AMD... Tri-Core?!?

morfinx said:
To be honest I haven't heard anything about any IDF announcements. But I'm on the design side so news sometimes take awhile to propagate to me.
Click to expand...

Are you sure you haven't heard anything about a demo that begins with "Ne" and ends with "halem"? Not that I've heard anything.... *whistles* ;)
 
APOLLO said:
Good question, I believe the memory controller was divided in two to access both memory channels independently of each other, thus data can flow to and from both memory channels simultaneously. If that's the case, then the individual memory controllers should have access to all cores, but this is pure speculation on my part, so don't quote me on this.
Click to expand...

If your right and it uses the MCs independently so it can flow from both memory channels simultaneously, that would mean (and this is only a guess) that atleast one of the MC can access all four cores, because if you where to run memory in single channel you would most likely have conflicts if both MCs where trying to access the RAM. Correct me if I am way off.
 
JackPack said:
Are you sure you haven't heard anything about a demo that begins with "Ne" and ends with "halem"? Not that I've heard anything.... *whistles* ;)
Click to expand...
Of course there will be Intel announcements at the IDF. I was asking morfinx if there would be any AMD announcements at that time because there is a rumor they'll be announcing something and the tri-core processor could be it. However, he's on the engineering team, not AMD's marketing division.

If you know anything about Nehalem, I suggest you post it in the Intel forum. Current rumors claim that Intel will have a Skulltrail demo at the IDF, but that can't be Nehalem; it's probably a dual Penryn utilizing non-registered RAM, which should be an interesting 'platform' in its own right.
 
debello_64 said:
If your right and it uses the MCs independently so it can flow from both memory channels simultaneously, that would mean (and this is only a guess) that atleast one of the MC can access all four cores, because if you where to run memory in single channel you would most likely have conflicts if both MCs where trying to access the RAM. Correct me if I am way off.
Click to expand...
OK, I did a bit of research and this is what I found:

"Behind this L3 cache sits an improved memory controller, still integrated into the CPU as with previous Opterons. AMD claims this memory controller is better able to take advantage of the higher bandwidth offered by DDR2 memory thanks to a number of enhancements, including buffers that are between 2X and 4X the size of those in previous Opterons and an improved prefetch mechanism. Perhaps most notably, the new controller can access each 64-bit memory channel independently, reading from one while writing to another, instead of just treating dual memory channels as a single 128-bit device."

Source: http://techreport.com/articles.x/13176

Here's more on the memory controller(s): http://www.realworldtech.com/page.cfm?ArticleID=RWT051607033728&p=2

And, the memory subsystem including the cache hierarchy: http://www.realworldtech.com/page.cfm?ArticleID=RWT051607033728&p=7

Now, how does all this relate to tri-cores and whether such a hypothetical microarchitecture will require one or both memory controllers, I have no idea... :D
 
This could be a great marketing move in one respect, it would make dual-cores take the place of single-cores, tri-cores as dual cores and then that leaves quad cores and up (octa?) for the high-end enthusiast
 
shyboi said:
Tri Core 939 please.
Click to expand...
That's wishful thinking as would be a tri-core 940. One can only dream...

Like someone else mentioned, I foresee a viable market for a tri-core processor as mobile performance chip. Shut down one or two cores while on battery power, and have all cores operational plugged in.
 
APOLLO said:
That's wishful thinking as would be a tri-core 940. One can only dream...

Like someone else mentioned, I foresee a viable market for a tri-core processor as mobile performance chip. Shut down one or two cores while on battery power, and have all cores operational plugged in.
Click to expand...


Interesting idea to be sure !
 
I hadn't thought of that: I suppose the core doesn't care what packaging in which it is implemented. Wouldn't power consumption be a bit much for mobile use, though?
 
holy shit, they actually did it. Triathlon anyone? I guess they are going to hold a Phenomenal Triathlon in Barcelona.
 
APOLLO said:
It's probably disabled to for a lower end of the market. Maybe it's also salvage for bad L3??
Click to expand...


Oh most certainly ! Yields (on Barcelona) must be horrifically bad.The press release is just.... comical to say the least. :D I had to wonder if this was some April Fools joke sprung many months too early. :p

But hey,if it'll sell and earn profits,go for it I say. :)
 
Manny, your negative comments are even beginning to annoy me. Why don't you just shut the hell up.
 
What negative comments ? I am merely using 'common sense' something not so common it would seem.As I said earlier,if it'll sell and make a profit why not ?

I merely pointed out that the motives behind this new product line are,suspect at best.

I have broken no forum rules,no name calling,no spamming,none of that crap.I (an AMD product user,like yourself) am merely sharing my views on a just newly announced AMD product,on the AMD sub forum. :)
 
Fuad is reporting that the first tri-cores will be arriving in Q1 '08. Also reported, Intel will be going tri-core as well. Seems like both companies think there is an untapped market out there. Either that, or it's monkey see, monkey do...
 
it's official...but i've read the whole thread and the best guess as to why is marketing taking a turd and making lemonade? i guess it's not a bad idea, just that i want a whole four cores. the point that matters most, price, has yet to be announced. bring it on, AMD!
 
mastercheeze said:
it's official...but i've read the whole thread and the best guess as to why is marketing taking a turd and making lemonade? i guess it's not a bad idea, just that i want a whole four cores. the point that matters most, price, has yet to be announced. bring it on, AMD!
Click to expand...

Look where the tri-core (obviously) is slotted: between dual and quad cores (I know this is obvious, but I think it's too obvious that people fail to realize the significance of it).

There will always be a market for "one better". We're human, we need to keep up with the Joneses.
 
It makes a lot of sense.

First, people like bigger numbers. A Tri-Core just sounds better than a Dual-Core. And even if it were slower than a dual core GHz wise, AMD could sell it off as being faster with more cores, which in certain processes would be very true in fact.

Also, they might make a lot of money off of otherwise failed Quad-Cores. For instance, a quad-core has a core that can only reach 1.8GHz while the rest reach 2.8GHz. Do you sell it as a quad core for $280 at 1.8GHz or do you sell it as a Tri-Core at 2.8GHz for $320? I don't know about you, but a 2.8GHz Triple Core sounds more appetizing than a 1.8GHz and that will certainly sell well.

Plus, if Intel is following in this field, then they certainly know that letting AMD have a free lunch isn't a good idea for themselves.
 
JackPack said:
If the message is that AMD is only selling these because they're defective quads, the average Joe isn't going to touch it. If marketing had researched this and there really was demand, it would have been on roadmaps. Instead, this tri-core is clearly reactionary to yield results.
Click to expand...



If it's cheaper than a 4... and faster than a 2.... i'd buy one, dependent on the benchmarks of course. It's not like this is a new practice in any way, like nvidia hasn't done the same
 
I don't think the average buyer cares that one core is disabled (Brokealona). Like mentioned above, 3 is bigger than two.

If AMD prices these right, they might do OK. But AMD isn't selling these in a vacuum, and has to deal with their own and Intel's low end quad core pricing.
 
Manny Calavera said:
Oh most certainly ! Yields (on Barcelona) must be horrifically bad.The press release is just.... comical to say the least. :D I had to wonder if this was some April Fools joke sprung many months too early. :p
Click to expand...

Heka, a 45nm product, for release in 2009 is completely unrelated to Barcelona in its current state and has *nothing* to do with current yields. Welcome to the idiot filter. It's taken far too long.
 
pxc said:
LOL You think this is a *good* thing, or is it a result of bad yields on 4 core Barcelonas? I guess theinq can spin, spin, spin anything about AMD to make it good.
Click to expand...

Exactly, "Native quad core....." this & that, but it reduces exponentially the number of good / fast cores you can get.

Lets say if they had a 1/10 chance of hitting 2.8Ghz, then you have 4 cores.
So 1/10th to the 4th power means, they only have a 1 in 10,000 chance of hitting 2.8Ghz on all 4 cores on the same die!!

Granted things change with the extra heat of all 4 cores on one die too.

But you get the idea.

Intel's "non-native" quad core may have some "speed" disadvantages (over native), but considering its still faster b/c all of the other tweaks, AND it cost Intel a lot less money to make. They can match (bin) their best dual cores together and make fast quad cores. If one core is bad, they can always disable it, and sell it as a single core Celeron. With less die space wasted., etc.

This just looks like Intel has AMD backed in to the corner with performance and MANUFACTURING (45nm + seperate packages) advantages.
 
The question is this, there are a lot of very smart people working at AMD. Could they screw up so bad that they would HAVE TO manufacture these chips to stay alive?

Here is something interesting, I manage a data center, 7000 hot servers humming along all the time day and night. I see every server that comes in and I have to place them in an area where they won't overheat/die (I know where am I going with this)
I have seen far more AMD servers installed over the past year than intel severs. Isn't the bread and butter of a chip maker in servers? Why would amd care about the desktop? If they are selling more servers? Also AMD vendors come along with all these programs for the developers showing them how to tweak the programs they are developing. I've never ever seen an intel vendor make half the effort the AMD guys do.

So again, where is the bread and butter?

I'll tell you what one of my developers told me, he believes (and heres really fucking smart too) that AMD is going to put a GPU on the die and 256 megs of cache. Take out the video card all together. That is why they purchased ATI. And they are developing their own systemboard too. Think of the advance!


chrisf6969 said:
Exactly, "Native quad core....." this & that, but it reduces exponentially the number of good / fast cores you can get.

Lets say if they had a 1/10 chance of hitting 2.8Ghz, then you have 4 cores.
So 1/10th to the 4th power means, they only have a 1 in 10,000 chance of hitting 2.8Ghz on all 4 cores on the same die!!

Granted things change with the extra heat of all 4 cores on one die too.

But you get the idea.

Intel's "non-native" quad core may have some "speed" disadvantages (over native), but considering its still faster b/c all of the other tweaks, AND it cost Intel a lot less money to make. They can match (bin) their best dual cores together and make fast quad cores. If one core is bad, they can always disable it, and sell it as a single core Celeron. With less die space wasted., etc.

This just looks like Intel has AMD backed in to the corner with performance and MANUFACTURING (45nm + seperate packages) advantages.
Click to expand...
 
I think you are grossly underscoring the processor manufacturing process. They arent just pounding out a bunch of processors hoping a couple of em are going to work. The tool their Die manufacturing for a specific purpose and core speed, and or a range of core speeds, as we've seen. Lets be a little more realistic, and say that theres a 1 to 5% chance that a die is produced containing a core that CANNOT hit its target, or is dead alltogether. Now recalculate.

1 in 10,000 chance? maybe if they are producing processors outside in the parkinglot.
 
Celsius said:
The question is this, there are a lot of very smart people working at AMD. Could they screw up so bad that they would HAVE TO manufacture these chips to stay alive?

Here is something interesting, I manage a data center, 7000 hot servers humming along all the time day and night. I see every server that comes in and I have to place them in an area where they won't overheat/die (I know where am I going with this)
I have seen far more AMD servers installed over the past year than intel severs. Isn't the bread and butter of a chip maker in servers? Why would amd care about the desktop? If they are selling more servers? Also AMD vendors come along with all these programs for the developers showing them how to tweak the programs they are developing. I've never ever seen an intel vendor make half the effort the AMD guys do.

So again, where is the bread and butter?

I'll tell you what one of my developers told me, he believes (and heres really fucking smart too) that AMD is going to put a GPU on the die and 256 megs of cache. Take out the video card all together. That is why they purchased ATI. And they are developing their own systemboard too. Think of the advance!
Click to expand...

I said that YEARS ago. I was wondering why motherboard makers couldn't solder on super fast memory with a wide path on the motherboard (like a videocard's memory), with a super all in one chip.

Why they couldn't integrate more stuff. If they had a system on a chip, with all the increased speed of having more high speed ram, high speed interconnects between everything. You wouldn't even need the super high speed 3Ghz 4 core chips.

I'm sure we're still a few years from that, but imagine a tiny VIA (mini-ITX) size motherboard with something like 12-30Mb of L2 cache, 512-768Mb of super high speed ram - effectively an L3 cache (think videocard speed & bit-width) all on a motherboard. No need to add ddr2 or ddr3 ram, b/c you have so much bandwidth & speed with the onboard memory. Though I guess you could make some ddr3 slots available.

Anyway, imagine a PS3 / Xbox360 sized boxed thats more powerful than anything you can imagine now. The 2 worlds are getting closer & closer together.

Look at what Xbox 360/PS3 does with its specs (not a lot of ram).

Ace123 said:
I think you are grossly underscoring the processor manufacturing process. They arent just pounding out a bunch of processors hoping a couple of em are going to work. The tool their Die manufacturing for a specific purpose and core speed, and or a range of core speeds, as we've seen. Lets be a little more realistic, and say that theres a 1 to 5% chance that a die is produced containing a core that CANNOT hit its target, or is dead alltogether. Now recalculate.

1 in 10,000 chance? maybe if they are producing processors outside in the parkinglot.
Click to expand...

I'm talking about for their top binned speed. Something like 2.8-3.0Ghz. And I used those numbers to show how the more cores the more expontially complicated it gets to have good speeds.

You can use different #'s. I didn't say that was the actual percentages. Though it COULD be that rare for a quad core chip that hits 2.8-3.0Ghz at default voltage.

Lets say for their 2.5Ghz bin, its something like 75% of chips hit that speed. 3/4th (75%) to the 4th power = only roughly 31% will work with all 4 dies at that speed.
 
chrisf6969 said:
I said that YEARS ago. I was wondering why motherboard makers couldn't solder on super fast memory with a wide path on the motherboard (like a videocard's memory), with a super all in one chip.

Why they couldn't integrate more stuff. If they had a system on a chip, with all the increased speed of having more high speed ram, high speed interconnects between everything. You wouldn't even need the super high speed 3Ghz 4 core chips.

I'm sure we're still a few years from that, but imagine a tiny VIA (mini-ITX) size motherboard with something like 12-30Mb of L2 cache, 512-768Mb of super high speed ram - effectively an L3 cache (think videocard speed & bit-width) all on a motherboard. No need to add ddr2 or ddr3 ram, b/c you have so much bandwidth & speed with the onboard memory. Though I guess you could make some ddr3 slots available.

Anyway, imagine a PS3 / Xbox360 sized boxed thats more powerful than anything you can imagine now. The 2 worlds are getting closer & closer together.

Look at what Xbox 360/PS3 does with its specs (not a lot of ram).
Click to expand...

You are still going to need lots of ram for large programs or else you are going to swapping to the HDD too much, even with fast/wide bandwidth you need the programs to remain in RAM.
 
maybe retool the ram so that its only accessed as secondary for programs and our applications that read in large quantities or preload information to be read later.
 
Celsius said:
The question is this, there are a lot of very smart people working at AMD. Could they screw up so bad that they would HAVE TO manufacture these chips to stay alive?
Click to expand...
I don't think it's so much a case of 'have to' as much as 'want to.' Meaning, they don't need to market tri-core processors to stay afloat, but if they can easily do it and make a profit, why not? Who knows, maybe we will see unexpected performance advantages with a tri-core processor. For example, it might OC better than a quad-core, but have the inherent capability to process more threads, have superior performance/watt, etc.

I remember a couple of years ago, THG performed an experiment in which they installed a dual-core Opteron and a standard Opteron processor on the same board. They obviously encountered some problems but the system was stable enough to perform a few benchmarks. Well, it was discovered that three cores proved superior to two cores and four cores in a few of the benchmarks. The conclusion at the time was that software available then did not fully utilize four cores efficiently even if there was full support.

I have seen far more AMD servers installed over the past year than intel severs. Isn't the bread and butter of a chip maker in servers? Why would amd care about the desktop? If they are selling more servers? Also AMD vendors come along with all these programs for the developers showing them how to tweak the programs they are developing. I've never ever seen an intel vendor make half the effort the AMD guys do.

So again, where is the bread and butter?
Click to expand...
Where does AMD make most of their profits, or revenue?
I'll tell you what one of my developers told me, he believes (and heres really fucking smart too) that AMD is going to put a GPU on the die and 256 megs of cache. Take out the video card all together. That is why they purchased ATI. And they are developing their own systemboard too. Think of the advance!
Click to expand...
The combination of CPU and GPU your friend is alluding to is probably the future Fusion product on AMD's roadmap. That is fairly well known as is their overall interest in acquiring ATI. The systemboard you mentioned is news to me, but if they are developing motherboards for their products, it's a long time in coming.
 
Sure do it to recover bad cores. But let us not pretend it is anything else. Think about the realities of pricing.

An Intel Q6600 is only $50 more than an E6600. So if AMD were to compete the same way, the tri core would be $25 more than dual?

This might have made sense if AMD got to quads first and was the only game in town where they could charge you $400 for quad, $300 for tri and $200 for dual, but with intel squeezing the quad core margins, I think they will pretty much have to sell at almost the same price as duals to move.
 
Snowdog said:
An Intel Q6600 is only $50 more than an E6600. So if AMD were to compete the same way, the tri core would be $25 more than dual?

This might have made sense if AMD got to quads first and was the only game in town where they could charge you $400 for quad, $300 for tri and $200 for dual, but with intel squeezing the quad core margins, I think they will pretty much have to sell at almost the same price as duals to move.
Click to expand...
There's more than one way AMD can market these processors. They can drop the price of all their products, effectively undercutting Intel across the board. IOW, the tri-core could compare to an Intel dual-core in price, the quad-core to Intel's forthcoming tri-core, etc. How successful this strategy will prove in the end, is anyone's guess.
 
APOLLO said:
There's more than one way AMD can market these processors. They can drop the price of all their products, effectively undercutting Intel across the board. IOW, the tri-core could compare to an Intel dual-core in price, the quad-core to Intel's forthcoming tri-core, etc. How successful this strategy will prove in the end, is anyone's guess.
Click to expand...

This would be known as the suicide solution. Intel has a big process/manufacturing advantage. It's 45nm cores cost less to make. AMD is losing money, extending the price war even further is not in their survival interest.

Who says AMD gets to choose, what if Intel prices quad cores at tri core prices. AMD will have to be very careful on pricing.

Personally I will start of with a quad capably motherboard with the cheapest dual I can find, eventually I will put a fast quad in there when they are cheaper and better supported. Tricore just doesn't enter into the picture.
 
You must log in or register to reply here.
Back
Top