![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
AMD demo's native quad-core
- Thread starter visaris
- Start date
Yea, but in other news:
http://www.xbitlabs.com/news/cpu/display/20061024001546.html
Somehow these articles try to make it sound like AMD is the first, while Intel already had its stuff working last month. Intel even provided benchmarks... where are AMD's benchmarks? If they demonstrate it, I want to know what they can do aswell. Intel's stuff performed excellently, like you'd expect from the finished product.
Which is also why I would find it strange if AMD would be first to market... that could only mean that AMD is rushing things, which can never be good for product quality.
Scali, I dont know if you noticed, but you might want to look back at the thread title again. This thread is clearly about AMD demoing a system...
This aint about Intel, or how you think Intel is better. Take it to the Intel forum, becouse I think I've had about as much Intel worship as I can handle for one day.
Please man. Lets keep this thread on topic. Please.
This aint about Intel, or how you think Intel is better. Take it to the Intel forum, becouse I think I've had about as much Intel worship as I can handle for one day.
Please man. Lets keep this thread on topic. Please.
FlintMurdock
n00b
- Joined
- Sep 20, 2006
- Messages
- 31
Plus this is native quad core. I don't know if it will make any difference compared to kentsfield, but if if Pentium D compared to K8 dual core is any indication it at least gives some hope that AMD is cookin up something special. It would be nice if some benches "leaked" out like the Conroes did. 
mzs_biteme
[H]ard|Gawd
- Joined
- Nov 7, 2001
- Messages
- 1,595
FlintMurdock said:Plus this is native quad core. I don't know if it will make any difference compared to kentsfield, but if if Pentium D compared to K8 dual core is any indication it at least gives some hope that AMD is cookin up something special. It would be nice if some benches "leaked" out like the Conroes did.
We'll probalby see "leaked" Yorkfield benches long before Barcelona if Intel has it their way...
FlintMurdock
n00b
- Joined
- Sep 20, 2006
- Messages
- 31
Yeah, for some reason AMD is really good about keeping secrets. That is a good thing for them, but bad for us enthusiasts that soak up every morsel of info.
freeloader1969
2[H]4U
- Joined
- Dec 13, 2003
- Messages
- 2,238
Maybe they'll actually show something today, besides task manager!
FlintMurdock
n00b
- Joined
- Sep 20, 2006
- Messages
- 31
I'm very optimistic about AMD's engineering brain trust. They've excelled in every way. Think of how much more competitive AMD would be if they had Intel's manufacturing capabilities. I've noticed that when AMD is super quiet, something good is happening there.
LstBrunnenG
Supreme [H]ardness
- Joined
- Jun 3, 2003
- Messages
- 6,676
FlintMurdock
n00b
- Joined
- Sep 20, 2006
- Messages
- 31
LstOfTheBrunnenG said:
I can't see the picture from your link, but if it's the same pic of Henri Richard holding up a Clovertown in one hand and a Barcelona in the other, they seem roughly the same size when you piece together the individual Conroe based die.
Edit: Strange enough, now I see the pic, lol.
chrisf6969
[H]F Junkie
- Joined
- Oct 27, 2003
- Messages
- 9,011
FlintMurdock said:
(about both Intel & AMD)
They're only good at keeping secrets when things are bad. When things are good they BOTH mysteriously leak them and don't do very much about hushing the leakers!!
Bloodgod42
Gawd
- Joined
- Nov 20, 2005
- Messages
- 592
Scali2 said:
AMD has been working on, and talking about, Barcelona for long before Intel's Quad -core. AMD is only taking thier time in releasing thier Quad-core because they wanne do it right (natively). Please do a little research if you want to post on the topic. Thanks.
FlintMurdock
n00b
- Joined
- Sep 20, 2006
- Messages
- 31
chrisf6969 said:
Yeah, Chris, looking back, I do have to agree with you. Same goes to ATI(oops AMD) and Nvidia. It;s just been awhile since AMD had to release something new, and so short that Intel has released new arch that I forgot.
Tawnos
2[H]4U
- Joined
- Sep 9, 2001
- Messages
- 3,808
Does it mean I've been on fark too much recently if I see the title of this thread and my immediate urge is to post bob the angry flower's guide to using the apostrophe (you idiots)?
On topic...LstOfTheBrunnenG, the link you put is throwing an error 500
On topic...LstOfTheBrunnenG, the link you put is throwing an error 500
w1retap
[H]F Junkie
- Joined
- Jul 17, 2006
- Messages
- 13,718
Well, good. Finally a little news that confirms that native quad cores are coming in 2007 for a Socket AM2. I'll be droppin one of those in. =)
FlintMurdock
n00b
- Joined
- Sep 20, 2006
- Messages
- 31
jon67 said:Yeah like when they kept quiet for 12 months before C2D was released... good sign
Um..not sure what you're saying. AMD had the performance crown before C2D was released. Why would they need to make noise? The only noise being made was Ka-Ching$$!.
Bloodgod42 said:
Well excuse me, but if AMD has been working on Barcelona long before Intel, then why can't they actually show the working thing at this demo?
As for 'natively', that doesn't mean much, as Kentsfield has shown. AMD just HAS to make it natively because they have the problem of the on-die memory controller. It's not possible for them to just glue two dies together, and make it work on the existing sockets.
Other than that they know (as we all do now, with 4x4) that just glueing a bunch of K8 together won't give them enough performance, so they have to rehash the core in order to try and stay in the race.
So I don't buy this whole 'native' talk. It's just marketing buzz. All I want to see is a working processor and benchmarks, and that's something they cannot provide... So is Barcelona any good with its 'native' design? Your guess is as good as mine, even after they 'demoed' the product.
With Intel we at least have the guarantee that their single-die quadcores will be at least as fast as their current MCM quadcores. We also don't have to wait that long, because we can already buy the MCM quadcores, and they already perform better than anything AMD can offer. So AMD just claiming 'native' on some product that apparently doesn't even work yet, is not enough to impress me.
They did. They had a working 16-core Opteron box, based on the Barcelona core on display. This move by AMD shows that the design of the quad-core parts is done, they have samples that not only work but work in a large SMP environment, and they have working 65nm parts with the new core design. How are they not "show[ing] the working thing at this demo"? I'm very confused by your remark here.Scali2 said:
That's the same argument that was used by many when the P4D came out. And guess what? It did matter. Just as it matters now for Kentsfield. Intel even knows it matters. If it didn't matter, then why would Intel be designing their own native quad-core part? Simple fact: it does matter. AMD knows it, Intel knows it, and I know you know it matters. That's why you'll be selling the hype when Intel releases their native quad-core parts in late 2007. See, as far as you are concerned, it doesn't matter when AMD has it, only when Intel does.Scali2 said:
That is not really true. AMD could glue the second core(s) onto the first with HT. AMD would only be able to use one mem controller, but this design is used on some MBs today. Still, I fail to see how that is different than Intel's situation: Intel's two chips share the same FSB and the same mem controller. AMD chose not to do an MCM hack, not because they couldn't do it, but because they don't want to. I for one applaud their design decision.Scali2 said:
Hey, you got one right!! Though, I think you miss the importance of 4x4. AMD's quad-FX platform is not a stopgap until Barcelona, Quad FX is a platform for use with Barcelona. Quad FX is a bit slower than Intel's quad-core now, but when Barcelona catches up with (and possibly passes) Core2, the Quad FX platform will boast 8 cores in a single desktop system. Do you intend to tell me that Intel's single quad-core chip will be better performing than 8 of AMD's newest cores?Scali2 said:
Bzzz, wrong. Intel is making their own native quad-core chip. Will you call their version "marketing buzz" as well? Somehow I doubt it.Scali2 said:
AMD said 40-70 percent better than old Opterons. Just as Intel first released their own Core2 benches. You ate Intel's numbers up off the floor a number of months ago. Want me to find a link? So what is wrong with AMD's?Scali2 said:
Are you telling me that you think that AMD's quad-core chips, based on a new revamped core, will perform worse than their current dual-core or FX chips? What are you implying here? Are you nuts?Scali2 said:
Fair enough.Scali2 said:
It does work, they publicly showed the unit, and gave some initial performance numbers verses their old product line. You are so full of FUD it is sad. Go post in the Intel forum and quit trashing the AMD threads with outright lies:Scali2 said:
visaris said:
Haha...Baghdad Bob at it's best!
AMD showed the Windows task manager ; the best you can say is that it load Windows
Secondly , by releasing estimates instead of hard number something is fishy there.My bet is they are so buggy and unstable that benchmarking was impossible.
Do you want to be linked to a Tigerton running POV-ray ?
yurimxpxman
n00b
- Joined
- Dec 1, 2006
- Messages
- 3
savantu said:
Might I remind you that this project is still under development? It's bound to have some kind of problems before they're finished. So even if you're right, it doesn't really matter at this point. Do you really think AMD would waste their time by releasing a buggy processor? puh-leez.
Lol. You are using the same arguments the AMD fans made about the initial Conroe benchmarks. Except, at that point in time, you bashed the AMD fans for it. It's funny how it's ok as long as you are doing it. Seriously though, I agree to some extent. I would like to see more information from AMD. However, AMD's secrecy does bring a bit of excitement with it.savantu said:AMD showed the Windows task manager ; the best you can say is that it load Windows
Secondly , by releasing estimates instead of hard number something is fishy there. My bet is they are so buggy and unstable that benchmarking was impossible.
No, this is the AMD forum, in a thread I started about AMD's quad-core CPUs. I want to talk about AMD and their products here. Pleae take your Tigerton and other Intel "information" over to the Intel forum. If you cannot stay on topic, kindly leave.savantu said:Do you want to be linked to a Tigerton running POV-ray ?
visaris said:
I've not seen any benchmarks of it... not even seen pictures of the working machine. All I saw was some guy holding up a single-die CPU, which could have been anything... and a screenshot of a Windows Taskmanager with 16 graphs... which again could have been anything. It could even have been an Intel system. There's no way to identify the system.
It's also suspicious that they come out with this news pretty right after 4x4 has tanked, don't you think?
It's just a way to try and give hope to their investors.. "Trust us, good things will come!".
Did it? How can you tell? The first Pentium D was actually single-die, so arguably it was 'native'. But if you claim it is not native, then how can you compare? It's the only Pentium D there is... there's no 'native' Pentium D to compare.
Other than that, Pentium D still performed quite well, compared to a dual CPU Xeon netburst, which it essentially was. It may not have compared favourably to the Athlon X2 at the time, but that was not anything to do with being 'native' or not.
Other than that, I don't really consider K8L 'native', if I compare it with Core2. Core2 has a shared L2-cache between cores... K8L basically takes 4 cores with non-shared L1 and L2-cache, just like the current K8, and then patches it with some shared L3-cache.
Hardly what I'd call a good design.
Intel will at least get a shared L2-cache across all cores, which will dramatically increase performance, just like Core2 Duo performs dramatically better than K8 at this time.
You're missing the point here.
Intel makes a single-die (let's not confuse matters by using 'native' when you mean single-die) quadcore because this will allow them to create a MCM octacore on 45 nm.
Other than that Intel won't try to market the single-die quadcore as 'native', trying to claim it will be better than MCM by default.
Which brings us to the actual point here... 'Native' only matters when it is implemented properly... You can take advantage of having all cores on one die, if you have a proper design. The Core2 Duo is a nice example of that, with its shared L2-cache.
Shared L3-cache is questionable... and considered poor when the competition will offer shared L2-cache.
If the design isn't different, then one or two dies, or even one or two sockets doesn't really matter much... Pentium D and Athlon64 X2 both prove that point.
No, it matters because in Intels case it will enable octacore computing on a single socket.
Leaving octacore aside, I would not care about nonsense like 'native' or not, I'd just look at benchmarking results to see which design is giving me the best performance.
Really? I just think they're being really stupid, because now everyone is going to buy Intel quadcores for the next 6 months, while AMD is only competing in the mid-end at best.
I hope they fired everyone who supported this 'no-MCM'-decision at AMD, if you are right (but I have more faith in AMD than you do, I trust they would have gone MCM if it was techically feasible).
I don't see why you would compare AMD's dual CPU system to Intels single-CPU systems... especially since Intel already sells dual CPU systems with the Xeon MP, based on Core2 Duo... which perform even better than Kentsfield, and leave 4x4 in the dust even more.
Besides, as mentioned, Intel will probably offer an octacore MCM chip at around the time that Barcelona is released, so that probably means AMD still needs to use two CPUs where Intel only needs one.
Marketing buzz is all about how you market your technology with buzz (gee, what a surprise).
Intel has never used the word 'native' to market any of their chips... They never said "Get a Core2 Duo instead of a Pentium D, because it's a 'native' dualcore!".
I don't expect them to start with quadcore. I don't think they'll have to, because there will be plenty of other good reasons to get their single-die quadcore than just the fact that it's 'native'.
Excuse me, but you must have me mistaken for some Intel !!!!!!... I didn't "eat Intel's numbers up off the floor" at all. I waited until they sent their systems to various independent review sites, and when they all experimented with the systems, did their own benchmarks, and all came to pretty much the same conclusions, I figured there must be something true about these figures.
So I'm now waiting for AMD to send its chips to these independent review sites, and see what the reviewers write.
How do you know their quadcore chips will be better? A quad FX system uses two memory controllers, a quadcore will use only one. So technically it has to make do with only half the bandwidth... Since the quadcore's caches are so small (rumour is 512 kb L2 per core and 2 mb shared L3 cache, which is basically just a 'shadow memory' of what is in the L2 caches, so effectively no extra cache size, just a buffer to synchronize L2-caches more efficiently).
Also, with 4 cores on one die, rather than in physically different packages, there may be a problem with power consumption/heat dissipation, meaning that the clockspeeds have to be lower (just like they were lower with the introduction of both Intels and AMD's dualcore processors, and again with Intels quadcore).
We also don't know how much impact the core improvements will have on per-core performance.
I'm simply implying that it's hard to say anything about performance right now.
With Intel it's far more straightforward. MCM already works with a single memory controller, and there are already 4 cores in one package. It can only get better when they go to a single die.
I'm glad we can have a mature and friendly technical discussion.
Sorry, but I must have missed these benchmarks and everything.
And this topic you've started on it, doesn't link to them either...
I just assumed that they didn't exist, like I usually do when I can't find the actual info on the net.
Provide the info if you want me to change my mind. I'd love to see how well K8L performs. I know Tigerton is impressive enough.
visaris said:
You know, it's quite sad that we even need an AMD and Intel forum... And it's people like you that are the reason why.
If we could all be mature and stick to technical arguments, then a single CPU-forum would have been enough.
I don't see why you don't want anyone to mention Intel in this thread? Isn't it about AMD's quadcore, which should compete against Intels quadcore? So isn't Intels quadcore the milestone for AMD?
For most sane people, the choice whether they will buy the AMD quadcore or the Intel quadcore will simply depend on which one performs best. So I'd say that the comparison is all that matters, and we should be talking about the Intel quadcore a lot here... at least half the time... where the other half is how AMD compares to it.
There's no point in just talking about the AMD quadcore... Your opening post basically covered that part already.
w1retap
[H]F Junkie
- Joined
- Jul 17, 2006
- Messages
- 13,718
Some of you guys need to write a book on this. Everyone is an expert in unreleased products all of a sudden.
yurimxpxman said:
Technically, yes...
Just check their technical documents for errata. Strictly speaking, pretty much everything they released is buggy.
Disclaimer: this is not unique to AMD in any way, all manufacturers have bugs.
Scali's law says that the amount of bugs is proportional to the pressure to get the product out to market. Best example of this is the Pentium III 1133 MHz.
I say:
Are you telling me that you think that AMD's quad-core chips, based on a new revamped core, will perform worse than their current dual-core or FX chips? What are you implying here?
The response:
How do you know their [AMD's] quadcore chips will be better?
I think this is a pretty good illustration of Scali's bias.
Are you telling me that you think that AMD's quad-core chips, based on a new revamped core, will perform worse than their current dual-core or FX chips? What are you implying here?
The response:
How do you know their [AMD's] quadcore chips will be better?
I think this is a pretty good illustration of Scali's bias.
I think the fact that you conveniently left out every single one of the many technical details I've used to support my view, and reply with such a personal rather than technical post, is a clear illustration of your bias.
If you can't win on technical arguments, you'll just attack personally, right?
I think there's nothing weird about me questioning AMD's quadcore performance, given the amount of factors in play, which I've mentioned above. It has nothing to do with bias, just with trying to be realistic and analytical.
I question Intel just as much.
If you can't win on technical arguments, you'll just attack personally, right?
I think there's nothing weird about me questioning AMD's quadcore performance, given the amount of factors in play, which I've mentioned above. It has nothing to do with bias, just with trying to be realistic and analytical.
I question Intel just as much.
pcfan10110
Limp Gawd
- Joined
- Nov 8, 2006
- Messages
- 233
Both of you SHUT UP!
pcfan10110
Limp Gawd
- Joined
- Nov 8, 2006
- Messages
- 233
Scali2 said:
Intel is now better performance wise than AMD but, like all good things it must come to an end. I am a !!!!!! for AMD that is true but, performance benchmarks do not lie. I am unbiased either way. Who really knows what AMD has planned it might put AMD back in 1st place or AMD will still be 2nd place. I currently have many AMD PC's in my house and only one Intel PC. Now, STOP YOUR BICKERING! JEEZ! YOU TWO ARE WORSE THAN A MARRIED COUPLE!
Excuse me, but it pisses me off when someone calls me biased, especially towards Intel.
In fact I was actually anti-Intel, and anti-x86 altogether (but back in the day, those were equivalent)... the only CPU brand I've ever been fan of is Motorola.
I could care less about either Intel or AMD, I still think x86 is a monstrosity, no matter who makes it... and I've bought either brand over the years, depending on which monstrosity gave best value-for-money at the time.
If you're gonna call me biased, at least call me Motorola-biased.
In fact I was actually anti-Intel, and anti-x86 altogether (but back in the day, those were equivalent)... the only CPU brand I've ever been fan of is Motorola.
I could care less about either Intel or AMD, I still think x86 is a monstrosity, no matter who makes it... and I've bought either brand over the years, depending on which monstrosity gave best value-for-money at the time.
If you're gonna call me biased, at least call me Motorola-biased.
pcfan10110 said:
Why? This makes absolutely no sense whatsoever.
pcfan10110 said:
Don't know why you respond to my post, because I pretty much said exactly this... We don't know what AMD's next CPU will bring... which is me being completely unbiased.
It can go either way.
Scali2 said:
It is not so much an issue of "AMD side" or "Intel side", it is an issue of respect for a user's thread. This is my thread, about AMD gear, and you are not able to stay on topic. If you want to start your own thread with a title of "Intel's whatever vs. AMD's whatever" and place it in the AMD forum, I couldn't care less. That, however is not what you are doing. You are comming into my thread, taking it off topic, and trashing it. Please make your own thread and leave mine alone, unless you are able to stay on topic and talk about AMD's quad-core chips, and AMD's quad-core chips alone.
--
Also, you mention that you think the shared L3 cache is a bad design. Would you care to explain why you do not like it?
I think the L3 cache design is a very elegent solution, right in the middle of a continuum:
- On one extreme of the continuum, there is a chip with totally independent caches. This design is great for workloads in which each core is working on something different. The cores do not need to fight each other over cache use, and this minimizes cache thrashing. This design is very well suited to applications designed with a pipelined threading model, such as multi-threaded Lame, as well as a handfull of other threading schemes. However, this extreme does not do well when multiple threads all work on the same set of data, which is also common in many threading schemes.
- On the opposite end, one has all cache shared. This design is ideal for sharing instructions and data that need to be accessed by multiple threads at a time. A lot of software I write uses this model actually. The down side is that this design is very susceptible to cache thrashing if all cores are working on different sets of data, and waste a lot of time fighting over the cache, throwing the other core's cache lines out.
- Now that we've looked at the two extremes, let's look at AMD's solution, somewhere in the middle. AMD uses an independent L2 cache, which minimizes cache thrashing on workloads where each thread uses different data (or instructions). AMD also uses a shared L3 cache which will keep data used by all threads in a cahce accessable to each of them. Further, AMD's L3 cache is said to be both exclusive and Inclusive, depending on access patterns. This should allow the L3 to cache shared instructions in an inclusive manner (ideal for shared instructions) and data in an exclusive manner when needed.
So, Scali, I think I've made my argument for why something in the middle of the cache continuum is a good idea, and why I think AMD has made the correct design decision. I did it all without mentioning anything from Intel. Why is it that you think AMD's cache is a poor design, and why do you think it is a hack? Can you even answer without comparing to Intel?
--
AMD's new core has a revamped and more advanced cache, an enhanced core with better FP, SSE, and more cache bandwidth. Bigger TLBs, more fast-path instructions, not to mention double the number of cores. The new core is better than the old one in many, many ways. The fact that you can keep a straight face while you claim that the new core will perform worse than the old ones amazes me. This is the fact that demonstrates your bias. Even when looking massive improvment in the face, you still say it is a step backwords. I invite all readers of your posts to make their own conclusion, however, I think the conclusion is clear.
pcfan10110
Limp Gawd
- Joined
- Nov 8, 2006
- Messages
- 233
visaris said:
Now that is a mature and educated response.
InorganicMatter
[H]F Junkie
- Joined
- Oct 19, 2004
- Messages
- 15,461
The problem with L3 cache is that it has a very, very high latency. Back in the early days of the Extreme Edition, Intel released an EE part that had 2MB of L3 cache, and I seem to recall that the L3 cache had such a high latency that it actually decreased performance compared to units that had no L3 cache.
Is K8L prepared to deal with L3's flaws? Maybe, but if it was I would think that AMD would have been pronouncing their engineering victory for months by now.
Is K8L prepared to deal with L3's flaws? Maybe, but if it was I would think that AMD would have been pronouncing their engineering victory for months by now.
brucedeluxe169
2[H]4U
- Joined
- Aug 24, 2004
- Messages
- 3,720
Scali2 said:
haha, talk about Orwellian revisionism.....
brucedeluxe169
2[H]4U
- Joined
- Aug 24, 2004
- Messages
- 3,720
InorganicMatter said:
as I recall, the Gallatin core EEs were the fastest Netburst processors until some of the later Prescott 2Ms were able to clock higher to beat them.
The L3 cache EEs were better than the ones w/o it...
visaris said:
Oh please. I think it's more of a question that you don't want to hear what I posted. It wasn't off-topic. I was putting AMD's quad-core into perspective, namely that of its competition.
AMD's quadcore processor won't exist in a void, it will be released in the real world, where other quadcores are available.
visaris said:
I already did, please read more carefully.
visaris said:
This shouldn't be an issue if the cache size is equal to the total of independent per-core caches, and the set-associativity of the caches is good.
Then you are in effect dynamically allocating the cachelines on demand of the cores. Tests with my own software concluded that Core2 Duo does an excellent job of this, so I expect that Intel will also be able to come up with decent performance for a shared cache between 4 cores.
visaris said:
I don't agree at all. All operations have to go through L2-cache, so you get more latency, which is not exactly middle-ground, but basically worst case everytime. In no situation will it perform as well as shared L2-cache will. On top of that, even the worst case of the shared L2-cache will be faster than L3-cache.
visaris said:
I already gave the answer, without having to compare to Intel. Shared L2-cache is one level higher in the hierachy, meaning less latency, more speed, more performance.
Other than that, I already said the cache is very small, so it can only be used to 'mirror' the L2-caches, it cannot provide additional caching.
But I prefer to also compare to Intel, because it puts things into perspective. Namely that Intel in fact DOES have shared L2-cache, and that Intel did have L3-cache solutions where L3's capacity was larger than that of L2, on the last Xeon netburst CPUs, which meant it would provide additional caching rather than just mirroring the core caches. This alleviated the extra latency costs of L3-cache, and gave some extra performance in return..
visaris said:
I can't draw any conclusions until I can see some benchmarks that show how much better this new core is. Anyone who can tell me how this new core performs, based on just this info, must have a crystal ball.
LstBrunnenG
Supreme [H]ardness
- Joined
- Jun 3, 2003
- Messages
- 6,676
This is how I picture you wanted the discussion to progress:visaris said:
Visaris: AMD demoed native quad core!
[H]: Okay.
And it looks like instead you got someone who, biased or not, tried to engage you in level-headed discussion about this product in relation to other products on the market. God help us all.