The AMD Execution Thread [2007 - 2017]
- Thread starter overclocked_enthusiasm
- Start date
-
- Tags
- amd
- Status
Oh, sorry, wrong "news"Yeah, it was posted alright... by you.
Go back 4 posts in this thread.
. Fixed! Thanks for pointing out!http://news.zdnet.com/2100-9584_22-6224419.html
AMD confirmed that Sandtiger is delayed to 2010.
Native octal-core Bulldozer chips will have to wait until then.
If there is a Bulldozer quad-core SKU, it might sneak in before Sandtiger, but the design effort would basically parallel the octal, so there aren't many reasons for there to be a massive difference in release dates.
Unless AMD switches things around and decides to release the new architecture for desktops before the server chips, Bulldozer and its variants seem likely to slip into 2010, perhaps very late 2009.
AMD confirmed that Sandtiger is delayed to 2010.
Native octal-core Bulldozer chips will have to wait until then.
If there is a Bulldozer quad-core SKU, it might sneak in before Sandtiger, but the design effort would basically parallel the octal, so there aren't many reasons for there to be a massive difference in release dates.
Unless AMD switches things around and decides to release the new architecture for desktops before the server chips, Bulldozer and its variants seem likely to slip into 2010, perhaps very late 2009.
We don't know how that combination of existing CPU and GPU will workout. Maybe it is better than the Bulldozer would ever be. An intermadiate solution might be even faster. Remember that AMD boasted Phenon as a native quad core against Intel "fake" quad core?
Anyway, what I really found bad about AMD on that article was this:
"According to sources, AMD's board has shown some irritation at the lack of AMD's execution, but a number of the members don't have a technical background in the PC and server industries. As a result, they rely heavily, like a number of corporate boards, on management's assessment and recommendations. Those recommendations haven't worked out so well over the past 12 months, and further missteps could be impossible to ignore. AMD declined to make any of its directors available for comment."
Ugh @ Sandtiger indeed being delayed to 2010. I still had some hopes it'd be there in mid-2009 and that roadmap was just bogus. Anyway, this part is just ridiculously screwed up:
It helps if you do a couple of nice things for them, as it builds up loyalty, but that's not the most important part. Case in point: Opteron. AMD's share isn't down to 12% only because they screwed up on execution; their roadmap in general was just quite subpar, yet I'm sure customers also "approved" of it. And then you've got other things like OEMs asking for socket compatibility which is why you see HT3 being wasted on Barcelona.
Anyhow, must be pretty tough being in the position of AMD's board of directors. At this point, they probably should ask for outside help from people/consultants with better industry experience; heck, I can see a few persons in the tech press that could even do a pretty good job at this kind of thing, so there's really no excuse here.
If I had to pin-point one 'culture' problem at AMD, it's that whole 'customer-centric' stuff. Here's a big hint: OEMs and customers in general don't have any clue whatsoever what they want. Asking them is just a waste of time, because what they *really* want is just better products.
It helps if you do a couple of nice things for them, as it builds up loyalty, but that's not the most important part. Case in point: Opteron. AMD's share isn't down to 12% only because they screwed up on execution; their roadmap in general was just quite subpar, yet I'm sure customers also "approved" of it. And then you've got other things like OEMs asking for socket compatibility which is why you see HT3 being wasted on Barcelona.
Anyhow, must be pretty tough being in the position of AMD's board of directors. At this point, they probably should ask for outside help from people/consultants with better industry experience; heck, I can see a few persons in the tech press that could even do a pretty good job at this kind of thing, so there's really no excuse here.
Intel slides about Nehalem
It will be quite difficult for AMD to fight against something like that. It'll take at least a year until it has something better than Shanghai and I'm not that sure if Bulldozer can help them much.
IMC helped AMD quite a bit. Later Intel could improve their platform enough so their MC could be very competitive with AMD solution in most cases besides bandwidth intensive ones. Now with Intel going to IMC they'll likely get some nice boost from it, at least with memory latency (it can only get better than their current solution) and most definitely with bandwidth.
Another thing I like is the additional memory slots. Having 12GB of RAM is quite sweet, even though it is only for the high-end.
It will be quite difficult for AMD to fight against something like that. It'll take at least a year until it has something better than Shanghai and I'm not that sure if Bulldozer can help them much.
IMC helped AMD quite a bit. Later Intel could improve their platform enough so their MC could be very competitive with AMD solution in most cases besides bandwidth intensive ones. Now with Intel going to IMC they'll likely get some nice boost from it, at least with memory latency (it can only get better than their current solution) and most definitely with bandwidth.
Another thing I like is the additional memory slots. Having 12GB of RAM is quite sweet, even though it is only for the high-end.
If the site is accurate, the high-end Bloomfield will not sport triple-channel DDR3 with an IMC, but rely on a northbridge.
At least in that respect, AMD's chips with IMCs won't be beaten by Intel's IMC products.
I'd expect there to be a latency penalty that would result from this, but Intel has already demonstrated its ability to make up for a fixed ~20 ns penalty.
The Quickpath connection will sport more bandwidth than the Core2 FSB, so it seems probable that even without an IMC Intel's top desktop chip will excel on bandwidth-limited tasks on the desktop.
Nehalem with IMC--if the controllers on AMD's chips are of similar latency--will potentially have better memory latency than AMD, since Nehalem doesn't go for the 3-level exclusive cache arrangement that took out most of the latency advantage A64 had over Core2.
On the moderately good news front, the Nehalem product that is paired with a GPU is a 2Q 09 product, which means AMD's Swift (if both chips are on schedule) should lag by at most 6 months, whereas before it wasn't certain if Intel would release the product in 4Q 08 or 1Q 09.
At least in that respect, AMD's chips with IMCs won't be beaten by Intel's IMC products.
I'd expect there to be a latency penalty that would result from this, but Intel has already demonstrated its ability to make up for a fixed ~20 ns penalty.
The Quickpath connection will sport more bandwidth than the Core2 FSB, so it seems probable that even without an IMC Intel's top desktop chip will excel on bandwidth-limited tasks on the desktop.
Nehalem with IMC--if the controllers on AMD's chips are of similar latency--will potentially have better memory latency than AMD, since Nehalem doesn't go for the 3-level exclusive cache arrangement that took out most of the latency advantage A64 had over Core2.
On the moderately good news front, the Nehalem product that is paired with a GPU is a 2Q 09 product, which means AMD's Swift (if both chips are on schedule) should lag by at most 6 months, whereas before it wasn't certain if Intel would release the product in 4Q 08 or 1Q 09.
From what slide did you deduce that? From those 2P/4P ones? Look at the last slide on page one, there IO hub and ICH are separate from RAM. I'm sure that on 4P image they put the RAM near IO hub just to save space on the slide
From what I've understood Intel will have direct memory connection with three and two channel CPUs. Single channel ones will use external memory controller.
[edit]
No, wait. The second page has something I didn't notice before and what is different from what I previously known.
[edit2]
Google translation:
Now that's really odd. At least the two and three channel models should all have direct memory connection (why else would they need different pin-count?). If indeed Intel decides to go with external memory controller it might indicate there are problems with their integrated one.
The information is kind of contradicting as the third page talks about three-channel integrated memory controller. With the highest-end CPU they need IO hub to connect more than one GPU (dual-channel one has integrated PCIe controller in CPU)
Last edited by a moderator:
Wolfdales seem to be trickling down the channel:http://www.xtremesystems.org/forums/showthread.php?t=171938.
So some more pain incoming.
So some more pain incoming.
vertex_shader
Banned
It depends. If you lost all of that money, would it break you? If not, Id ride it out at this point, if yes wait a few days they should bounce back to 7 relatively soon.
You could do one other thing. Consult a tax advisor and see if you can avoid paying some taxes by claiming a loss from the sale of these stocks.
Heh, I only had $1000 invested in them so it's not a big deal.
I bought AMD around $14. Thankfully I also bought Intel around $20 so that should soften the loss somewhat, as well as the Best Buy stock I bought at the time time. Well, they were until I just checked and both dropped about $2.00 today.
Invest for the long term. You could also dollar cost average your stock purchase. Get $500 in amd stock now and it should average to $9.9/share. Or put $1000 and it averages down to $8.6/share.
That's assuming the stock prices won't continue to fall ... ala SCOX.
I dont see amd going down too much longer. At some point the pieces would be more valuable than the whole. A fund of some sort would buy it just to break it up and sell the pieces. I think we are very close to that point.
Except AMD competes in the high tech silicon market, a market where being 2nd place means you're worthless and a single provider can scale to cover the entire market. I'd say AMD's engineers are about the only real value the company has right now.
Fabs are worth a lot. They're not cheap to make or equip by any stretch of imagination. AMDs are fairly advanced and would be interesting in themselves, and certainly not very cheap. I don't think AMD will die as in disappear. Hector and the gang might go away and ownership might change, but they'll still be around, same name, new masters, new management team, possibly new focus.
For those investing for the long-term, Phil Hester has done one of the following things:
1) confirmed many month's worth of rumors
2) misspoke
3) was misquoted
in this blog entry.
http://www.mercextra.com/blogs/taka...sday-experiencemeeting-with-amds-phil-hester/
Specifically, it's the part where he states that AMD is "on schedule" for the introduction of 45nm products in 2009.
That's right, not the fuzzy 2H 08, just a solid year running with its uninspiring process and uninspiring design.
I'm also running across rumors that the B3 stepping has failed to fix the TLB bug, so it's possibly a full 6 months of a buggy version of an uninspired design.
I really don't know where the break-even part of their previous conference call is going to come from.
AMD is managing to underpromise and still underdeliver.
1) confirmed many month's worth of rumors
2) misspoke
3) was misquoted
in this blog entry.
http://www.mercextra.com/blogs/taka...sday-experiencemeeting-with-amds-phil-hester/
Specifically, it's the part where he states that AMD is "on schedule" for the introduction of 45nm products in 2009.
That's right, not the fuzzy 2H 08, just a solid year running with its uninspiring process and uninspiring design.
I'm also running across rumors that the B3 stepping has failed to fix the TLB bug, so it's possibly a full 6 months of a buggy version of an uninspired design.
I really don't know where the break-even part of their previous conference call is going to come from.
AMD is managing to underpromise and still underdeliver.
- Status
