Sign in with
Sign up | Sign in
Your question
Closed

Why the xbitlabs Thuban review is junk - author exposed.

Last response: in CPUs
Share
a b à CPUs
April 30, 2010 11:58:48 AM

The pre-review "Thuban is instanbul for desktop"

Remember that from 6 months back? How the author oh so confidently told us

Quote:
we expect AMD Thuban processors to work at about 3.0 GHz. And as we have seen during our today’s test session, it won’t be enough for them to outperform neither competitor’s six-core solutions, nor top quad-core Nehalem based CPUs, which cope perfectly well with heavy multi-threaded load due to Hyper-Threading technology support.


I guess if you already have stuck your neck on the line 6 months before the chip is even available...

The real review of Thuban


Same website, same author. I don't know what is up at xbit but at the very least a different reviewer should have reviewed the actual chip after the joke that was the "istanbul for desktop" article. That very article was the basis for people like Fazers, jimmysmitty and Chad to declare Thuban would be no better than the i5 750. :lol: 

As usual, you need to be very careful about who you believe. Some websites cant be trusted - xbit has proven itself untrustworthy over the recent months.
a b à CPUs
April 30, 2010 4:01:59 PM

^ Actually it was the much more recent Taiwanese review that sorta confirmed for me that Thuban, which is basically a stopgap CPU until Bulldozer comes out, would not be the "world's fastest desktop CPU". Which incidentally falls under your "need to be very careful about who(m) you believe" cautionary statement :D .

AMD took some economies with Thuban - same size L3 cache despite 50% extra cores, rather simplistic turbo solution, no improvement in memory bandwidth, no IPC improvements, etc. To me, that indicates AMD did a quick, minimal job with it in order to keep the AMD loyal in their camp until BD arrives. I would imagine BD is getting all the R&D funds & effort for some time now.
a c 131 à CPUs
April 30, 2010 4:20:17 PM

Haha. I read that article a long time ago. The performance numbers make sense but how would it be a preview of future possible thuban if he didn't overclock the opteron to at least 3GHz? His conclusion makes it sound like he did overclock to 3GHz when he didn't.
a b à CPUs
April 30, 2010 4:20:19 PM

Wow you totally avoided the point altogether there didn't you fazers?

The point is, the xbit review is a sham review because the author had already made up their mind long before the actual chip was released.

So that's one less anti-thuban to subtract from the legions of pro-thuban reviews out there. :) 
a b à CPUs
April 30, 2010 4:23:13 PM

but hell, you can't argue that it is not better than the 750, you can argue that the 930 is the king of gaming and get away with it, but not when the 1055 beats the 750 on price of platform and can be OCed regardless of what others are saying about it.

from my point of view, either clock for clock comparision, price per price comparision (then it'd be 965 or 955), or OC comparision

and in all instances, amd wins, either with a 955/965 or a 1055T


with the 1090T there is less of a win against the i7-920/930, and can be said to be on more equal footing.

but considering that the 1055 don't cut much from the 1090 (same platform, no gimped cores, same turbo, no fucked pcie 2 lanes >.> stares at 1156s) after OC it becomes 1055T vs 920/930 and in that comparision, if i was building new I'd go 1055T every way for the savings to put towards a better gfx. for me personally tho, not so much for anyone else with bigger budgets.


but i have to say, remember that intel can release a 300 dollar 6c/12t with turbo i7-935 and say $500 i7-965 that can simply blow things away. will they do it is another issue tho. Or if you are an intel fanboi, believe that intel can bring in 8 core neth-ex (without the added QPI links) to the desktop on demand and completely crush amd's higher end again.
a b à CPUs
April 30, 2010 4:29:39 PM

jennyh said:

The point is, the xbit review is a sham review because the author had already made up their mind long before the actual chip was released.


Why are you bashing him now instead of 6 months ago?
a b à CPUs
April 30, 2010 4:38:07 PM

What makes you think i wasn't bashing him 6 months ago?
a b à CPUs
April 30, 2010 4:41:08 PM

And come on fazers.

Do you actually believe GF would put all that effort into perfecting their 45nm for what you call a 'stop gap' cpu? That doesn't make sense.
a b à CPUs
April 30, 2010 5:08:51 PM

jennyh said:
Wow you totally avoided the point altogether there didn't you fazers?

The point is, the xbit review is a sham review because the author had already made up their mind long before the actual chip was released.

So that's one less anti-thuban to subtract from the legions of pro-thuban reviews out there. :) 


No, I rebutted your point that it was the XBitLabs article that made me think Thuban wasn't going to be as spectacular as you are spending far too much time trying to make it appear to be.

C'mon - look at the facts: AMD has confirmed there are no core tweaks to Thuban - it's basically a Phenom 2 quad with 2 additional cores, no additional L3 cache, but with low-K dielectric so as to reduce power consumption as well as a sort of 3-on/3-not-quite-so-on Turbo feature.

Taking your Neoseeker benchmark analysis from the other thread, the 1090T was 27.5% ahead of the i7-920 in the highly threaded benchmarks. Seeing as how all the cores were being used, I doubt think either CPU was in turbo mode, so comparing them IPC-wise they would be about equal, since the Thuban has a 23% higher base clock rate. So in other words, it takes 6 physical cores for AMD to match 4 physical + 4 'fake' cores (which offer a range of -10 to +25% advantage depending on the app being run, according to the AMDZone "experts" like Scientia).

In the gaming & other low-threaded apps, it seems Thuban is about the equal of the 920, despite the 600MHz stock clock advantage and the 100MHz turbo clock advantage, which doing a simple average is a 650MHz total clock advantage.

If turbo is turned off and both Thuban & the 920 oc'd to 4GHz, then it appears the 920 will beat Thuban in most of the benchies. Of course, very few users oc.

My point is - it took AMD 18 months to catch up to the lowest-speed Nehalem, at the expense of far larger die size and 50% more physical cores. The latter is a losing game for AMD since Intel is a generation ahead on process and can cram a higher number of more efficient cores onto a given die size than AMD can. Yes it does appear that AMD gets its CPUs fabbed cheaply from GF nowadays, but that is a deal that cannot last forever as GF has to spend the same (or more, since Intel is a large shareholder in some of the fab equipment manufacturers and probably gets something of a discount) on capital equipment during the process node upgrades.

So IMO Bulldozer is the make-or-break CPU for AMD and that's where they are spending most of their time & dollars. If BD lives up to the hype, I'll switch to AMD for my next build. However Thuban is simply not a compelling choice for me.
a b à CPUs
April 30, 2010 5:15:09 PM

jennyh said:
And come on fazers.

Do you actually believe GF would put all that effort into perfecting their 45nm for what you call a 'stop gap' cpu? That doesn't make sense.


No - I'm sure GF is working on yields for their own profitability and to fill up the Dresden fab with orders. Isn't it about 30% underutilized as of last quarter? I seem to recall somebody mentioning that during the AMD Q1 statement thread.

It was AMD's design effort that I was referring to, when I said IMO it was a stopgap CPU to tide people over until Bulldozer.

BTW, did you see the report yesterday about TSMC making a large jump in their quarterly profits for Q1? I guess it's all the other foundry work they do and not so much the 40nm yield issue for AMD & nVidia GPUs...
a b à CPUs
April 30, 2010 5:15:33 PM

Why is it always 'because intel COULD' fazers?

If intel COULD they WOULD. Yields are pathetic on Gulftown thats why you can't get hold of them. Why do you think Apple are so upset with intel too? A total lack of 32nm chips except the *tiny* clarkdales.

No fazers, intel *can't* just make mainstream 6 cores - at least not profitably. If they could, they would.
a b à CPUs
April 30, 2010 5:37:53 PM

jennyh said:
Why is it always 'because intel COULD' fazers?

If intel COULD they WOULD. Yields are pathetic on Gulftown thats why you can't get hold of them. Why do you think Apple are so upset with intel too? A total lack of 32nm chips except the *tiny* clarkdales.

No fazers, intel *can't* just make mainstream 6 cores - at least not profitably. If they could, they would.


Hmm, the two Intel engineers who post here unofficially, Sonoran and Archibael, have both disputed your claim about 32nm being "pathetic" for Intel - I tend to believe them since they are in a far better position to actually know rather than make a lot of assumptions. For one thing, I think it was Sonoran who mentioned that during the 32nm ramp-up they have to prioritize which CPUs to make, and that is based on a market analysis and the decisions have a rather lengthy lead time of 6 months or more.

Where do you get that "Apple are so upset with Intel"? All I've seen are a bunch of rumors based on some AMD execs riding on an Apple bus around Cupertino, or some such. Could be that their 1970's Gremlin broke down and they had to thumb a ride with what turned out to be the Apple bus :D . Or it could be that AMD has shown some ES samples to Apple and they are interested.

But, let's turn the tables, shall we?? Since Thuban is basically just a P2 with 2 extra cores, why did it take AMD a year or more to produce it, esp. after Istanbul which is also 45nm and 6 cores?

OK, this is starting to degenerate into hyperbole and ridiculousness, instead of my newfound persona of exuding thoughtful insight :p , so pardon me while I take a tranquilizer and meditate upon my navel lint a while. I suggest you do the same, OK?? :kaola: 
April 30, 2010 5:52:34 PM

jennyh said:
Why is it always 'because intel COULD' fazers?

If intel COULD they WOULD. Yields are pathetic on Gulftown thats why you can't get hold of them. Why do you think Apple are so upset with intel too? A total lack of 32nm chips except the *tiny* clarkdales.

No fazers, intel *can't* just make mainstream 6 cores - at least not profitably. If they could, they would.


Gulftown would not have made it to market if the yields were "pathetic", because they indeed would have been less profitable were that the case and a Return on Investment calculation would have been done and determined that there was better money to be made making more Arrandale and Clarkdale.

Who can't get hold of them? I see them on Newegg right now.

It's a premium part, and it costs more to reflect 1) the expected price people would be willing to pay to obtain it and 2) the R&D which had to go into producing it. But don't fool yourself into thinking yields drive up the costs on this chip.
April 30, 2010 6:10:08 PM

jennyh said:
That very article was the basis for people like Fazers, jimmysmitty and Chad to declare Thuban would be no better than the i5 750.

Jenny you insane liar, where/when did I ever predict the Thuban would be no better than the i5-750?

Jenny, do you feel embarrassed for saying that with Thuban, AMD's share price would overtake Intel's within one week of Thuban's release? :lol: 

Quote:

we expect AMD Thuban processors to work at about 3.0 GHz. And as we have seen during our today’s test session, it won’t be enough for them to outperform neither competitor’s six-core solutions, nor top quad-core Nehalem based CPUs, which cope perfectly well with heavy multi-threaded load due to Hyper-Threading technology support.

This quote turned out to be perfectly true, so what are you whinging about?

a c 99 à CPUs
April 30, 2010 6:10:32 PM

jennyh said:
Why is it always 'because intel COULD' fazers?

If intel COULD they WOULD. Yields are pathetic on Gulftown thats why you can't get hold of them. Why do you think Apple are so upset with intel too? A total lack of 32nm chips except the *tiny* clarkdales.

No fazers, intel *can't* just make mainstream 6 cores - at least not profitably. If they could, they would.


fazers_on_stun said:
Hmm, the two Intel engineers who post here unofficially, Sonoran and Archibael, have both disputed your claim about 32nm being "pathetic" for Intel - I tend to believe them since they are in a far better position to actually know rather than make a lot of assumptions. For one thing, I think it was Sonoran who mentioned that during the 32nm ramp-up they have to prioritize which CPUs to make, and that is based on a market analysis and the decisions have a rather lengthy lead time of 6 months or more.


You are both correct in part. Sonoran's commentary about Intel being early in their 32 nm fab ramp-up and needing to prioritize what chips get made on 32 nm means that Intel can't make mainstream 32 nm chips. It's not that they are unable to make said chips (they are making bigger and smaller ones for revenue, so the process obviously works), they simply don't yet have as many fab lines as they would need to yield those chips in the necessary quantities. I couldn't tell you what yields on Gulftowns/Westmeres are, but they will be lower than 45 nm yields simply because it's a new process and hasn't had time to be optimized like 45 nm has. That's a well-known phenomena that affects everybody, not just Intel. AMD didn't roll out 45 nm to their entire chip line all at once, either.

Quote:
Where do you get that "Apple are so upset with Intel"? All I've seen are a bunch of rumors based on some AMD execs riding on an Apple bus around Cupertino, or some such. Could be that their 1970's Gremlin broke down and they had to thumb a ride with what turned out to be the Apple bus :D . Or it could be that AMD has shown some ES samples to Apple and they are interested.


Apple would be stupid to not occasionally talk to AMD and "investigate" using their products. At the very least, it would allow them to get a better bargaining position with Intel when they negotiate the next order for CPUs.

Quote:
But, let's turn the tables, shall we?? Since Thuban is basically just a P2 with 2 extra cores, why did it take AMD a year or more to produce it, esp. after Istanbul which is also 45nm and 6 cores?


I'll bet it has to do with AMD doing a new stepping for the Magny-Cours server CPU. MC was a long time coming since it was a major platform change for AMD. The dies in the Magny-Cours are virtually identical to the Thuban die with DDR3 support and the D1 stepping. They also were released quite recently as well (March 29th.) The D0 Istanbul had about a 140 W TDP at 2.8 GHz. The new-stepping D1 dies are able to do 3.2 GHz at a 125-watt TDP envelope, which is far more competitive than a 2.8 GHz, 140-watt unit would have been. Plus, I bet AMD's yields have improved over the yields they initially had when Istanbul was released, so they could release at a lower price and still make money.
a c 111 à CPUs
April 30, 2010 6:19:53 PM

That xBit review/comparison blew chunks when it came out --- and looks even more silly today.

Don't know what they could have been thinkin' ...
a b à CPUs
April 30, 2010 7:04:54 PM

fazers_on_stun said:
No, I rebutted your point that it was the XBitLabs article that made me think Thuban wasn't going to be as spectacular as you are spending far too much time trying to make it appear to be.

C'mon - look at the facts: AMD has confirmed there are no core tweaks to Thuban - it's basically a Phenom 2 quad with 2 additional cores, no additional L3 cache, but with low-K dielectric so as to reduce power consumption as well as a sort of 3-on/3-not-quite-so-on Turbo feature.

Taking your Neoseeker benchmark analysis from the other thread, the 1090T was 27.5% ahead of the i7-920 in the highly threaded benchmarks. Seeing as how all the cores were being used, I doubt think either CPU was in turbo mode, so comparing them IPC-wise they would be about equal, since the Thuban has a 23% higher base clock rate. So in other words, it takes 6 physical cores for AMD to match 4 physical + 4 'fake' cores (which offer a range of -10 to +25% advantage depending on the app being run, according to the AMDZone "experts" like Scientia).

In the gaming & other low-threaded apps, it seems Thuban is about the equal of the 920, despite the 600MHz stock clock advantage and the 100MHz turbo clock advantage, which doing a simple average is a 650MHz total clock advantage.

If turbo is turned off and both Thuban & the 920 oc'd to 4GHz, then it appears the 920 will beat Thuban in most of the benchies. Of course, very few users oc.

My point is - it took AMD 18 months to catch up to the lowest-speed Nehalem, at the expense of far larger die size and 50% more physical cores. The latter is a losing game for AMD since Intel is a generation ahead on process and can cram a higher number of more efficient cores onto a given die size than AMD can. Yes it does appear that AMD gets its CPUs fabbed cheaply from GF nowadays, but that is a deal that cannot last forever as GF has to spend the same (or more, since Intel is a large shareholder in some of the fab equipment manufacturers and probably gets something of a discount) on capital equipment during the process node upgrades.

So IMO Bulldozer is the make-or-break CPU for AMD and that's where they are spending most of their time & dollars. If BD lives up to the hype, I'll switch to AMD for my next build. However Thuban is simply not a compelling choice for me.


+1
a c 127 à CPUs
April 30, 2010 7:39:25 PM

jennyh said:
The pre-review "Thuban is instanbul for desktop"

Remember that from 6 months back? How the author oh so confidently told us

Quote:
we expect AMD Thuban processors to work at about 3.0 GHz. And as we have seen during our today’s test session, it won’t be enough for them to outperform neither competitor’s six-core solutions, nor top quad-core Nehalem based CPUs, which cope perfectly well with heavy multi-threaded load due to Hyper-Threading technology support.


I guess if you already have stuck your neck on the line 6 months before the chip is even available...

The real review of Thuban


Same website, same author. I don't know what is up at xbit but at the very least a different reviewer should have reviewed the actual chip after the joke that was the "istanbul for desktop" article. That very article was the basis for people like Fazers, jimmysmitty and Chad to declare Thuban would be no better than the i5 750. :lol: 

As usual, you need to be very careful about who you believe. Some websites cant be trusted - xbit has proven itself untrustworthy over the recent months.


Um hey, don't throw me in there I said on par with a Core i7 930 and I am not that far off. Most of the reviews show the 930 and the 1090T trading blows both at stock and both OCed and in most cases, are on par.

But the reviewer was right on one part. Thuban cannot beat Gulftown. Thats a true fact. In most cases a stock 980X beats a 4GHz Thuban.

I do love your guesses on Intels yields.

http://www.newegg.com/Product/Product.aspx?Item=N82E168...

That alone says its available.

So I guess now you will go through every review not praising Thuban as an amazing chip and try to discredit the authors?

BTW, you can't get mad at an author for making a prediction. They do it all the time.
a b à CPUs
May 1, 2010 12:24:40 AM

fazers_on_stun said:

AMD took some economies with Thuban - same size L3 cache despite 50% extra cores, rather simplistic turbo solution, no improvement in memory bandwidth, no IPC improvements, etc. To me, that indicates AMD did a quick, minimal job with it in order to keep the AMD loyal in their camp until BD arrives. .


This is really kind of silly statement. You obviously haven't followed processor development much over the years. This type of incremental improvement is par the course from intel and AMD.

Far from have taken "economies" or "quick minimal" as you put it, the addition of even a primitive turbo function to an architecture not designed with it in mind was a surprise. Memory bandwidth improvements would not be expected at this stage, nor would core level IPC improvements. L3 cache increase was precluded by the size of the die at 45nm.

This is the type of post we see too often. People just make stuff and post it like its fact. I don't get it.





a b à CPUs
May 1, 2010 12:36:00 AM

FALC0N said:
L3 cache increase was precluded by the size of the die at 45nm.

This is the type of post we see too often. People just make stuff and post it like its fact. I don't get it.


Do You have a link to back that statement up ? Smaller cache size or less is seen in both AMD and Intel chips up and down the product lines. And all else being similar more expensive cpu's have more cache and the lower variants less.
Celerons and Athlons.
e8400's vs e7400.
a b à CPUs
May 1, 2010 1:11:14 AM

FALC0N said:
This is really kind of silly statement. You obviously haven't followed processor development much over the years. This type of incremental improvement is par the course from intel and AMD.

Far from have taken "economies" or "quick minimal" as you put it, the addition of even a primitive turbo function to an architecture not designed with it in mind was a surprise. Memory bandwidth improvements would not be expected at this stage, nor would core level IPC improvements. L3 cache increase was precluded by the size of the die at 45nm.

This is the type of post we see too often. People just make stuff and post it like its fact. I don't get it.


And of course that is merely your opinion. You should indicate it as such, as I did mine.

If the 45nm die size is such a constraint, then explain Magny Cours with its whopping 346 x 2 mm die size. Clearly AMD could have made a bigger die with more L3, although they would also likely have to raise the price. So, yes - AMD took economies with the design. And I'm not the only one with this opinion - from http://www.brightsideofnews.com/news/2009/5/14/amds-mag...:

Quote:
So, is AMD architecturally doing a better job than Intel for these MCMs? Let's look at what we should be having here: The G34 substrate features two 45nm dies - each with six cores and private 512K L2 per core, sharing a - probably undersized - 6M L3 cache per die. I feel that AMD, having been on the forefront of dense cache memory cell technologies [anyone remember Z-RAM?], should have added a bit more here when they decided to jump from 4 to 6 cores. Remember, Core i7 and Gainestown Nehalems [Nehalem-EP] have 8 MB L3 per four cores, fed by a 216-bit Triple-Channel DDR3 while Beckton [Nehalem-EX] will have 8 cores and 24 MB L3, fed by a 288-bit Quad-Channel memory controller. So, for more memory intensive threads, where Istanbul has 6 MB L3 per 6 cores, and fed just by 144-bit Dual-Channel DDR3, the cache increase would have been helpful as long as the latency is managed.


Perhaps you should take note that Thuban uses the same old 3-issue K8 core, albeit with a few tweaks, which means they have been sitting on that design since what - the last 8 years now? So yes most people would have expected some core-level IPC improvements by now - after all, Intel went from P4 to 4-issue Core2 during that time period, and Sandy Bridge is also expected to make some significant core changes. If SB makes it to market before Bulldozer, that would be two major IPC improvements for Intel and zero for AMD in roughly 9 years time.

I would hardly call the 30+% IPC improvement of Core2 "incremental", but then I'm not an AMD-favoring history revisionist either.

And seriously - a bolt-on turbo function is a surprise to you? What I find surprising is how long it took AMD to release Thuban, in view of Istanbul coming out early last summer, and yes I have heard about validation cycles.

You are relatively new here, but already you are coming across as yet another AMD know-it-all who thinks he's a walking encyclopaedia of CPU facts & history, which is another thing we are seeing too often. You obviously have two very different standards of performance - a low one for AMD and likely an impossible one for Intel.
a b à CPUs
May 1, 2010 1:16:35 AM

notty22 said:
Do You have a link to back that statement up ? Smaller cache size or less is seen in both AMD and Intel chips up and down the product lines. And all else being similar more expensive cpu's have more cache and the lower variants less.
Celerons and Athlons.
e8400's vs e7400.


This is what I am talking about! Nobody who follows chip development would have to ask. These types of incremental development are common.

I will give you two examples from each manufacturer.

Intel --

1) The I7 980X! Its nothing more than a I7 quad with 2 additonal cores and extra L3 which was only possible due to the die shrink.

2) Core 2 quads. Two core 2 quads on a single die, talking though the northbridge.

AMD

1) Single core A64 to the X2. No significant changes except 2 integrated cores on a single die.

2) Agena Phenoms to Deneb Phenoms. This was nothing but a die shrink.


You go on for hours if you really wanted to. This is how chips usually evolve. A little bit at a time.
a b à CPUs
May 1, 2010 1:25:31 AM

FALC0N said:
This is what I am talking about! Nobody who follows chip development would have to ask. These types of incremental development are common.

I will give you two examples from each manufacturer.

Intel --

1) The I7 980X! Its nothing more than a I7 quad with 2 additonal cores and extra L3 which was only possible due to the die shrink.

2) Core 2 quads. Two core 2 quads on a single die, talking though the northbridge.

AMD

1) Single core A64 to the X2. No significant changes except 2 integrated cores on a single die.

2) Agena Phenoms to Deneb Phenoms. This was nothing but a die shrink.


You go on for hours if you really wanted to. This is how chips usually evolve. A little bit at a time.

You had a hissy fit about people 'making up things' and then and I may add poorly tried to reason what AMD did or didn't do , on three points with Thuban. When asked to back that up, with links you can't.
So all you have is your opinion, after whining about other people. Hypocrite.
a b à CPUs
May 1, 2010 1:51:45 AM

fazers_on_stun said:
If the 45nm die size is such a constraint, then explain Magny Cours with its whopping 346 x 2 mm die size. Clearly AMD could have made a bigger die with more L3, although they would also likely have to raise the price. So, yes - AMD took economies with the design. And I'm not the only one with this opinion - from http://www.brightsideofnews.com/news/2009/5/14/amds-mag...:


Once again simple common sense and logic evades you Fazers.

Die size isn't the contraint, if anything it's TDP. The 1090T is a 125w CPU at 3.2ghz and 6 cores. Magny cours is 130w 12 cores at 2.3ghz (or thereabouts). Do I need to remind you about Gulftowns 6 32nm cores at 140w TDP btw?

One of these makes a good desktop cpu, the other makes a good server cpu - and it has got nothing to do with die size constraints.
a b à CPUs
May 1, 2010 2:29:05 AM

fazers_on_stun said:
And of course that is merely your opinion. You should indicate it as such, as I did mine.

If the 45nm die size is such a constraint, then explain Magny Cours with its whopping 346 x 2 mm die size. Clearly AMD could have made a bigger die with more L3, although they would also likely have to raise the price. So, yes - AMD took economies with the design. And I'm not the only one with this opinion - from http://www.brightsideofnews.com/news/2009/5/14/amds-mag...:

No its not my opinion. It would have been foolish to add L3 cach at this die size. With 12 megs of L3 cache, the thuban would be almost 450 mm! And thats less than 10% speed improvement.

350 mm is a huge die as it is. Die size affects power usage, cost and yield amongst other concerns. The problems presented by the additonal die space needed far far outweigh the small speed gain.

If you think this was a viable option on a desktop chip, then you don't understand the technology.

Quote:

Perhaps you should take note that Thuban uses the same old 3-issue K8 core, albeit with a few tweaks, which means they have been sitting on that design since what - the last 8 years now? .


Actually, the K8 is nothing more than a K7 with 64 bit instruction set and an on die memory controller.

And they have made core level IPC improvements in that time. Don't know where you get the idea they didn't but they have been INCREMENTAL.

Quote:
Intel went from P4 to 4-issue Core2 during that time period, and Sandy Bridge is also expected to make some significant core changes..........................................I would hardly call the 30+% IPC improvement of Core2 "incremental", but then I'm not an AMD-favoring history revisionist either.

No your not an AMD favoring revisionist. Unfortunately, your not well versed on the topic either.

Core 2 didn't just magically appear in 2006. It was developed INCREMENTALLY as an extension of the P3 architecture and later Pentium M for laptops and was simply moved to the desktop.

It appears to be a huge jump ONLY because it replaced Netburst. But if you follow its parallel development history, it improved INCREMENTALLY.

Quote:
And seriously - a bolt-on turbo function is a surprise to you? What I find surprising is how long it took AMD to release Thuban


Funny that no knowledgeable source shared this view. Like I said, you don't understand the technology.
Quote:

You are relatively new here, but already you are coming across as yet another AMD know-it-all who thinks he's a walking encyclopaedia of CPU facts & history......


Your being wrong doesn't make me a walking encyclopedia. But thanks anyways for the complement. :) 
a b à CPUs
May 1, 2010 2:37:16 AM

notty22 said:
You had a hissy fit about people 'making up things' and then and I may add poorly tried to reason what AMD did or didn't do , on three points with Thuban. When asked to back that up, with links you can't.
So all you have is your opinion, after whining about other people. Hypocrite.


Wait a second notty22. You could post links supporting your position. Instead you simply demand that I support my position instead of supporting your own. So if I'm a hypocrite, you are too by that logic.

Maybe I'm just more knowledgeable than I give myself credit for. But the incremental development of processors is not a state secret. Its common knowledge. Forgive me for being reluctant to spend the time taking you step by step by step though the history of the cpu. Your more than able to do the research on your own.
a b à CPUs
May 1, 2010 2:35:37 PM

FALC0N said:
No its not my opinion. It would have been foolish to add L3 cach at this die size. With 12 megs of L3 cache, the thuban would be almost 450 mm! And thats less than 10% speed improvement.

350 mm is a huge die as it is. Die size affects power usage, cost and yield amongst other concerns. The problems presented by the additonal die space needed far far outweigh the small speed gain.

If you think this was a viable option on a desktop chip, then you don't understand the technology.


Funny, apparently Intel with the i7-980X, with 6 cores and 12 megs L3, doesn't "understand the technology" either :sarcastic:  And yes I know it's a process node ahead of Thuban.

Quote:
And they have made core level IPC improvements in that time. Don't know where you get the idea they didn't but they have been INCREMENTAL.


Hmm, I guess you didn't understand what was meant by "with a few tweaks" which is how I referred to these incremental core improvements.

Quote:
No your not an AMD favoring revisionist. Unfortunately, your not well versed on the topic either.


Hmm, attempting a rebuttal with a mere unsupported personal conclusion. How about actually addressing the points next time, eh?

Quote:
Core 2 didn't just magically appear in 2006. It was developed INCREMENTALLY as an extension of the P3 architecture and later Pentium M for laptops and was simply moved to the desktop.

It appears to be a huge jump ONLY because it replaced Netburst. But if you follow its parallel development history, it improved INCREMENTALLY.


It appears your definition of "incremental" is one of convenience, changing to suit your arguments. Personally I consider an improvement in the single digits to be "incremental" whereas one in double digits or more would be significant, and of use as a defining moment in CPU development. Sorta like the Intel 4004 which you so blithely dismiss as of little importance :p , despite the fact that before same there were no complete general purpose microprocessors commercially available. I guess that was an "incremental" improvement by your definition as well - let's see, zero IPC to what, 90K? Yep, that's merely incremental :D .

Quote:
Funny that no knowledgeable source shared this view. Like I said, you don't understand the technology.


Again, you should post some evidence rather than just resorting to a dismissive insult. We veterans see far too much of that here on THG :p .

But since this thread has already quoted from Brightside, why not continue the tradition?

Quote:
Your being wrong doesn't make me a walking encyclopedia. But thanks anyways for the complement. :) 


You're quite welcome :D 
a b à CPUs
May 1, 2010 2:51:48 PM

jennyh said:
Once again simple common sense and logic evades you Fazers.

Die size isn't the contraint, if anything it's TDP. The 1090T is a 125w CPU at 3.2ghz and 6 cores. Magny cours is 130w 12 cores at 2.3ghz (or thereabouts). Do I need to remind you about Gulftowns 6 32nm cores at 140w TDP btw?

One of these makes a good desktop cpu, the other makes a good server cpu - and it has got nothing to do with die size constraints.


Ahh, Jenny - didn't we flog the old ACP vs. TDP debate in countless other threads here :p . C'mon - it's a beautiful day here and too nice to waste. Not to mention I'm quite busy evading that "common sense & logic" which is lunging at me like a runaway behemoth :D .

Even if for argument purposes I were to concede that yes, the 980 always runs at 140W and the 1090T at 125W, let's figure out the actual electricity costs of that difference, OK? Lessee now, assuming a desktop computer to be running 8 hours per day, maybe 6 days a week (even fanboys have to take the day off occasionally :p ), times 50 weeks per year (again, even fanboys take vacations :p ). that's 36 kilowatt-hours of extra energy usage per year. Dunno about where you live, but here in Northern Virginia, Dominion Electric charges me something like 10 cents per kilowatt-hour. Probably less than that but I'm too lazy to go look up my last electric bill :D 

So that's a whopping thee dollars and 60 cents extra per year. About the cost of a Starbucks medium latte. And before you say "What about the cooling costs for that added heat", let me state that half of the year here, the gas heat is on so I'm thankful for the extra free heat since the cost has already been amortized by the computer usage. Such as it is :D .

And I could always open the window during the summer :p .

Bottom line - all I hafta do is cut down on those lattes by one per year and open the window, and boom! - the electricity differential is covered :D .

a b à CPUs
May 2, 2010 2:04:23 AM

Looks like a factual review to me (pretty realist) and mirrors the majority of the results out there.

:) 
a b à CPUs
May 2, 2010 7:48:41 PM

fazers_on_stun said:
Funny, apparently Intel with the i7-980X, with 6 cores and 12 megs L3, doesn't "understand the technology" either :sarcastic:  And yes I know it's a process node ahead of Thuban.


Intel understands it just fine. YOU don't. Notice how you just causally add at the end that you know its a smaller node. It NOT enough to know. You must understand what it means, and you just don't.

The 980X is only a 250 mm die at 32nm. You can claim you understand until your blue in the face but you would not posted this argument if you did.

Its the chip equivalent of saying "2X2=4 and 2+2=4 therefore 4+4 and 4X4 must also equal the same number". It only sounds good IF you don't what your doing.

Quote:
Hmm, attempting a rebuttal with a mere unsupported personal conclusion. How about actually addressing the points next time, eh?


Your failure to understand the topic is not a personal conclusion. I have demonstrated it several times and its obvious to any person who does understand the topic.

And here is a great example of once again, you just making stuff up. I DID address your point. The WHOLE answer, instead of you cherry picked sentence read:

"No your not an AMD favoring revisionist. Unfortunately, your not well versed on the topic either.

Core 2 didn't just magically appear in 2006. It was developed INCREMENTALLY as an extension of the P3 architecture and later Pentium M for laptops and was simply moved to the desktop.

It appears to be a huge jump ONLY because it replaced Netburst. But if you follow its parallel development history, it improved INCREMENTALLY. "


Looks like I addressed your point just fine.

Quote:
It appears your definition of "incremental" is one of convenience, changing to suit your arguments. Personally I consider an improvement in the single digits to be "incremental" whereas one in double digits or more would be significant, and of use as a defining moment in CPU development. Sorta like the Intel 4004 which you so blithely dismiss as of little importance :p , despite the fact that before same there were no complete general purpose microprocessors
commercially available. I guess that was an "incremental" improvement by your definition as well - let's see, zero IPC to what, 90K? Yep, that's merely incremental :D .


My definition of incremental is to follow the technical development of the chip. Core 2 for example wasn't new. It was just new to desktops. To a layperson, Its a huge jump. But technically, it had been around for years.

By the way, Thuban is double digit. So was the Phenom I, II, I7, C2Q, C2D, Pentium D, X2. All in the last 5 years. Seems like a lot of defining moments to me. But particular topic is indeed a matter of opinion.

One chip that DOESN'T meet your criteria was the 4004. It was slower and more expensive than similar offerings of the time. It WASN'T the first cpu. It was the first all-in-one cpu. Good on paper, not so good in practice.

As I said the first time we discussed this, its an interesting historical note. But a la Phenom I and its first integrated quad, it wasn't nearly as significant as it looked on paper. In fact, the phenom has fared much better than the 4004 ever did.

Quote:
Again, you should post some evidence rather than just resorting to a dismissive insult. We veterans see far too much of that here on THG :p .

But since this thread has already quoted from Brightside, why not continue the tradition?


You don't understand the technology Fazer. Its apparent with almost every post. You double back over your own arguments without even realizing it and defend your untenable positions like your in a life of death struggle.



a b à CPUs
May 2, 2010 7:53:48 PM

fazers_on_stun said:
..............................

Bottom line - all I hafta do is cut down on those lattes by one per year and open the window, and boom! - the electricity differential is covered :D .


Maybe if your a business with 1000 units running 24/7 it might matter. But im with you here. For home users, power savings just isn't going to matter much.
a b à CPUs
May 2, 2010 8:10:13 PM

i think everybody is goin off topic here, point is jenny was right about the review, it seemed biased, and wrong, which seems to be the intial concern. also, yes, the thuban is just to shut amd customers up until bulldozer, which seems to be crucial for amd, so obviously the 1090T is just a mid point, and is just a 955+2 cores...the other things you are arguing about are goin all over the place. IMO, the 1090T is an overall better cpu than the i7 930, but they really do trade blows in most things, gaming going to i7, raw power to 1090T. however, considering the 1090T is cooler, cheaper (overall), much more future-proof, and a few other things, gives the advantage to the 1090T by just a little, IMO. also, shouldnt we be mentioning that amd now has a 200$ 6 core, where intel can barely get a 200$ 4 core. im no amd fanboy, but im gonna stick with the $700+ cheaper cpu, and the company that doesnt change sockets every christmas...
a b à CPUs
May 2, 2010 8:19:14 PM

ares1214 said:
i think everybody is goin off topic here, point is jenny was right about the review, it seemed biased, and wrong, which seems to be the intial concern. also, yes, the thuban is just to shut amd customers up until bulldozer, which seems to be crucial for amd, so obviously the 1090T is just a mid point, and is just a 955+2 cores...the other things you are arguing about are goin all over the place. IMO, the 1090T is an overall better cpu than the i7 930, but they really do trade blows in most things, gaming going to i7, raw power to 1090T. however, considering the 1090T is cooler, cheaper (overall), much more future-proof, and a few other things, gives the advantage to the 1090T by just a little, IMO. also, shouldnt we be mentioning that amd now has a 200$ 6 core, where intel can barely get a 200$ 4 core. im no amd fanboy, but im gonna stick with the $700+ cheaper cpu, and the company that doesnt change sockets every christmas...


Your first sentence, that might be the ops claims, but its baloney. Haven't you followed her reasoning, she feels the same way about Anands review to, so basically anything but glowing praise is labeled, bias and wrong ? I've yet to see ANY proof about something being 'wrong' with that review.
a b à CPUs
May 2, 2010 8:29:56 PM

im not saying theres nothing wrong with the 1090T, theres plenty wrong, like not very good amount of L3 cache, its 45 nm, oc'ing hasnt been very good, and on and on. however, what they predicted was wrong, then what other people were saying about the i5 being better was wrong, which i believe, unless jenny meant otherwise, is what she was getting at. also, xbit isnt biased, or even biased, however i think they worded what they said wrong, and the author of that one might be a bit biased. also not the smartest thing to make a prediction 6 months in advance, when there is hardly any info on any of the cpus out yet. and no, i dont argue the intel cpus beat the 1090T, nor do i dispute any benchmark that says that, however, it seems its split in bench's, but due to reasons previously stated, i give the slight advantage to the 1090T.
a b à CPUs
May 2, 2010 8:58:17 PM

Every day countless tech sites report on tech news, they speculate on press releases and road maps and then they eventually also test hardware. The op is spinning this to look like some kind of journalistic CRIME. I don't see ANY hint of bias, that YOUR claiming as well. There is nothing in that review that sounds like, I told you so, as reported here, I,we knew . The op is planting this scenario that the reporter is desperately trying to make his predictions come true, and I don't see it. Other review sites also came to the same conclusions as him. Those sites would have also talked about Thuban numerous times in earlier reports.

a b à CPUs
May 2, 2010 9:36:11 PM

jennyh said:
Why is it always 'because intel COULD' fazers?

If intel COULD they WOULD. Yields are pathetic on Gulftown thats why you can't get hold of them. Why do you think Apple are so upset with intel too? A total lack of 32nm chips except the *tiny* clarkdales.

No fazers, intel *can't* just make mainstream 6 cores - at least not profitably. If they could, they would.

Don't forget the i7 apple laptops are running up to 107 degrees, thats gotta make people happy, 3rd degree burns from setting a laptop on your lap.

The 1090T and the 1055T are not ground breaking except for the fact that its 2 extra cores at the same power consumption, same heat, higher OC possibilities, ...

The only thing Intel support can come up with is "its still not as fast as our $1100 980X. No one expected it to be. Yes the cpu can be seen as a stopgap, but its improvements are huge where they are.

New 45nm >> old 45nm. The 965 C3 (after being out 2 months?)could barely hit 4.1 with a lucky chip. the 1090T are hitting 4.4 stable and up with a lucky chip within the first week. I used the term lucky describing a batch that will hit those speeds on good air cooling, and the fact that not all chips will oc the same.

The only thing fanboys can counter with is ... "that must be a lie."
a b à CPUs
May 2, 2010 10:27:36 PM

noob2222 said:

The 1090T and the 1055T are not ground breaking except for the fact that its 2 extra cores at the same power consumption, same heat, higher OC possibilities, ...

The only thing Intel support can come up with is "its still not as fast as our $1100 980X. No one expected it to be. Yes the cpu can be seen as a stopgap, but its improvements are huge where they are.


Few chips are truly groundbreaking though. They evolve a little bit at a time. New stepping, extra cores, new special instructions, more cache, etc. The big jumps are usually a matter of perception and not chip design leaps.

The I7/I5 for example were little more than a core 2 quad with all four cores integrated, on die memory control, and hyperthreading. All technology that was almost a decade old. Which is an eternity in the computer world. Only the turoboost was new.

However, given all the trouble AMD had from 2006 to 2008 with the low performing brisbane die shrink and the delays and unexpectedly low performance with the release of the Agena quads, its nice to see them get a couple of solid release in a row with Deneb and Thuban. Competition breeds lower prices and more options and thats good for everyone.

a b à CPUs
May 4, 2010 11:26:47 PM

FALC0N said:
Intel understands it just fine. YOU don't. Notice how you just causally add at the end that you know its a smaller node. It NOT enough to know. You must understand what it means, and you just don't.

The 980X is only a 250 mm die at 32nm. You can claim you understand until your blue in the face but you would not posted this argument if you did.

Its the chip equivalent of saying "2X2=4 and 2+2=4 therefore 4+4 and 4X4 must also equal the same number". It only sounds good IF you don't what your doing.


No, you are mistaking what is possible with what is practical. It is possible for AMD to maintain the L3/core ratio as in the P2, with comcomitant performance increase, but they chose not to for reasons of economy. Ergo - AMD took economies with Thuban. You can argue out of both ends all you like, but that doesn't change the fact that AMD could have done it but didn't.

As for the 450mm die size, just take a look at Intel's Itanium with its 2B+ transistors to see what is doable, if you're willing to set pricing accordingly. Another "economy" argument, in addition to the 980 keeping the same L3 cache/core ratio. And as I linked to previously, I'm not the only one with this opinion. I notice you don't provide any countervailing links - just some ad hominem.

Quote:
Your failure to understand the topic is not a personal conclusion. I have demonstrated it several times and its obvious to any person who does understand the topic.

And here is a great example of once again, you just making stuff up. I DID address your point. The WHOLE answer, instead of you cherry picked sentence read:

Core 2 didn't just magically appear in 2006. It was developed INCREMENTALLY as an extension of the P3 architecture and later Pentium M for laptops and was simply moved to the desktop.

It appears to be a huge jump ONLY because it replaced Netburst. But if you follow its parallel development history, it improved INCREMENTALLY.
Looks like I addressed your point just fine.


IIRC Pentium M and Yonah (Core Duo/Solo) were 3-issue cores whereas Core2/Merom was the first 4-issue core. Do you even know what that entails in terms of checking for code dependencies?

Quote:
My definition of incremental is to follow the technical development of the chip. Core 2 for example wasn't new. It was just new to desktops. To a layperson, Its a huge jump. But technically, it had been around for years.


Umm, riiiiiight. Yep, just because Falcon sez so.

Quote:
By the way, Thuban is double digit.


Really? When the reviews show that its IPC is the same or lower than the P2? Unless you subscribe to the "aggregate IPC" nonsense of course. I guess it's not only me, but most of the review sites who don't know anything either.

Quote:
One chip that DOESN'T meet your criteria was the 4004. It was slower and more expensive than similar offerings of the time. It WASN'T the first cpu. It was the first all-in-one cpu. Good on paper, not so good in practice.

As I said the first time we discussed this, its an interesting historical note. But a la Phenom I and its first integrated quad, it wasn't nearly as significant as it looked on paper. In fact, the phenom has fared much better than the 4004 ever did.


LOL - now it's the history books that are ignorant too. BTW, in case you didn't know - the 4004 was actually 4 separate chips, if by "all-in-one" you mean integrated on one die. And as the first complete microprocessor sold commercially (as opposed to the incomplete ALUs, etc which required dedicated LSI logic), then yes it was a huge jump up from zero IPC.

And seriously - P1? BTW, the preferred tech term is "monolithic", not "integrated" since all quads are "integrated" circuit chips. And if you're attempting to regurgitate the tired old monolithic vs. MCM debate for quads, give it up already. History (and sales) show that AMD was incredibly stupid for going the monolithic route for desktop, giving Intel the entire quad market for some 12+ months, when Core2 showed that communicating through the FSB didn't hinder performance significantly. Yep, P1 certainly was 'significant' alright - made AMD a laughingstock for a couple years there, esp. when their VP was tooting the "40% improvement over Core2 across a wide variety of workloads".

Quote:
You don't understand the technology Fazer. Its apparent with almost every post. You double back over your own arguments without even realizing it and defend your untenable positions like your in a life of death struggle.


And you keep repeating the same old mantra while maintaining your indefensible positions :p . No, I am willing to publicly admit when I'm wrong, which is more frequent that I would like. I doubt we'll ever see anything similar from you, as is apparent from every one of your posts.

What you have to learn here is that your opinion doesn't count for more than anyone else's - it's providing facts backed up with links that is the gold standard here. Just because you, along with Sharikook and other so-called "experts" too numerous to enumerate, hold yourselves out as some self-alleged fonts of tech wisdom, does not make it so.
a b à CPUs
May 5, 2010 1:11:05 AM

fazers_on_stun said:
No, you are mistaking what is possible with what is practical. It is possible for AMD to maintain the L3/core ratio as in the P2, with comcomitant performance increase, but they chose not to for reasons of economy. Ergo - AMD took economies with Thuban. You can argue out of both ends all you like, but that doesn't change the fact that AMD could have done it but didn't.

As for the 450mm die size, just take a look at Intel's Itanium with its 2B+ transistors to see what is doable, if you're willing to set pricing accordingly. Another "economy" argument, in addition to the 980 keeping the same L3 cache/core ratio. And as I linked to previously, I'm not the only one with this opinion. I notice you don't provide any countervailing links - just some ad hominem.


In business, what is impracticable IS impossible. Little Caesars could drop their price for a large pizza from $5 a pizza to $2 and they would soon find themselves out of business.

The Itanium is a failure. Are you sure you want to use that example? Not to mention targeted at a completely different market.

Quote:
Really? When the reviews show that its IPC is the same or lower than the P2? Unless you subscribe to the "aggregate IPC" nonsense of course. I guess it's not only me, but most of the review sites who don't know anything either.


Aggregate IPC is nonsense? Thats exactly how you measure IPC. Its usually measured by the cpu, not the core. Though it can refer to a core if its specified.

Quote:
LOL - now it's the history books that are ignorant too. BTW, in case you didn't know - the 4004 was actually 4 separate chips, if by "all-in-one" you mean integrated on one die. And as the first complete microprocessor sold commercially (as opposed to the incomplete ALUs, etc which required dedicated LSI logic), then yes it was a huge jump up from zero IPC.


No the history books are not ignorant. Your interpretation of the history books is ignorant.

No the 4004 wasn't 4 separate chips. It was one chip. Thats its place in history. IF it was 4 chips, it wouldn't be a microprocessor. That its claim to fame. IT WAS ITS ONLY CLAIM TO FAME. We wouldn't even be talking about it otherwise.

Quote:

.....................What you have to learn here is that your opinion doesn't count for more than anyone else's - it's providing facts backed up with links that is the gold standard here........


Here are a couple of links and quotes from said links on common topic of our discussion.
---------------------------------------------------

http://en.wikipedia.org/wiki/Instructions_per_cycle

Well there you go. Notice how they are describing aggregate IPC.

-------------------------------------------------------------------
The 4004

The Intel 4004.......was the first complete CPU on one chip, and also the first commercially available microprocessor. Such a feat of integration.......


Looks like one chip to me.

http://en.wikipedia.org/wiki/Intel_4004
--------------------------------------------------------------------
The 8008

]the chip was commissioned by Computer Terminal Corporation (CTC) to implement an instruction set of their design for their Datapoint 2200 programmable terminal.....The subsequent Intel 8080 and 8085 CPUs were also heavily based on the same basic design; even the 8086 architecture (a non-strict extension of the 8080) loosely resembles the original Datapoint 2200 design. This not only means that each 8008 instruction has an equivalent in the larger instruction set of the 8080, but also that in most cases an 8080 instruction corresponds directly to one modern x86 instruction (although the encodings are different in all three).

http://en.wikipedia.org/wiki/Intel_8008

ahh yes. The chip that made intel. Notice it say 8008 not 4004.
---------------------------------------------------------------


But the truth Fazer is that none of this matters. If you REALLY wanted to know, you could have looked it up yourself. It took me less than 5 minutes to dig up these links and quotes.

I don't think you really want to know. You confuse what you want to believe with facts. They are not the same thing.
May 5, 2010 1:42:57 AM

Oh god not this again, If i hear thuban or X6 and IPC in the sentance, I'm going to hit something....
a b à CPUs
May 5, 2010 5:54:11 AM

fazers_on_stun said:

Quote:
Your failure to understand the topic is not a personal conclusion.....

Core 2 didn't just magically appear in 2006. It was developed INCREMENTALLY as an extension of the P3 architecture and later Pentium M for laptops and was simply moved to the desktop.

It appears to be a huge jump ONLY because it replaced Netburst. But if you follow its parallel development history, it improved INCREMENTALLY.........Core 2 for example wasn't new. It was just new to desktops. To a layperson, Its a huge jump. But technically, it had been around for years.


Umm, riiiiiight. Yep, just because Falcon sez so.


Really? Lets look at what Hardware secrets had to say:

"The new Intel Core microarchitecture, which.... will be used on all new CPUs from Intel, like Merom, Conroe and Woodcrest. It is based on Pentium M’s microarchitecture..............Pentium M is based on Intel’s 6th generation architecture, a.k.a. P6, the same used by Pentium Pro, Pentium II, Pentium III and early Celeron CPUs and not on Pentium 4’s.......You may think of Pentium M as an enhanced Pentium III. Thus you may think of Core microarchitecture as an enhanced Pentium M."

http://www.hardwaresecrets.com/article/313
-------------------------------------------------------------

"Since all new CPUs from Intel will use Pentium M’s architecture, studying this architecture is very important to understand the......... forthcoming Core microarchitecture"


http://www.hardwaresecrets.com/article/270/1
-------------------------------------------------------------

How about Anandtech:

"Core 2 is clearly a descendant of the Pentium Pro, or the P6........ Those CPUs inherited the bus of the Pentium 4, but are still clearly children of the hugely successful P6 architecture"

http://www.anandtech.com/show/1998/1
-----------------------------------------------------

Thats the difference between us Fazers. I know what I am talking about. You just make up whatever choose to believe.

The funny part is, this particular story is common knowledge. Everyone who really follows cpu development knows the story of how Intel's Israeli design team and the Pentium M bailed intel out of the Netburst nightmare. Everyone except you of course.
May 5, 2010 9:21:11 AM

jennyh said:
Once again simple common sense and logic evades you Fazers.

Die size isn't the contraint, if anything it's TDP. The 1090T is a 125w CPU at 3.2ghz and 6 cores. Magny cours is 130w 12 cores at 2.3ghz (or thereabouts). Do I need to remind you about Gulftowns 6 32nm cores at 140w TDP btw?

One of these makes a good desktop cpu, the other makes a good server cpu - and it has got nothing to do with die size constraints.


Your a little off in the TDP department.


Intel® Core™ i7-980X Processor Extreme Edition (12M Cache, 3.33 GHz, 6.40 GT/s Intel® QPI) No 130 Watts 6C / 12T Yes Yes $999.00 Launched

You can find the information here: http://ark.intel.com/ProductCollection.aspx?familyID=39...
a b à CPUs
May 5, 2010 9:28:13 PM

why does jenny take this stuff so seriously? they are just pieces of sand and plastic........
a b à CPUs
May 5, 2010 10:49:30 PM

dont forget small amounts of metals! :D  that must be it.... :lol: 
a b à CPUs
May 6, 2010 5:58:47 PM

FALC0N said:
In business, what is impracticable IS impossible. Little Caesars could drop their price for a large pizza from $5 a pizza to $2 and they would soon find themselves out of business.

The Itanium is a failure. Are you sure you want to use that example? Not to mention targeted at a completely different market.


Hmm, well AMD has been known for dropping their prices, losing vast sums of money, but still staying afloat. However none of what you posted disputes what I said - AMD took economies with Thuban. All your arguments to the contrary notwithstanding, as is readily apparent to most posters here.

Quote:
Aggregate IPC is nonsense? Thats exactly how you measure IPC. Its usually measured by the cpu, not the core. Though it can refer to a core if its specified.


Try a bit harder nex time:

Quote:
Four Instructions per Clock
Core 2 is a "four wide" architecture, meaning that it can issue and retire four instructions per clock cycle. The Pentium and Athlon 64 architectures are "three-wide." That's not to say that Core 2 is immediately 33% faster, because pipeline stalls still occasionally happen. Still, the ability to issue four instructions per clock gives Core 2 an immediate leg up.


Clearly that "four instructions per clock cycle" is on a per-core basis, since it would have been eight or sixteen if a dual or quad core was being discussed.

And from your favorite source:

Quote:
A superscalar CPU architecture implements a form of parallelism called instruction-level parallelism within a single processor. It therefore allows faster CPU throughput than would otherwise be possible at a given clock rate. A superscalar processor executes more than one instruction during a clock cycle by simultaneously dispatching multiple instructions to redundant functional units on the processor. Each functional unit is not a separate CPU core but an execution resource within a single CPU such as an arithmetic logic unit, a bit shifter, or a multiplier.


Quote:
No the 4004 wasn't 4 separate chips. It was one chip. Thats its place in history. IF it was 4 chips, it wouldn't be a microprocessor. That its claim to fame. IT WAS ITS ONLY CLAIM TO FAME. We wouldn't even be talking about it otherwise.


OK, on this one you are correct (finally! :D ). After I checked it does turn out that the 4 chips I mentioned were support chips for implementing a complete system, not the CPU itself. However, no matter how stubbornly you persist in attempting to minimize or marginalize the fact (obviously since you feel discrediting Intel raises AMD up a notch or two), the fact remains that Intel invented the first complete microprocessor, and is generally credited by the history books for doing so. Whether that invention was commercially successful is beside the point. Obviously you want to keep flogging that horse because it's the only one you have still breathing, I guess.

Quote:
http://en.wikipedia.org/wiki/Instructions_per_cycle

Well there you go. Notice how they are describing aggregate IPC.


No, I don't notice. No mention of "core(s)" anywhere in the article - feel free to point out where in case I missed it. What I see is that the author failed to make any distinction between cores and CPUs. However the ones I linked clearly make a distinction between what's internal to a core vs. what's external to the core. And I have yet to see any AT or other review site publishing an article where they mention IPC as a "aggregate" IPC - again, feel free to link one in case I missed it.
a b à CPUs
May 6, 2010 7:33:02 PM

fazers_on_stun said:
Hmm, well AMD has been known for dropping their prices, losing vast sums of money, but still staying afloat. However none of what you posted disputes what I said - AMD took economies with Thuban. All your arguments to the contrary notwithstanding, as is readily apparent to most posters here.



Yes they could manufacture a 450mm die or a 600mm or whatever, but its every bit as impractical as your local toyota dealer selling a Camry for $5000. There is a reason that that Thuban is the biggest Die in memory. And its not because your right.

Intel waited for the die shrink for a reason.

Quote:

Try a bit harder nex time:

Four Instructions per Clock
Core 2 is a "four wide" architecture, meaning that it can issue and retire four instructions per clock cycle. The Pentium and Athlon 64 architectures are "three-wide." That's not to say that Core 2 is immediately 33% faster, because pipeline stalls still occasionally happen. Still, the ability to issue four instructions per clock gives Core 2 an immediate leg up.

Clearly that "four instructions per clock cycle" is on a per-core basis, since it would have been eight or sixteen if a dual or quad core was being discussed.

And from your favorite source:

"A superscalar CPU architecture implements a form of parallelism called instruction-level parallelism within a single processor. It therefore allows faster CPU throughput than would otherwise be possible at a given clock rate. A superscalar processor executes more than one instruction during a clock cycle by simultaneously dispatching multiple instructions to redundant functional units on the processor. Each functional unit is not a separate CPU core but an execution resource within a single CPU such as an arithmetic logic unit, a bit shifter, or a multiplier."


Your main problem here is CPU IPC vs core IPC. Its valid to measure both ways, but they are not interchangeable. Somehow you got the idea that only core level is valid.


Quote:
"No the 4004 wasn't 4 separate chips. It was one chip. Thats its place in history. IF it was 4 chips, it wouldn't be a microprocessor. That its claim to fame. IT WAS ITS ONLY CLAIM TO FAME. We wouldn't even be talking about it otherwise."

OK, on this one you are correct (finally! :D ). After I checked it does turn out that the 4 chips I mentioned were support chips for implementing a complete system, not the CPU itself.


............HE ADMITTED I WAS RIGHT!!!!!!.............( Falcon runs off and parties for a week in celebration! )............ :bounce: 


Quote:
However, no matter how stubbornly you persist in attempting to minimize or marginalize the fact (obviously since you feel discrediting Intel raises AMD up a notch or two), the fact remains that Intel invented the first complete microprocessor, and is generally credited by the history books for doing so. Whether that invention was commercially successful is beside the point. Obviously you want to keep flogging that horse because it's the only one you have still breathing, I guess.


Never tried to minimize what they have done. Just point out that the first microprocessor and their current cpu dominance, though they appear closely related, were largely separate events.



Quote:
http://en.wikipedia.org/wiki/Instructions_per_cycle

Well there you go. Notice how they are describing aggregate IPC.

No, I don't notice. No mention of "core(s)" anywhere in the article - feel free to point out where in case I missed it. What I see is that the author failed to make any distinction between cores and CPUs. However the ones I linked clearly make a distinction between what's internal to a core vs. what's external to the core. And I have yet to see any AT or other review site publishing an article where they mention IPC as a "aggregate" IPC - again, feel free to link one in case I missed it.


It says:

"The number of instructions per second for a processor can be derived by multiplying the instructions per cycle and the clock speed (measured in cycles per second or Hertz [Hz]) of the processor in question."

This is clearly a processor wide, aggregate calculation.

But like I said earlier, they are both valid. Aggregate is more common when comparing the complete cpu.
a b à CPUs
May 6, 2010 7:55:04 PM

FALC0N said:
Really? Lets look at what Hardware secrets had to say:

"The new Intel Core microarchitecture, which.... will be used on all new CPUs from Intel, like Merom, Conroe and Woodcrest. It is based on Pentium M’s microarchitecture..............Pentium M is based on Intel’s 6th generation architecture, a.k.a. P6, the same used by Pentium Pro, Pentium II, Pentium III and early Celeron CPUs and not on Pentium 4’s.......You may think of Pentium M as an enhanced Pentium III. Thus you may think of Core microarchitecture as an enhanced Pentium M."

http://www.hardwaresecrets.com/article/313
-------------------------------------------------------------

"Since all new CPUs from Intel will use Pentium M’s architecture, studying this architecture is very important to understand the......... forthcoming Core microarchitecture"


http://www.hardwaresecrets.com/article/270/1
-------------------------------------------------------------

How about Anandtech:

"Core 2 is clearly a descendant of the Pentium Pro, or the P6........ Those CPUs inherited the bus of the Pentium 4, but are still clearly children of the hugely successful P6 architecture"

http://www.anandtech.com/show/1998/1
-----------------------------------------------------

Thats the difference between us Fazers. I know what I am talking about. You just make up whatever choose to believe.

The funny part is, this particular story is common knowledge. Everyone who really follows cpu development knows the story of how Intel's Israeli design team and the Pentium M bailed intel out of the Netburst nightmare. Everyone except you of course.


LOL - where do you get that I'm unfamiliar with the history behind Core2??

In case you don't recall, the discussion was whether Core2 was a mere incremental improvement over Pentium M. The links you provided just show that yes, Core2 is based on what came before it, just as can be said about any x86 processor. Even the Extremetech article I linked to pointed out how C2 uses micro-op fusion the same as Core Duo.

Carried to an extreme, we could say that all digital processors are mere incremental improvements over the original Turing concept.

I noticed you failed to address my point about C2 being 4-issue compared to Pentium M being 3-issue. The reason that is signficant - and not merely incremental - is that the hardware necessary to check for code dependencies within a thread on an OoO CPU increases exponentially, which not only means more transistor budget but also slows down the process. That's why you don't see any 5 or 6-issue cores on desktop at the moment, at least for CISC processors like x86. Much easier and - you guessed it - economical to just bolt on more cores and let the software devs worry about inter-thread dependencies.

Maybe you've heard of the CPU industry business principle, that the company with the performance crown generally gets to set pricing for the competition?? If AMD thought they stood any chance at capturing the crown from the 980X, they would not have skimped on any possible increases, such as keeping the L3/core ratio the same, because then they would be the ones selling the $1K DT chips, not Intel. But no, AMD realized that being stuck on 45nm with no HKMG, and with no significant IPC improvements over P2, they had to sell the 6-core CPU at bargain prices, so they economized where possible.

Why this rather obvious chain of logic seems to escape you, is inexplicable unless you're some of of AMD fanbois with an agenda to discredit Intel wherever you think possible. And seriously, I have nothing to prove to you - you are the newbie here, wading in with an attitude and not even knowing the difference between monolithic and integrated. There is a primetime example of you talking about something of which you are clearly ignorant. So stop with the pompousness already.
May 6, 2010 8:09:56 PM

It's a walk-off!
a b à CPUs
May 6, 2010 8:21:05 PM

FALC0N said:
Yes they could manufacture a 450mm die or a 600mm or whatever, but its every bit as impractical as your local toyota dealer selling a Camry for $5000. There is a reason that that Thuban is the biggest Die in memory.


Really?

Quote:
Manufacturing process: 90 nm.
High-speed transistors only where necessary, to lower dissipation.

Number of transistors: 1720 million:

core logic: 2x 28.5 = 57 million,
L1 and L2 caches: 106.5 million,
24 Mbyte L3 cache: 1550 million,
bus logic and I/O: 6.7 million.

Die size: 27.72 x 21.5 = 596 mm2.


Quote:
Your main problem here is CPU IPC vs core IPC. Its valid to measure both ways, but they are not interchangeable. Somehow you got the idea that only core level is valid.


No I didn't. What I said was that when referring to "IPC", all the review web sites that I've seen mean on a per-core basis. Ergo, the most common meaning, esp. since we already know the number of cores in the CPU. Since there are 3 fundamental ways of comparing CPUs - clock speed, # of cores, and the IPC or architectural efficiency, it's just silly to talk about "aggregate IPC".

Quote:
............HE ADMITTED I WAS RIGHT!!!!!!.............( Falcon runs off and parties for a week in celebration! )............ :bounce: 


LOL - hopefully you are being sarcastic here..

Quote:
Never tried to minimize what they have done. Just point out that the first microprocessor and their current cpu dominance, though they appear closely related, were largely separate events.


Well I could go dig up the other thread where you did marginalize Intel's achievement, but I'll be content with just pointing out that I never tied Intel's marketshare or financial success to the 4004. It was just giving them credit for being first.

Quote:
It says:

"The number of instructions per second for a processor can be derived by multiplying the instructions per cycle and the clock speed (measured in cycles per second or Hertz [Hz]) of the processor in question."

This is clearly a processor wide, aggregate calculation.

But like I said earlier, they are both valid. Aggregate is more common when comparing the complete cpu.


And my interpretation of that wiki article is that the author was lazy and using "processor" where he or she meant core. Since the author never did mention "core" anywhere, then s/he fails to make a distinction. In contrast, the wiki article I linked was clear and definite about the distinction. However it's a free country so choose whatever definition you want :p  - if you like typing "aggregrate" in front of every reference to "IPC" so as to be clear, be my guest..
a b à CPUs
May 6, 2010 8:23:44 PM

roofus said:
It's a walk-off!


LOL - you're right - this equine has been flogged far past the point where rigor mortis sets in :p ..
!