Bit-tech AMD six core review

[SEA]

Shintai
OK, how windows will dispatch threads across the sockets?
I just realized that Bit-tech used dual socket config.
And if windows will spread 4 threads across 2 cpus - every cpu will have ONLY 2 cores busy. Therefore it will run 3.33 GHz.

Is it correct?

And only if test has more than FOUR(?) threads it will degrade frequency to 3.2 GHz

[savantu]

guys , these boards don't have HT3 yet . Don't jump to conclusions.

Hypertransport has little to no relevance in 2 socket configs.

The bottleneck isn't the interconnect; but the cores.

[SEA]

Hypertransport has little to no relevance in 2 socket configs.

The bottleneck isn't the interconnect; but the cores.

Or frequency?

[Smartidiot89]

Fixed!

Don't worry, maybe things would get better with DDR3, and HT 3.0, and whatever else you guys are looking forward to. I like what Hans termed "(temporary) Server platform limitation;" to me it reads, a poorly thought-out strategy/design which does nothing but make AMD look bad when old gen intel is beating it in benches. In many ways, that is why some of you have no choice but always look to the future for hope and solace. Thing is though, if you want high-end AMD, you can only seriously consider Istanbul.

I really couldn't care less, I didn't buy AMD for the lower price if thats what your saying. All I did point out was that the test was horrible.

I wonder if any site can pull it off to build 2 dual-socket systems based on Nehalem and Istanbul at the same pricepoint and keep it unbiased

[MrMojoZ]

Fixed!

Don't worry, maybe things would get better with DDR3, and HT 3.0, and whatever else you guys are looking forward to. I like what Hans termed "(temporary) Server platform limitation;" to me it reads, a poorly thought-out strategy/design which does nothing but make AMD look bad when old gen intel is beating it in benches. In many ways, that is why some of you have no choice but always look to the future for hope and solace. Thing is though, if you want high-end AMD, you can only seriously consider Istanbul.

I'm amazed you read into a statement about an AMD product and came out with that assesment, you are usually very un-biased.

[Shintai]

Shintai
OK, how windows will dispatch threads across the sockets?
I just realized that Bit-tech used dual socket config.
And if windows will spread 4 threads across 2 cpus - every cpu will have ONLY 2 cores busy. Therefore it will run 3.33 GHz.

Is it correct?

And only if test has more than FOUR(?) threads it will degrade frequency to 3.2 GHz

If you run 4 threads or less out of the possible 16 yes in almost all cases. Unless coded with some exception.

[duploxxx]

Fixed!

Don't worry, maybe things would get better with DDR3, and HT 3.0, and whatever else you guys are looking forward to. I like what Hans termed "(temporary) Server platform limitation;" to me it reads, a poorly thought-out strategy/design which does nothing but make AMD look bad when old gen intel is beating it in benches. In many ways, that is why some of you have no choice but always look to the future for hope and solace. Thing is though, if you want high-end AMD, you can only seriously consider Istanbul.

i wonder if you were actually working with server hardware for the last 6 years orso, perhaps you would know who dominated the server market performance/price/power wise for a very very long time.... and even never lost that crown to some extend only lost parts of it due to design/process issue and is still capable to hold on in the volume market with latest tech of the competitor on a 6 year old design. It's only the brand name that kept them alive for all those years.... oh yes and those off course never happend sales tricks.

how does that sound from a strategy/design perspective. blue fanboys they keep amazing....

I'm amazed you read into a statement about an AMD product and came out with that assesment, you are usually very un-biased.

I think you are confusing him with someone else, has been like that from the moment he arrived on the news forum

If you run 4 threads or less out of the possible 16 yes in almost all cases. Unless coded with some exception.

from a technical point of view, sure and now reality ? no clue about final speed, seen these multipliers go back and forth so many times it actually gets you wacko.

[SEA]

If you run 4 threads or less out of the possible 16 yes in almost all cases. Unless coded with some exception.

That's exactly the problem in their review - most of tests are not using maximum threads, that is in gaming, and in encoding where performance of 8 and 12 cores are equal.

And only tests that use max threads show difference. BUT (!) there is another factor that would make difference - cpu interconnect speed. Depend on program nature it is can be more or less significant.
For example cinebench, where 12x Istanbul @2.6 DOES 25595, and W5580 @3.2 does 2898. Recalc performance of latter to 2.6 - it you will get IPC less than Istanbul.
Again, W5580 is NOT suppose to be @3.2 GHz in cinebench (all cores are loaded). If so - that is yet another flaw of that review.

In addition - in couple of month we will see the HT3 MBs and this will make a difference in true server tests

[villa1n]

Hardly belivable considering 8 instances of app runnig on 4 cores (+HT). I rather belive that most of the time it ran at 2.93 GHz with high CPU utilization.

Thats an opinion, based on your technical capabilities. These submissions, have a lot of wiggle room in regards to tweaking and cherry picking. Just because the average retail chip can't do that, doesn't mean a cherry picked one cant. One has to assume, that it is as close to the high frequencies as possible, trending away from the norm, in an effort to make it appear as the norm. Simple marketing and statistics.. and i don't see any evidence your posited theory is any more likely, if anything, its counter-intuitive, and less likely given the motivations behind the submissions.

It states that turbo mode was active. It doesn't mean that CPU was in turbo mode during the entire test.

But how long was it active? it could be anywhere from 99.99999999% to 0.000001% its not specified, one has to assume, as long as possible, again given the criteria and motivation.

Care you to explain how it is possible to ajust bios settings/kernel in such way, that additional 133-266MHz of frequency will be equivalent to one cpu core (since we already know that 3.33GHz isn't possible when app loads all 4 cores of x5570).

Again, where is the fact, it can't be loaded on all 4 under any circumstances.? Citation?

Your nit picking something that doesn't make much sense. Given the rules of the submissions, and the motivation to show the best possible scores...one has to assume tweaks versus, off the shelf performance..its pretty common sense?

[kl0012]

Thats an opinion, based on your technical capabilities. These submissions, have a lot of wiggle room in regards to tweaking and cherry picking. Just because the average retail chip can't do that, doesn't mean a cherry picked one cant. One has to assume, that it is as close to the high frequencies as possible, trending away from the norm, in an effort to make it appear as the norm. Simple marketing and statistics.. and i don't see any evidence your posited theory is any more likely, if anything, its counter-intuitive, and less likely given the motivations behind the submissions.

OK. Let see. Someone made a statement. You know, it is common to be asked for evidence when you made a statement. So I asked for an evidence supporting this statement. Now I get the answer: "It is you, who should provide an evidences". Isn't it funy? But I don't complain. My only statement (3.33GHz is not possible for x5570 when 4 cores loaded) was already supported by evidence. See link to anandtech above.

But how long was it active? it could be anywhere from 99.99999999% to 0.000001% its not specified, one has to assume, as long as possible, again given the criteria and motivation.

So do you agree with me that Hans statement is incorect since we can't tell for sure that CPU ran at 3.33GHZ 100% of time (or even 50% of time).

Again, where is the fact, it can't be loaded on all 4 under any circumstances.? Citation?

Your nit picking something that doesn't make much sense. Given the rules of the submissions, and the motivation to show the best possible scores...one has to assume tweaks versus, off the shelf performance..its pretty common sense?

Again, it is not my busines to provide evidence. I asked for simple explanation. Till now I didn't get any evidence (or explanations) except some "conspiracy theory" about cherry-piked cpus and mystical BIOS settings.

[YukonTrooper]

Hmmm... hopefully we'll get different results with other reviews, but results are results. Satisfactory, I suppose. Then again, I guess I didn't expect miracles.

[duploxxx]

In addition - in couple of month we will see the HT3 MBs and this will make a difference in true server tests

the ht3 mobo's with the new amd chipset sr5670-5690 should be released within a few weeks, the question will be if many server vendors will create new design in stead off keeping the long known chipsets for a few months knowing that in q12010 a whole new chipset and cpu line (c32-g34) will be available. I doubt we will be seeing many of the new 1207 chipset. On the other hand all major server vendors will start the new line as fast as possible which is based on the same chipset series as 56xx.

[informal]

Hmmm... hopefully we'll get different results with other reviews, but results are results. Satisfactory, I suppose. Then again, I guess I didn't expect miracles.

It's been already stated the "test" was a joke since it didn't test the CPUs in workloads that they were supposed to do,but in poor excuse for client level of apps(the ones we do on our desktops). Server workload does not equal client workload,so the test that was supposed to do the evaluation of these CPUs in former but provides a set of apps that belong to latter group is pointless. Not to forget to mention that even some of those client apps failed to scale well up to or past 8 threads...

[Shintai]

That's exactly the problem in their review - most of tests are not using maximum threads, that is in gaming, and in encoding where performance of 8 and 12 cores are equal.

And only tests that use max threads show difference. BUT (!) there is another factor that would make difference - cpu interconnect speed. Depend on program nature it is can be more or less significant.
For example cinebench, where 12x Istanbul @2.6 DOES 25595, and W5580 @3.2 does 2898. Recalc performance of latter to 2.6 - it you will get IPC less than Istanbul.
Again, W5580 is NOT suppose to be @3.2 GHz in cinebench (all cores are loaded). If so - that is yet another flaw of that review.

In addition - in couple of month we will see the HT3 MBs and this will make a difference in true server tests

You cant calculate performance backwards like that.

And if you look on memory. The nehalem system seems quite wierd. 16GB with trichannels aint easy to put it mildly. The opterons systems would need 4 dimms total. 2 for each CPU for max performance. The nehalem system 6 dimms, 3 for each for optimal performance.

Its simply an extremely bad review for several reasons beyond this.

[Hans de Vries]

OK. Let see. Someone made a statement. You know, it is common to be asked for evidence when you made a statement. So I asked for an evidence supporting this statement. Now I get the answer: "It is you, who should provide an evidences". Isn't it funy? But I don't complain. My only statement (3.33GHz is not possible for x5570 when 4 cores loaded) was already supported by evidence. See link to anandtech above.


So do you agree with me that Hans statement is incorect since we can't tell for sure that CPU ran at 3.33GHZ 100% of time (or even 50% of time).


Again, it is not my busines to provide evidence. I asked for simple explanation. Till now I didn't get any evidence (or explanations) except some "conspiracy theory" about cherry-piked cpus and mystical BIOS settings.

You do not read my posts, even though you react to them....

1) The SPEC submission says that Turbo Mode up to 3.33 GHz is used.

2) It can not be expected from the reader of the SPEC submission
that he knows (somehow magically) at what frequencies the test
were run. The "reader" that includes you and me.

The goal of a SPEC submission is to score as much points as possible
This will make a vendor's system look better compared to his competitors
which have mostly identical systems. The vendor has a number of possibilities
to increase the score which he can (and should) use. Running at the highest
possible frequencies is one of them. This has nothing to do with a conspiracy
theory.


Regards, Hans

[Chad Boga]

You do not read my posts, even though you react to them....

1) The SPEC submission says that Turbo Mode up to 3.33 GHz is used.

2) It can not be expected from the reader of the SPEC submission
that he knows (somehow magically) at what frequencies the test
were run. The "reader" that includes you and me.

The goal of a SPEC submission is to score as much points as possible
This will make a vendor's system look better compared to his competitors
which have mostly identical systems. The vendor has a number of possibilities
to increase the score which he can (and should) use. Running at the highest
possible frequencies is one of them. This has nothing to do with a conspiracy
theory.


Regards, Hans

Just because the incentive is there to run Turbo mode at it's highest level doesn't necessarily mean that the workload of the benchmark allowed that to happen though, that is the point you are obfuscating.

[SEA]

Oh, I just realized, that W5580 has base freq. 3.2 GHz, and in turbo - up to 3.6 GHz

[Shintai]

Oh, I just realized, that W5580 has base freq. 3.2 GHz, and in turbo - up to 3.6 GHz

Its certainly dont have a turbomode that high. Its 3.33ghz and 3.46Ghz for 3-4 and 1-2 cores.

[SEA]

OK. Nevertheless it still affects many their tests, particularly games

[Zucker2k]

I really couldn't care less, I didn't buy AMD for the lower price if thats what your saying. All I did point out was that the test was horrible.

I wonder if any site can pull it off to build 2 dual-socket systems based on Nehalem and Istanbul at the same pricepoint and keep it unbiased

The test was horrible because we actually saw how the hardware would react in a workstation environment?

I'm amazed you read into a statement about an AMD product and came out with that assesment, you are usually very un-biased.

How can I be biased when I've been proved right with my prediction of every AMD cpu in the last two years or so? Listen, with the current, and rather unrealistic emotional attitude that some of you bring to these discussions, I should be on a fine streak for a while more to come.

It's been already stated the "test" was a joke since it didn't test the CPUs in workloads that they were supposed to do,but in poor excuse for client level of apps(the ones we do on our desktops). Server workload does not equal client workload,so the test that was supposed to do the evaluation of these CPUs in former but provides a set of apps that belong to latter group is pointless. Not to forget to mention that even some of those client apps failed to scale well up to or past 8 threads...

Oops, there goes my idea for an Istanbul based workstation.

OK. Nevertheless it still affects many their tests, particularly games

Yay, Turbo works like Intel intended it to work - on servers and workstations - in games.

[Smartidiot89]

The test was horrible because we actually saw how the hardware would react in a workstation environment?

Thats regular Desktop environment for gamers, not a workstation that can utilize all the potential either the Istanbul or Nehalem has to give.

[Zucker2k]

I really couldn't care less, I didn't buy AMD for the lower price if thats what your saying. All I did point out was that the test was horrible.

I wonder if any site can pull it off to build 2 dual-socket systems based on Nehalem and Istanbul at the same pricepoint and keep it unbiased

Thats regular Desktop environment for gamers, not a workstation that can utilize all the potential either the Istanbul or Nehalem has to give.

Utilize all the potential? And that comes automatically in a server environment? My friend, stop the excuses, Istanbul was not the only cpu tested in this environment, you know. And even in pure server environment, how many benches do you think it wins?

[Smartidiot89]

Utilize all the potential? And that comes automatically in a server environment? My friend, stop the excuses, Istanbul was not the only cpu tested in this environment, you know. And even in pure server environment, how many benches do you think it wins?

What I think isn't interesting, give me an unbiased review of the 2 instead of trolling then you'll shut me up

[informal]

Oops, there goes my idea for an Istanbul based workstation.

Yeah,how strange (not) that AMD actually doesn't tout Istanbul as workstation CPU... The high clocks(turbo/no turbo- it doesn't matter) of Nehalem makes it a better fit for workstation environment than Istanbul at present. After Istanbul gets to ~3Ghz in SE variant,AMD will probably launch 1P boards for Suzuka equivalent of Istanbul MPU.Add in the HT3/DDR3 in H2 09 and you will have a very very capable workstation platform based on RevD.

[Zucker2k]

Yeah,how strange (not) that AMD actually doesn't tout Istanbul as workstation CPU... The high clocks(turbo/no turbo- it doesn't matter) of Nehalem makes it a better fit for workstation environment than Istanbul at present. After Istanbul gets to ~3Ghz in SE variant,AMD will probably launch 1P boards for Suzuka equivalent of Istanbul MPU.Add in the HT3/DDR3 in H2 09 and you will have a very very capable workstation platform based on RevD.

"Very very capable;" you should be a politician. Unfortunately, AMD is trying to keep up with a moving target, and there I say, going about it all wrongly - at least for now.



One thing I find very very interesting is the disparity between max clocks on AMD server cpus and desktop counterparts. They need the clocks on the server front, but you don't see it, it's rather on the desktop.

To me this is indicative of the strength of the architecture; the very stringent and demanding desktop environment exposes what AMD has been getting away with on the desktop platform. And before anybody starts a "power consumption" debate, here's a question: What does AMD need more?

1. A 3.4Ghz server part with a 125w (genuine tdp)

or

2. A 2.8Ghz server part with a 95w iirc (suspect tdp or acp)

Keep in mind that, it still gets beat in performance/watt, and in performance by the competition. So how effective is this power-consumption strategy I've heard being made?

Of course since I've always been very suspicious of AMD's tdp/oc headroom numbers, I'm easily led to conclude that latent architectural weaknesses are the reasons behind the disparity in clocks on server vs desktop AMD. Simply put, AMD is not playing any strategic game, but is being forced to work with what it has. It is the only sound explanation because with its current IPC, IF AMD could release a 3.4Ghz chip on the server front and get away with it, they would.