Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Alan, re: Don't really know [when DDR2-800 registered is due to become available in volume] but as mentioned it will only be a few percentage points different in performance...
For QC it will be quite a bit more than a few percentage points. For DC I think you are quite right.
Regards,
Rink
Duke, I don't (eom)
Kate, I can only reiterate: Relax a bit please.
Regards,
Rink
Heinz,
Uncertainty on the street about NGA vs H2'06 K8 performance isn't going away this quarter. And when it becomes certain it might well be in favor of NGA. So unless AMD's market share gains duplicate those of previous quarter AMD isn't going to go over $40 again.
IMO AMD's roadmap is better for say Q2 '07 and after that until Intel gets to 45nm; Intel's roadmap is better for H2 '06 and into H1 '07 until AMD releases rev G and 65nm parts that bin above their 90nm counter parts. For now I'm planning my investments accordingly.
I agree with Chris Tom that it's suspect that Intel used boards with an ATI chipset with ATI drivers. The advantage will probably not be specified any time soon.
Regards,
Rink
Wbmw, re: Go ahead and assume zero latency for HTT. It has very little to do with the memory subsystem anyway, except for being on the path to remote memory, but the HTT contributor for this has always been small to begin with.
...except for being on the path to remote memory... Exactly right. It's a significant contributor there. Any decrease in HTT latency will be relavant for the performance of 2 and especially 4 socket servers. I agree though that a more sophisticated HTT protocol is an order of magnitude more important and that it's not due to arrive anytime soon.
I agree with the rest of your post.
Regards,
Rink
Kate, If you relax a bit in your postings I'll reply.
Regards,
Rink
Alan, I agree largely again. Some small details: Kentsfield is a MCM. 1GB/s is for current processors and for specint.
I do have a question though. I wonder when DDR2-800 registered is due to become available in volume. I thought it would be just slightly after Kentsfield / QC Opteron intro, say ~ Q1 or so '07, but I have never known for sure. So, am I wrong?
Regards,
Rink
Wbmw, re: 21GB/s for QC Opteron vs. 17GB/s for QC Clovertown
Remember 17GB/s for Cloverton includes Cloverton to Cloverton traffic as well, plus QC might see higher frequency HTT than currently available.
re: However, Clovertown's memory is all off of a single memory controller, while QC Opteron will have half local and half remotely on the other socket. That might end up leveling the latency landscape as well.
FBDIMM adds latency as well. The fact that Intel needs a NB adds latency compared to AMD's solution as well. Again any higher frequency HTT reduces latency for AMD beyond current solutions.
re: You can assume the test was completely designed to give Intel an advantage, but given the desktop comparison, I don't think AMD investors should bank on this scenario.
If you are refering to the same desktop comparison as I think you are (Conroe vs FX60 at IDF) that comparison was completely designed to give Intel an advantage as well.
AMD used to be 100% of my portfolio. That started to change this January. About a week ago I sold more chunks. I'm now at 25% AMD.
I might trade a bit in Intel after Q1 earnings. I might even go start to own it for medium term prospecs starting around end Q2.
re: I expect AMD to substantially underperform in desktop and mobile, and slightly underperform in servers. At least until G-step.
It's clear we don't agree with regards to servers. You might be excluding virtualization. QC too. I agree though that in desktop and mobile AMD will underperform but not by more than roughly 15% overall.
Regards,
Rink
Alan, nice summary. Doesn't change anything in my thinking though as it is in line with it. It's a bit strange to see your post lacking any comment on AMD's quad core that's coming relatively soon on the heels of Woodcrest and won't have the bandwidth bottleneck issues of Kentsfield. Quad core will put momentum in the 2-socket space back to AMD's side after it went to Woodcrest for a while.
Regards,
Rink
Wbmw, I think you are forgetting that AMD has the right bandwidth-latency combination to do quad core in servers with more than 1 socket and Intel has not until CSI. This added to core revisions, and HTT revisions, and better performing virtualization will rather easily allow AMD to keep momentum for years to come UNLESS Intel's 45nm is spectacularly good (which it might be because of high-k but I'm not exactly holding my breath).
Just to be clear for the desktop environment I'm expecting somewhere between 3-10% generic performance increase for rev F on AM2 compared to rev E on s939, based on rumors and early sample test results published by Anand.
With regards to the transition from 90 to 65nm AMD is not in the same situation as when they were with the transition from 130nm to 90nm. I'm not expecting 65nm parts exceeding 90nm top bin until into H1 '07. Rev G IPC and frequency improvement come on top of that.
All in all I expect AMD to slightly underperform in desktop and mobile until rev G. I expect AMD to continue to excell in the server environment.
Even more generally I think that server + increased DT and Mobile volume + decreasing depreciation will generate increasing profits next year.
Regards,
Rink
Wbmw, I've seen the Anand review a couple of times yes. I'm not so sure as you are that it'll represent an accurate picture of Conroe's performance vs. 90nm rev F.
1. Anand used an early (dec) sample that was said to have a sub-par mem contr.
2. We haven't seen *any* Conroe vs. Rev F review.
Yes Intel gave performance estimates both in general terms and more specific ones, and yes Intel shows some confidence in showing off Conroe vs rev E, and yes Intel gave a nice summary of the technical advances in cache and computing units. But all in all there's not sufficient information for me to be able to have a solid picture of the extent of Conroe's performance advantages over 90nm rev F (let alone 65nm shrink of rev F).
I was hoping Chipguy could have been a bit more specific as he has the background to form a better idea than me. If he feels he can't do that just yet I think there's not much hope for most posters here (including me) to get a solid picture.
Regards,
Rink
Chipguy, I was hoping you could have been more specific.
About those SPEC numbers. Do they depend more on SSEx or old x86 ALU/FPUs?
Regards,
Rink
Chipguy, thanks for the info! Happy I was mostly out in time. I need to know what it means compared to H2 DC Turion, Athlon X2, and Opteron though. Do you have any idea in terms of tens of percentages of Intel performance increase compared to competition?
Regards,
Rink
HailMary, dumped three quarters of my position on open as well. Only one quarter remaining. I'm sceptical short term because of Intel's marketing but might get back in again before earnings. I think AMD potentially has more earnings / market share surprises left this year.
Regards,
Rink
Doug, they didn't provide a P4 ref system because that would amount to negative publicity and more Osbourneing than necessary.
Regards,
Rink
I'm not going to comment to that (eom)
Chipguy, re: Yeah AMD is having sweet little run of a few great quarters going but that isn't the big picture.
You call that board approval the big picture? Sorry, but you lost me at ~$20.
Regards,
Rink
"may pick up" (eom)
Imho you're not so humble.
Here's an example of what you wrote: Furthermore, look at what extent AMD has gone to give the impression that it is "serious" about a plant in India:
http://www.financialexpress.com/fe_full_story.php?content_id=110223 : "SemIndia, AMD to chip in with $3 bn. AMD’s chip plant in India to put pressure on rival Intel."
This was around the time Intel announced its India investments. Talk about, hype. Talk about smoke and mirrors. Believe what you want to believe. I think AMD is starting to show its true colors.
That wasn't an AMD press release and neither were those anywhere near AMD's words. I mentioned why those words could not have come from AMD's mouth. Yet you claim now that AMD intentionally mislead the authors of articles like the one you mentioned by being purposely vague so that most of them are taking up the wrong message that AMD has a new plant in India?!! This is the first article that I've seen that mentions this. I think you're purposefully placing a single quote out of the proper context of the many related articles of much better quality to support your cry that "the majority of what comes out of AMD's mouth is pure BS".
Regards,
Rink
Tom's first look at AM2: http://www.siliconinvestor.com/readmsg.aspx?msgid=22188539
He did indeed use a Dec '05 engineering sample with less than ideal memory speed settings.
Regards,
Rink
Paul, I think Tom's evaluating the broken version because of this quote:
"However, it is clear that Socket AM2 is not just about performance, at least not until the arrival of DDR2-800 in the mainstream market. With current DDR2-667 memory, very little improvement should be expected as the integrated memory controller suffers from relaxed memory timings."
From what we heard about why a new stepping was needed and why AM2 was delayed till June was that the MC needed performance enhancements (source: Charlie/Inq). The above sounds to me like those improvements are not yet available in that engineering sample.
Regards,
Rink
Imho, I think you should manage your expectations better. AMD isn't reversing any previous stance which is what I think you imply; it's only providing more detail.
AMD signed an agreement that SemIndia can use it's process IP for 90nm when that new fab is finally finished (>18 months from now). It never said it intended to take a stake in SemIndia, just that it was an option that they now seem not to be using. Like we say in Holland there are a thousand ways to Rome. Like subcontracting (Chartered model comes to mind for Geode's).
That other line you quoted ("AMD’s chip plant in India to put pressure on rival Intel") is completely fabricated by some short circuits in the authors brain. Like a sound bite with almost no truth behind it. AMD never had a chip plant in India, and never signed any agreement that it would build or own a plant there.
Regards,
Rink
Chipguy, iirc K7 was designed for two levels of cache although only one level was first implemented.
Although as you say AMD currently doesn't have the fab capacity add moster L3 caches on mainstream volume products fab capacity is expanding at a rather steep pace this year (fab 36 ramp + Chartered fab 7) and will allow reasonable sized L3 caches (like the 4MB the Inq mentioned yesterday) in FX and Opteron lines even if on 90nm.
Wow 7ns 12MB L3 is incredibly fast. That isn't what Itanium is using though, is it?
Regards,
Rink
Here's some nice digging by Petz/SI on x86 servers vs. Itanium: http://www.siliconinvestor.com/readmsg.aspx?msgid=22171200
Regards,
Rink
SP/Keith, Chipguy has been claiming for some time that ISS contains all Windows based servers (among other things) including Itanium based ones. Even if true this should be a rather small percentage of ISS compared to x86 based servers.
Does anyone here know why ESS is down 5% rev.-wise compared to the preceeding quarter? Is that because of big tin declines?
Regards,
Rink
Advantages rev F K8 vs. NGA:
- Memory bandwidth
- Memory latency
- IO Virtualization (servers/workstations)
- Mature 64b implementation
- Inter socket latency (servers/high end workstations)
- Mobo design simplicity (servers/high end workstations)
Disadvantages:
- Power
- Cache size/density
- Lack of 'Genuine Intel' CPUID
Bottom line:
- NGA is not exactly a full generation ahead of K8 and K10 will be a full generation ahead of NGA.
- Intel is significantly earlier to 65nm but AMD's 90nm SOI + SiGe in addition to increasing fab capacity fully negate the disadvantage, and once 65nm starts maturing AMD will have a significant process advantage again.
- Intel is the dominant player willing to apply effective measures that leverage this position to control cpu demand. Some cracks appeared in the control building, but not in the fundament.
Regards,
Rink
Mas, could you explain why you think the 3GHz frequency of the coming SC Opteron 256 is the result of the addition of SiGe?
Regards,
Rink
Does anyone know when the NGA EE parts will be available?
Regards,
Rink
Wmbw,
Well where do you think the bonusses of Intel flash people and some other indirect costs are kept?
Don't you think separation of Spansion might have cost some money?
Lastly are you sure that the cost structure of Spansion did not in fact increase permanently because of the separation?
Regards,
Rink
[Edited :)] Tx Buggi!!
BTW, Merom probably slipped to Q4: http://www.siliconinvestor.com/readmsg.aspx?msgid=22149035
Regards,
Rink
Buggi, you don't think that's the effect of added SiGe already right?
Regards,
Rink
Another cross post: Rev F grew to 220mm^2, 13% die increase, 4% more trannies, 7% L2 die size reduction. Pics of Rev E, and rev F side by side. Discussion starts here: http://www.siliconinvestor.com/readmsg.aspx?msgid=22146135
Regards,
Rink
Keith, that rumor appears to be true.
Report of ThinkEquity with same title as I posted a full quote from appeared on Ameritrade (so that quote was probably authentic): http://www.siliconinvestor.com/readmsg.aspx?msgid=22145488
I wrote Charlie but he was already on to the reason behind the delay (FBDIMMs being late, hot, and in short supply): http://www.theinquirer.net/?article=29576
I'm seeing that specific reason as a positive for AMD even after Dempseys launch date for obvious reasons.
Regards,
Rink
Well it sure would be nice if Merom wouldn't be compatible with Napa provided Santa Rosa is indeed delayed till '07 as the article claims.
Regards,
Rink
Chipdesigner, the core of the claim that Merom is delayed till 2007 is actually a bit different, namely that Merom platform Santa Rosa is delayed till 07. What makes you think that Merom won't fit in current Yonah platform Napa in the mean time?
I posted that Dempsey delay claim both here and on AMD/SI boards but must admit I haven't seen it anywhere else. It can't be googled yet either.
Regards,
Rink
x-post SI: 08:09 INTC Intel: downgrade details (20.67 ) -Update-
ThinkEquity is downgrading INTC to Accumulate from Buy and lowering their tgt to $26 from $27 saying Intel's Dempsey processor appears to have slipped from March to May 2006 (6- 8 weeks). ...
http://www.siliconinvestor.com/readmsg.aspx?msgid=22144920
The report claims Dempsey is Intel's first "true dual core". What a laugh.
This delay is btw at least as significant as the Montecito launch.
Regards,
Rink
Wbmw, you rather obviously didn't read the article I posted. The configs are in it (page 2). I can only repeat that hundreds of percents improvement for the 64b version of LightWave is a rough avg from what I read yesterday.
Regards,
Rink
Wbmw/Buggi, here's one more. VERY nice performance improvements in 64b version of LightWave: http://www.planetx64.com/index.php?option=com_content&task=view&id=5&Itemid=49&limit...
Hundreds of percents.
Regards,
Rink
Maybe the 3GHz is single core? I mean that would make sense, although I admit I don't give this much chance.
Regards,
Rink