies to improve and keep prices low. I'm not an employee of Intel or AMD, and I'm not a fanboy of either, I like them both... But I think that without AMD, Intel wouldn't be anywhere near what it is today. It would be the Microsoft of chipmakers, making over-priced mediocre products and releasing them at longer intervals. Competition is good.

My whole point was, you don't have to mention how you don't like hypertransport in every post. We got it the first few times. Now it's just annoying and immature.

---

AMD never had a lead of 30%, dude! when they got 8% on gaming, it was a big deal, isn't it? so, 30% is a super mega big deal :)

who?

You can prove anything with statistics. Fact is, for a long while AMD had a better arch (first with x86-64 too), cooler chips that had better performance and were competitively priced. Few people are arguing that Netburst is better than K8 (which is 2003 tech, btw - of course it's not competing with Conroe).

But anyway, I don't have a conflict of interest here, I'm just happy that there's competition that is pushing both companies to improve and keep prices low. I'm not an employee of Intel or AMD, and I'm not a fanboy of either, I like them both... But I think that without AMD, Intel wouldn't be anywhere near what it is today. It would be the Microsoft of chipmakers, making over-priced mediocre products and releasing them at longer intervals. Competition is good.

My whole point was, you don't have to mention how you don't like hypertransport in every post. We got it the first few times. Now it's just annoying and immature.

I couldn't agree more with you. Competition is the backbone of all innovation and progression.

But perhaps we should let it rest now, before this converstation about little things of the core 2 architecture get out of control.

---

What do you actually develop at intel? Software? Hardware?

Who? is Francois Piednoel (previously ran here under that name, he found a lowest predictor back in September.

do a search on the net for that name and you'll know what he does...
ro just goto seti@home forum.

Interesting is if he is the same Francois Piednoel that worked on Max Payne then he may already know his nemesis Mats Peterson who he is have some tit for tat over since Mats also helped out on it.

Amazing what you stumble over on the internet.

---

Team mauisun.org

But perhaps we should let it rest now, before this converstation about little things of the core 2 architecture get out of control.

Absolutely, my goal was never to start a big debate.

I've just always disliked it when discussions about the technological marvels that modern CPUs are turn into low-brow sports cheering ("My team is better than yours, your guys suck!") and the accomplishments of others are denigrated (see Slashdot). I think with this subject matter, reason is more appropriate than emotion.

Fact is, the K8 was a great design for 2003 and it gave a good kick in the ass to the industry, and the Conroe is a great design for 2006 and it's doing the same thing now. We should all be happy about that.

The worse thing that could happen to all of us would be for Intel to crush AMD, so Intel fanboys should be careful what they wish for.

---

What do you actually develop at intel? Software? Hardware?

Who? is Francois Piednoel (previously ran here under that name, he found a lowest predictor back in September.

do a search on the net for that name and you'll know what he does...
ro just goto seti@home forum.

Interesting is if he is the same Francois Piednoel that worked on Max Payne then he may already know his nemesis Mats Peterson who he is have some tit for tat over since Mats also helped out on it.

Amazing what you stumble over on the internet.

I never hide it, I am senior performance analyst at Intel, I designed SSE2 instructions and optimize code. Yep, i worked on many games, many video codecs, and many scientify application. but keep in mind, Rosetta is my hobby.
I dedicated my life to performance studies, in computers or cars.
I am an expert in "performance".
From articulating an instruction need and getting it develloped to implementing it in software, that's what i do.
I was honnored to work with the Markus's team, on Max Payne, and I can't wait to see Alan Wake coming to life. Very exciting time for the game industry, finally using multi-core and getting closer to reality.

when I go and say that the FSB is not saturating, I spent 2 years of my life making sure it is not happening, you can trust me on this one, it does not happen! Same for the predicators, The sad part is to get those FUD ppt files saying the other way around in the case you put a niddle on Mars and divide the square root of the universe size by 1.23324534 and remove the cubic value of a atom of water ...

Of course i did meet Matt before (Trade shows), and Mike G too, In finland in the case of Mike, we had quit some beer together. Those are fine people.
What people need to remember is simple: people working for i or A all have famillies, we very often are part of the same neibourghood and community; and share common interest: computer.3 house away, I have a apple VP, and 1 block away, a VP of AMD. We shop in the same grosery store, and when we meet, we just avoid to speak about work, and it is fun.
well, with apple, the situation is kind of different now, I like it better :)
Competition is good if you keep the olympic spirit into it, I try my best to do so.

When i see FUD, it is different, I jump on it and mess it up, because FUD is misleading, and it is bad for the consumer.

and I am very proud of my findings in Rosetta predicator :), i wish to find more!

Merry XMAS to all!!!

who?

PERSONAL RAC = 7374, 23th place with 4 machines
Top 20 machine ranking:
TOP 1 Machine is RAC = 3464
TOP 2 Machine is RAC = 2572
1 PREDICTION