tag:blogger.com,1999:blog-2602471396566186819.post1750233727926806463..comments2023-10-26T15:06:30.940+00:00Comments on AIMeD Corporation: AMD's Q1 2008 Earnings (or lack of) ReportRoborat, Ph.Dhttp://www.blogger.com/profile/04845879517177508741noreply@blogger.comBlogger126125tag:blogger.com,1999:blog-2602471396566186819.post-62256746658477336502008-04-30T17:56:00.000+00:002008-04-30T17:56:00.000+00:00“You've hit 4GHz very easily. Are you increasing t...“You've hit 4GHz very easily. Are you increasing the CPU multiplier, or the FSB to OC at this stage?”<BR/><BR/>The 4 Gig run was done strictly by a 10X multiplier, with memory set at the board natively assigned DDR3-1333 bios parameter. Incidentally, also listed in those options are, DDR3- 1600 *DDR3-1600 O.C.*, and *DDR3-1800 O.C.* I had to manually assign this parameters, but the board SAW the 1600 native.<BR/><BR/>Subsequently, I keyed in the DDR3-1600 native and checked the latency, it went down to 57ns. That’s well within reach IMC. <BR/><BR/>There is an interesting option I have, frankly, never seen before. The frequency multiple can be increased or decreased by .5. I always felt that a full multiple was too much of a jump; ASUS has addressed this issue quite nicely.<BR/><BR/>“BTW, have you picked up an equally impressive video card do go with this monster CPU?”<BR/><BR/>Unfortunately, no, I haven’t. I am still using the 1900XTX Crossfire set up which really isn’t bad. The scores I got with the setup along with the Q6600 were 11,490. With this chip the scores increased to 12,857, not too bad for 2 year setup. They’ve got some new things on the horizon in the interim. I really would like a substantial increase. <BR/><BR/>The ATI purchase really turned the graphics industry sideways. <BR/><BR/>My next purchase will be that “electric cooler” we spoke. GURU’s Electromigration, and carrier mobility abstracts have me pissing my pants. The next thing you know I’ll be wearing a dress and high heels.<BR/><BR/>With that in mind, that E8400 is absolutely beautiful, spectacular, in fact. I thought that Q6600 was something irreplaceable and unique. Man, was I all wet, it was only the beginning.<BR/><BR/>I’ll keep you posted as I develop a relationship with the new chip. Next stop, 1800 FSB, than to 4Gig and beyond.<BR/><BR/>SPARKSSPARKShttps://www.blogger.com/profile/05535419513995195852noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-73715360131198850812008-04-30T03:41:00.000+00:002008-04-30T03:41:00.000+00:00"Actually GURU scared me a bit when he gave me a n..."Actually GURU scared me a bit when he gave me a new concept to read, carrier mobility"<BR/><BR/>I wouldn't worry so much about mobility - if you are OC'ing you may have to eventually worry about electomigration though!<BR/><BR/>Sweet dreams while you think about that...Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-11183177374152778482008-04-29T14:57:00.000+00:002008-04-29T14:57:00.000+00:00Hey GURU! It just dawned on me! I personally am go...Hey GURU! It just dawned on me! I personally am going to put this 'low bin' (old)Q6600, 'top bin' QX9770 thing to the ultimate test!<BR/><BR/>Same motherboard, powersupply, memory, bios, RAM, basically the whole f**king magilla, out of a retail box. No spin, just numbers on air, then with some di hydrogen oxide.<BR/><BR/>Now we have top bin vs. top bin!<BR/>HOO YAA!<BR/><BR/>Any takers on Super Pi 1M @ 11 second?<BR/><BR/>BTW: We've got to go over this carrier mobility thing. Christ, those formulas scrambled my brain, or what's laft of it. Looks like my little buddies (electrons) start running into each other and don't where to go, from heat, no less!<BR/><BR/>SPARKSSPARKShttps://www.blogger.com/profile/05535419513995195852noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-20192698439294407362008-04-29T14:20:00.000+00:002008-04-29T14:20:00.000+00:00Giant, it’s good to hear from you. Ah, Q6600 the n...Giant, it’s good to hear from you. <BR/><BR/>Ah, Q6600 the new overclocking darling of the Twenty First Century. I’m afraid to take the sweetie higher than 3.4, incredible! Actually GURU scared me a bit when he gave me a new concept to read, carrier mobility. Whoa! It was like a surgeon, who specializes in carcinogenic abnormalities, suggesting some ‘Light Reading!” I imagined thermal runaway times a couple of hundred million transistors, not a happy day. I backed it off a bit to your similar 3.3 clocks. (I’ll wait for the anti carcinogenic hafnium and some water treatment on the big boy.) I OC’d with the 8X and the 9X multiplier, it’s all the same, the thing just cranks.<BR/><BR/>What make the Q6600 so versatile you can breeze it up to gaming speeds for a gaming rig, or drop it in a media machine (with limited cooling) and let the four cores do all the work @ stock speeds for rendering, etc The P35 seems to be ideal for this application. <BR/><BR/>There is some FUD running around the net that Q6600 will be sold till the third quarter, as if it were a bad thing. (INTC 45nM shortages, I think, frankly, I think they got caught with their pants down; its demand) Whatever the reason, it will be a sad day when Q6600 gets EOL’d.. Actually, 65nM (GO) is running so well they could keep it in the lineup for another quarter. That said, I know, it’s back to business, yields, DPW, margins, and all that, but man, what a ride.<BR/><BR/>Q9300 is out and from what I’m reading it cranks just as well, if not better, with lower thermals. F**K’N A BUBBA!<BR/><BR/>Wait! Wait! Is that the UPS truck?!? Ah, false alarm.<BR/><BR/>Stay in touch, Giant, that other guys has got GURU’s nuts twisted again. This time GURU came out swinging, and it hasn’t been pretty. <BR/><BR/>I’ll keep ya posted.<BR/><BR/>SPARKSSPARKShttps://www.blogger.com/profile/05535419513995195852noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-46541039652086641932008-04-29T10:32:00.000+00:002008-04-29T10:32:00.000+00:00hyc... you twist and twist these arguments when yo...<I>hyc... you twist and twist these arguments when you start to realize you are wrong... so let's take a step back to your original post:<BR/><BR/>"Given that (a few) folks are now showing Phenoms overclocking to 3.5GHz, I think it's also reasonable to believe that AMD actually expected to be able to deliver 3.0GHz parts to the market."<BR/><BR/>you are the one who made a 'resonable' claim, I was merely pointing out that:<BR/><BR/>A) This is a real bad conclusion (use of an OC to forecast stock speed parts or expectations thereof)<BR/>B) It is impossible to tell whether the 3.5GHz overclock is the norm or an outlier... given the 1.328V Vcore and use of water cooling, and a very limited data set of folks who have achieved this... I'm skeptical and even if this were 'normal' (which there is no evidence one way or the other) go back to point A<BR/><BR/>It appears as though you are trying to excuse AMD for their false claims by drawing conclusions on an incomplete data set. While that may be the norm at Scientia's blog, you will get challenged here.<BR/><BR/>So do you stand by your above claim? <BR/></I><BR/>I thought I already made it clear that I conceded on your points and that I agreed that you were right. What part of "I'm not disagreeing" did you not understand?<BR/><BR/><I><BR/>So now you can go on and on about trying to discredit minor points to distract people from your initial erroneous and/or unsubstantiated claims - this seems remarkably familiar to a recent string of posts about 64bit and is getting tiring.</I><BR/><BR/>I think you may find it's tiring because you don't know to stop attacking after you've already won the point.hychttps://www.blogger.com/profile/15473250487285924085noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-13060525224895378492008-04-29T04:53:00.000+00:002008-04-29T04:53:00.000+00:00My Q6600 came out of a retail box; I seriously dou...<I><BR/><BR/>My Q6600 came out of a retail box; I seriously doubt the AMD chip did. Also, if I recall Giant has achieved the same result with his Q6600. Thats 2 for 2.</I><BR/><BR/>That's correct. Furthermore, that is a year old B3 stepping Q6600 on air cooling. With a G0 stepping and a great motherboard and ram like Sparks has 3.6GHz should be possible on air cooling. My 3.3GHz OC is 24/7 100% stable on a near two year old P5B Deluxe board. The same board is good for 4.5GHz with my E8400 (though the nForce 790i board in my main machine is better at OCing, the Q6600 just becomes thermally limited past 3.3GHz, so it wouldn't be wise to take it any further)<BR/><BR/>Comparing a once off OC on a Phenom that wasn't entirely stable using water cooling to an OC that is easily achievable on most Q6600s using a high end air cooling is just stupid.<BR/><BR/><I><BR/><BR/>The big fella's coming soon, the Q6600 is coming out and going back to wifes machine. </I><BR/><BR/>Now that's we've been waiting for! Can't wait to see some OCing results on the QX9770! 4GHz should be a walk in the park for that monster!Unknownhttps://www.blogger.com/profile/04674699447174785970noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-81504871256135027352008-04-29T02:38:00.000+00:002008-04-29T02:38:00.000+00:00hyc... you twist and twist these arguments when yo...hyc... you twist and twist these arguments when you start to realize you are wrong... so let's take a step back to your original post:<BR/><BR/>"Given that (a few) folks are now showing Phenoms overclocking to 3.5GHz, I think it's also reasonable to believe that AMD actually expected to be able to deliver 3.0GHz parts to the market."<BR/><BR/>you are the one who made a 'resonable' claim, I was merely pointing out that:<BR/><BR/>A) This is a real bad conclusion (use of an OC to forecast stock speed parts or expectations thereof)<BR/>B) It is impossible to tell whether the 3.5GHz overclock is the norm or an outlier... given the 1.328V Vcore and use of water cooling, and a very limited data set of folks who have achieved this... I'm skeptical and even if this were 'normal' (which there is no evidence one way or the other) go back to point A<BR/><BR/>It appears as though you are trying to excuse AMD for their false claims by drawing conclusions on an incomplete data set. While that may be the norm at Scientia's blog, you will get challenged here.<BR/><BR/>So do you stand by your above claim? Given that AMD themselves did not have a 3.0GHz part on their roadmap through Q2'08 BEFORE ALL THE TLB AND INITIAL RELEASE OF K10, I'm not sure how you can make the assertion that it's reasonable that AMD expected to have 3GHz parts, especially, again, when it wasn't on their own roadmaps.<BR/><BR/>So now you can go on and on about trying to discredit minor points to distract people from your initial erroneous and/or unsubstantiated claims - this seems remarkably familiar to a recent string of posts about 64bit and is getting tiring. <BR/><BR/>"...Is that what you were trying to say?"<BR/><BR/>No what I was trying to say is taking an OC on a part and using that to predict what you should be able to do in the future is not reasonable and to use that to defend a chip manufacturer (who would certainly know better) is also wrong.<BR/><BR/>Then to compound that by comparing an OC on a top bin part vs the OC on a bottom bin part is also not a very good comparison (typically bottom bin parts catch most of the crap that falls out of the upper bins).<BR/><BR/>It's frustrating that given the lack of data you want to interpret everything in the best possible light... I generally go by the point of view that until something is proven (or at least substantiated by significant data) you should go by a more conservative point of view.<BR/><BR/>We both agree there is 'limited' (to put it favorably) data on a 3.5GHz overclock... given the limited data I would not be using it to make arguments and excuse AMD and use it to support AMD 'reasonably' thinking they could get to 3GHz on 65nm. (This is the logic issue... then of course there are the technical issues which I beat to death about using an OC to predict stock performance)Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-86246732197038607862008-04-29T02:06:00.000+00:002008-04-29T02:06:00.000+00:00By the way fellas, if need to know anything more a...By the way fellas, if need to know anything more about the Q6600, let me know I try it out, right here on this keyboard. <BR/><BR/>The big fella's coming soon, the Q6600 is coming out and going back to wifes machine. <BR/><BR/>THEN we'll see what the top bin badboy can do. HOO YAA!<BR/><BR/>SPARKSSPARKShttps://www.blogger.com/profile/05535419513995195852noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-88476257308490118872008-04-29T02:00:00.000+00:002008-04-29T02:00:00.000+00:00I was not intentionally FUDing. I just missed the ...I was not intentionally FUDing. I just missed the significance of one of your statements. I apologize for not reading more carefully.<BR/><BR/>Q6600 is the bottom bin, for both speed and thermals. 9850 is the top for current speed, but more importantly, thermal limits. So while the 9850 may have headroom for frequency, in its current state it'd be unlikely that any faster chip would fit in the thermal limits, and therefore a 3GHz product is unlikely.<BR/><BR/>Is that what you were trying to say?hychttps://www.blogger.com/profile/15473250487285924085noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-89348735606623060362008-04-29T01:55:00.000+00:002008-04-29T01:55:00.000+00:00“And then you say 3.3 is a comparable OC on the 66...“And then you say 3.3 is a comparable OC on the 6600? with water cooling? on a small population chips? (or on a good majority?)”<BR/><BR/><BR/>Ah----no way. I have a Q6600 @ 3.38 1.4125 Vcore it does SuperPi 1M in 15 seconds. It PLAYS stable all day. I could be sandbagging here, the X48 chipset has no problem with a 1600 FSB. Not many folks have such a board.<BR/><BR/>Here is a link that did have the Phenom X4 9850 clocked to 3.5 but it was unstable, “but it preserved stability only at 3.2 GHz”<BR/><BR/>At 3.5 it ran Pi 1m in 22 sec.<BR/><BR/>My Q6600 came out of a retail box; I seriously doubt the AMD chip did. Also, if I recall Giant has achieved the same result with his Q6600. Thats 2 for 2.<BR/><BR/>Both chips are on air, the Phenom is on water.<BR/><BR/>Obviously, the thermals are lower, and the architecture is more efficient, clock for clock. But, we already knew this.<BR/><BR/>The point is yours.<BR/><BR/>http://xtreview.com/index.php<BR/><BR/><BR/>SPARKSSPARKShttps://www.blogger.com/profile/05535419513995195852noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-50310139738477078742008-04-29T01:27:00.000+00:002008-04-29T01:27:00.000+00:00I think you're changing the subject; we weren't ta...I think you're changing the subject; we weren't talking about dual cores.<BR/><BR/>Your point I was addressing was - does a X% overclock give you a fair indication on what future models will do. Can you reasonably assume from anything from the fact that a 65nm 2.4GHz Q6600 commonly overclocks to 3.3GHz with water cooling? Can you reasonably assume from this fact alone, that 3GHz parts in the same family will arrive down the road? Your answer was No. But in hindsight, those 3GHz parts were indeed introduced.<BR/><BR/>As for fair comparisons: 3.5GHz for the 9850 may turn out to be exceedingly rare, we don't know yet. (Though there are at least two different people reporting results in that range on xtremesystems right now, so it's not a completely isolated case.) But given that 3.5GHz has been demonstrated, it's logical to assume that 3.2, 3.3GHz will be achieved in greater numbers. So yes, I think it's fair to say that 3.5GHz over 2.5GHz implies that 3.3GHz will be a more common OC, and that puts it in a range that's pretty comparable to what is commonly seen with the Q6600.hychttps://www.blogger.com/profile/15473250487285924085noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-22002043681892677492008-04-29T00:39:00.000+00:002008-04-29T00:39:00.000+00:00"and overclockers appear to be getting to about 3...."and overclockers appear to be getting to about 3.3GHz with it on water. So 3.5GHz with a 2.5GHz 9850 seems like rough parity."<BR/><BR/>You're kidding right? You pull up a rare good OC on 9850 (granted there isn't a lot of chips AMD has made, but @3.5 there is not enough data to suggest this is an 'expected' overclock).<BR/><BR/>And then you say 3.3 is a comparable OC on the 6600? with water cooling? on a small population chips? (or on a good majority?)<BR/><BR/>You are taking THE MOST FAVORABLE data point on the AMD chip and comparing it to a common data point on the Intel chip and then say they are comparable? <BR/><BR/>"it seems if Intel had expected their 65nm process to get to 3GHz or more they would have released such a product already"<BR/><BR/>I know from your past comments you have a bias toward AMD, but please stop with the FUD. You won't get away with the spin on this site - I would like to also note you are comparing the OC on Intel's BOTTOM bin with the OC on AMD's top (best) bin.<BR/><BR/>No 3GHz on Intel 65nm? How about (mix of dual and quad cores):<BR/>http://www.newegg.com/Product/Product.aspx?Item=N82E16819115026<BR/>http://www.newegg.com/Product/Product.aspx?Item=N82E16819115028<BR/>http://www.newegg.com/Product/Product.aspx?Item=N82E16819117134<BR/>http://www.newegg.com/Product/Product.aspx?Item=N82E16819117133<BR/>http://www.newegg.com/Product/Product.aspx?Item=N82E16819117092Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-45334310919763831302008-04-28T23:39:00.000+00:002008-04-28T23:39:00.000+00:00My point on the Intel OC's (ignoring the LN2 resul...<I>My point on the Intel OC's (ignoring the LN2 results) is that I wouldn't expect those overclocks to translate directly to future stock parts either (but if you assume the AMD results are indicative of future gains, then why wouldn't you expect similar gains based on Intel OC's). I think NEITHER OC (minus some correction factor) should be considered a forecast of future clocks - it simply is indicative of the headrooom an architecture potentially has, ultimately the process performance will dictate how close you can get to that point. And at this point, esp on 65nm, AMD has not demonstrated that they could get there - nor was it a reasonable expectation a year ago... and they're own leaked roadmaps were reflecting that,</I><BR/><BR/>I understand your point, and I'm not disagreeing. I linked to the Q6600 results simply for context (which really was largely for my own benefit). Intel lists the Q6600 as a 2.4GHz part http://www.intel.com/products/processor_number/chart/core2quad.htm and overclockers appear to be getting to about 3.3GHz with it on water. So 3.5GHz with a 2.5GHz 9850 seems like rough parity.<BR/><BR/>It seems if Intel had expected their 65nm process to get to 3GHz or more they would have released such a product already. Instead, they shifted to 45nm, to produce the 3.2 GHz QX9770. Everything you're saying makes sense.<BR/><BR/>Also looking at the photo of the AMD 3GHz demo, there's a huge bank of fans being used. http://www.theinquirer.net/en/inquirer/news/2007/07/26/amd-shows-off-30ghz-barcelona<BR/>So OK, I agree on all your points.hychttps://www.blogger.com/profile/15473250487285924085noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-88460049298127268782008-04-28T22:20:00.000+00:002008-04-28T22:20:00.000+00:00HYC - you ignore the forest and focus on the trees...HYC - you ignore the forest and focus on the trees... do you really think OC's represent (with some factoring down) what you should expect for future stock?<BR/><BR/>Yes water is no longer 'extreme'... but for AMD to release these chips commercially we are talking stock air cooling AND low enough TDP's to fit their 'drop-in' to existing MOBO's in the field marketing campaign.<BR/><BR/>I don't see even a relatively mild Vcore increase like to 1.3V (or 1.33V) fit AMD's stock TDP bins. Just look at the Vcores on their current 135Watt bin parts. <BR/><BR/>And by the way the link you provide was 1.328V not 1.3V (unless you were referring to the Vcore for 3.0GHz). I also have no idea if this is typical performance, or an outlier, but quite frankly I have not seen that many reports of AMD chips at 3.5GHz at 1.33Volts<BR/><BR/>Finally I think it would not hold up with an AMD stock cooler. Let's face it, if it could - AMD would be selling them at 3.0GHz and helping their margins, instead of letting folks try to get there on their own with 'black editions'.<BR/><BR/>My point on the Intel OC's (ignoring the LN2 results) is that I wouldn't expect those overclocks to translate directly to future stock parts either (but if you assume the AMD results are indicative of future gains, then why wouldn't you expect similar gains based on Intel OC's). I think NEITHER OC (minus some correction factor) should be considered a forecast of future clocks - it simply is indicative of the headrooom an architecture potentially has, ultimately the process performance will dictate how close you can get to that point. And at this point, esp on 65nm, AMD has not demonstrated that they could get there - nor was it a reasonable expectation a year ago... and they're own leaked roadmaps were reflecting that,Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-65195886034168712102008-04-28T21:54:00.000+00:002008-04-28T21:54:00.000+00:00"Take for example QX9770 rated for 136W as we incr..."Take for example QX9770 rated for 136W as we increase the Vcore and the clocks, does the TDP rise on a linear level, or is there a rise on some esoteric exponential progression?"<BR/><BR/>Power is typically exponentially related to clock speed (and I believe also Vcore - someone correct me if I'm wrong here)<BR/><BR/>Heat has all sorts of impacts - it will impact carrier mobility negatively and exponentially (which in turn lowers overall speed), for those with a semiconductor physics background it also increases minority carriers, which can hurt some of the parasitic capacitances, I think (it's been awhile since I studied this) and of course you touched upon general resistances vs temperature issues.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-81663163408424350732008-04-28T21:18:00.000+00:002008-04-28T21:18:00.000+00:00Sorry, forgot to paste the first linkhttp://www.xt...Sorry, forgot to paste the first link<BR/><BR/>http://www.xtremesystems.org/forums/showthread.php?t=185909<BR/><BR/>3.5GHz @1.3V, water cooling.hychttps://www.blogger.com/profile/15473250487285924085noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-12049429001532272632008-04-28T21:16:00.000+00:002008-04-28T21:16:00.000+00:00I'm not sure I follow the logic here - these overc...<I>I'm not sure I follow the logic here - these overclocks I can only assume have Vcore jacked way up and may have some aggressive cooling(?). AMD's process appears to have a leakage problem which would mean even slight changes to Vcore would mean large changes in TDP (and thus difficulty in getting higher clocks without exploding the thermals).<BR/><BR/>Overclocks are difficult to use as an indication of future stock performance - I think even Scientia correctly points this out on many occasions. It shows the architecture has the capability but it is not necessarily indicative of future products as you have to consider process and thermal and binsplit implications.<BR/><BR/>If I carry your logic and then look at the Intel 45nm overclocks, what clocks would you reasonably assume for Penryn?<BR/><BR/>From the early leaked roadmaps (I think this may have been from HKPEC?), most of them had 2.6 in Q4'07 or Q1'08, and only a modest increase to 2.8GHz in Q2'08 - based on this I would have to assume AMD knew what they could do and even then couldn't deliver on it.<BR/></I><BR/>I can't argue with your fundamental point. But in this case, I don't think 1.3V is out of line. Water cooling is certainly not commonplace yet, but it's also not massively extreme. E.g., it's not something ridiculous like liquid nitrogen...<BR/><BR/>This result seems a bit over the top...<BR/>http://pinoypureview.blogspot.com/2007/10/penryn-overclocked-to-6ghz.html<BR/><BR/>I would certainly not assume from these results that products in the 5GHz range are coming out any time soon.<BR/><BR/>These seem a bit more comparable<BR/>http://www.xtremesystems.org/forums/showthread.php?t=142430hychttps://www.blogger.com/profile/15473250487285924085noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-42737717491488425222008-04-28T20:16:00.000+00:002008-04-28T20:16:00.000+00:00Jack, recall above, my example of AMD’s market sha...Jack, recall above, my example of AMD’s market share loss sighting CRAY as customer who will be lost indefinitely, perhaps I was lucky in my timing and analysis. Why do I get the feeling Nehalem is further along than most suspect.<BR/><BR/>http://www.theinquirer.net/gb/inquirer/news<BR/>/2008/04/28/cray-goes-intel-hpc<BR/><BR/>This is big.<BR/><BR/><BR/>SPARKSSPARKShttps://www.blogger.com/profile/05535419513995195852noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-43297410335029239522008-04-28T19:21:00.000+00:002008-04-28T19:21:00.000+00:00“even slight changes to Vcore would mean large cha...“even slight changes to Vcore would mean large changes in TDP (and thus difficulty in getting higher clocks without exploding the thermals).”<BR/><BR/>I’ve wondered about this dynamic for quite some time. <BR/><BR/>Let’s exclude the lower binned chips, if I’m using the correct terminology, for the sake of argument. Take for example QX9770 rated for 136W as we increase the Vcore and the clocks, does the TDP rise on a linear level, or is there a rise on some esoteric exponential progression? <BR/><BR/>Additionally, why does removing heat enable one to raise the clock rate substantially? Given that the leakage is part of the chips static design parameter, why does cooling or super cooling enable one to clock faster at higher voltages, aside from lowing resistance as a function of the chip operation? Does it lower leakage? Or am I confusing the two functions as interrelated, when they aren’t.<BR/><BR/>Obviously, circuit pathways in the chips design play a role in limiting frequency. I would like to exclude this as a variable, if possible.<BR/><BR/>SPARKSSPARKShttps://www.blogger.com/profile/05535419513995195852noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-8136902651813810232008-04-28T18:21:00.000+00:002008-04-28T18:21:00.000+00:00"Given that (a few) folks are now showing Phenoms ..."Given that (a few) folks are now showing Phenoms overclocking to 3.5GHz,I think it's also reasonable to believe that AMD actually expected to be able to deliver 3.0GHz parts to the market"<BR/><BR/>I'm not sure I follow the logic here - these overclocks I can only assume have Vcore jacked way up and may have some aggressive cooling(?). AMD's process appears to have a leakage problem which would mean even slight changes to Vcore would mean large changes in TDP (and thus difficulty in getting higher clocks without exploding the thermals).<BR/><BR/>Overclocks are difficult to use as an indication of future stock performance - I think even Scientia correctly points this out on many occasions. It shows the architecture has the capability but it is not necessarily indicative of future products as you have to consider process and thermal and binsplit implications.<BR/><BR/>If I carry your logic and then look at the Intel 45nm overclocks, what clocks would you reasonably assume for Penryn?<BR/><BR/>From the early leaked roadmaps (I think this may have been from HKPEC?), most of them had 2.6 in Q4'07 or Q1'08, and only a modest increase to 2.8GHz in Q2'08 - based on this I would have to assume AMD knew what they could do and even then couldn't deliver on it.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-39281441711819782662008-04-28T13:20:00.000+00:002008-04-28T13:20:00.000+00:00hycIt's entirely possible that, averaged over some...<B>hyc</B><BR/><BR/><I>It's entirely possible that, averaged over some (unspecified) mix of workloads, the overall advantage was just 40% at the time they published those claims.</I><BR/><BR/>The quoted claims were that Barcelona would "be 40% faster than Clovertown across a wide variety of workloads" and "would blow away Clovertown in every dimension". Those statements are <I>not</I> consistent with any scenario in which Barcelona beats Clovertown in only three or four benchmarks like SPEC or OpenLDAP. Those statements clearly imply a <I>general</I> level of performance superiority in the server space and that's exactly why so many were so excited and looking forward to the K10 launch. And it's also why everyone was so let down. IMO, AMD had deliberately fostered an expectation in the public that they knew would not be fulfilled.<BR/><BR/>Now you're free to give AMD the benefit of the doubt and believe what you wish regarding their veracity & honesty, but I believe (as do many others) that AMD essentially lied and deliberately misled the public with those statements. They have lost all credibility with me and I no longer trust anything they say. Not to mention a couple other huge claims they backpeddled on:<BR/>- Promised that Quad FX support would continue through the K10 generation. Now it has been EOLed.<BR/>- Demonstrating K10 at 3.0 GHz and one AMD rep later even referred back to the demo as evidence of healthy K10 clocks.Axelhttps://www.blogger.com/profile/15126742407361053721noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-49530878872796676252008-04-28T13:11:00.000+00:002008-04-28T13:11:00.000+00:00axel: "1. Asset Light and Asset Smart are the same...<B>axel:</B> <I>"1. Asset Light and Asset Smart are the same thing, but AMD has recently chosen to use the latter in order to avoid the implication that it involves the sale of fab capacity."</I><BR/><BR/>If they really are the same thing, I wonder if the name change was designed to make them appear separate and give AMD more time without drawing even more ire from investors. I recall that one of the pointed questions directed at Hector was that they'd gone a year without much detail on "asset lite", and would they see this process repeated with "asset smart."<BR/><BR/>Hector's reply was along the lines of providing more details "soon", after which he clarified that "soon" meant anywhere from 90 days to 9 months. So it could be a case of not wanting to reveal anything yet, but finding a way to keep investors from becoming even more agitated than they must already be.<BR/><BR/>I think that it's not going to be anything revolutionary, probably something pretty mundane. But Ruiz would rather roll out the details when AMD is in better financial health and the investors have put the pitchforks down. If AMD hasn't made some progress on the financial front by year's end I don't know if he can delay revealing the details behind asset lite/smart any longer.Tonushttps://www.blogger.com/profile/01082528970434639776noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-60639672473963632272008-04-28T07:39:00.000+00:002008-04-28T07:39:00.000+00:00Axel said...As we know from the Barcelona propagan...Axel said...<BR/><I><BR/>As we know from the Barcelona propaganda experience in 2007, AMD are quite willing and able to spin and hype things one way (and even lie e.g. "outperform Clovertown by 40%") and the reality turns out completely different. </I><BR/><BR/>AMD has certainly messed up in a lot of ways, but I don't think the 40% "hype" was a lie. The current SPECweb2005 result shows Barcelona is clock-for-clock 40% faster than Harpertown, and 50% clock-for-clock faster than Clovertown.<BR/><BR/><A HREF="http://www.spec.org/web2005/results/res2007q4/web2005-20071117-00100.txt" REL="nofollow">Clovertown</A><BR/><A HREF="http://www.spec.org/web2005/results/res2008q1/web2005-20080225-00104.txt" REL="nofollow">Harpertown</A><BR/><A HREF="http://www.spec.org/web2005/results/res2008q2/web2005-20080409-00107.txt" REL="nofollow">Barcelona</A><BR/><BR/>My own tests with OpenLDAP showed Barcelona with a 66% lead, clock-for-clock, over Clovertown.<BR/><BR/><A HREF="http://connexitor.com/blog/pivot/entry.php?id=191" REL="nofollow">OpenLDAP Scaling</A><BR/><BR/>It's entirely possible that, averaged over some (unspecified) mix of workloads, the overall advantage was just 40% at the time they published those claims. While there are certainly a lot of workloads where the advantage is less, or nonexistent, there are probably also other workloads where the advantage is even higher than 66%.<BR/><BR/>You can criticize them all you want for the botched launch, and failing to deliver faster clock speeds, but I think there's plenty of evidence that they weren't making these numbers up. Given that (a few) folks are now showing Phenoms overclocking to 3.5GHz, I think it's also reasonable to believe that AMD actually expected to be able to deliver 3.0GHz parts to the market. This is pretty significant; it would probably take a 5GHz Harpertown system to match the performance of a 3GHz Barcelona for my workloads. I shudder to think of the performance-per-watt metrics for such a 5GHz server.hychttps://www.blogger.com/profile/15473250487285924085noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-87024305178210415892008-04-27T23:48:00.000+00:002008-04-27T23:48:00.000+00:00In The Know-I responded to his reply. Not quite wh...In The Know-<BR/><BR/>I responded to his reply. Not quite what he would expect. I was a bit more objective and less tongue in cheek. However, I invite you to read the post as it is a summary on my part conserning both blogs. I centainly hope he doesn't delete this one for the sake of postarity.<BR/><BR/>SPARKSSPARKShttps://www.blogger.com/profile/05535419513995195852noreply@blogger.comtag:blogger.com,1999:blog-2602471396566186819.post-22964588628037994972008-04-27T23:18:00.000+00:002008-04-27T23:18:00.000+00:00Sparks, Scientia deleted your post, but he did res...Sparks, Scientia deleted your post, but he did respond to it. That was the surefire approach that got me to stop posting there, but I'm sure he doesn't miss me. <BR/><BR/>But just for kicks, let's go through his points and see if they apply to what I posted. <BR/><BR/><I>Roborat's blog is a great home for trolls which he fosters with his reliance on anonymous posts. </I><BR/><BR/>I have never posted anonymously here, or on his blog. So I must not be one of the trolls. <BR/><BR/><I>1.) Scientia must be wrong because no one knows what Asset Smart means.</I><BR/><BR/>I'll plead partially guilty to this one. Someone at AMD knows what asset smart is, but despite his assertions to the contrary, I don't think he knows what it is any better than I did. <BR/><BR/>The document he refers to doesn't convince me since it clearly refers to <B>Manufacturing SMART</B>, not <B> ASSET SMART</B>. Last I checked, assets were something I owned while manufacturing was something I do. They don't seem quite the same to me. <BR/><BR/><I>2.) Scientia must be wrong because he doesn't work in chip fabrication.</I><BR/><BR/>You wont find his SPC--> APC--> APM --> LEAN --> SMART progression anywhere in the documentation he linked to. This is his own creation and has nothing to do with what industry he works in. <BR/><BR/>So what I actually said is that Scientia is wrong because he doesn't know what SPC is or does. Lots of people outside of the semiconductor industry know and use SPC and have for decades. So in this case he is wrong because he is making stuff up. Working in the industry is not a necessary precondition to understanding SPC. <BR/><BR/>3.) These ideas won't work because they are not unique to AMD or because they don't apply to every company.<BR/><BR/>Nope, I never said that. <BR/><BR/>4.) The numbers are useless because they do not include industry averages.<BR/><BR/>I didn't say that either. <BR/><BR/>5.) Switching to smaller batch or single wafer tooling isn't free.<BR/><BR/>No, it is not free, but if he thinks that was my point, he needs to re-read what I said. I said changing batch size will decrease the number of wafers you get out with the same amount of tooling. So you either need more tooling, or you need to change the process to get around that detail. <BR/><BR/>For the record, I have gone through the documents that AMD provided. Nowhere do I see a direct contradiction to anything I posted. <BR/><BR/>AMD has proposed to get around the lot size issue by getting the equipment manufacturers to make tools designed to run batches of smaller lots (i.e. mini-batch tools). That is a process change in my book, so I stand by my assertion. <BR/><BR/>There are also simpler solutions to this problem. To look at AMD's presentation, you would think that lots are fixed blocks of wafers that can never be broken down. That isn't true at all. Lots are split and reassembled all the time. This can be done to help move things through batched processes. <BR/><BR/><I>Again, these are AMD's documents and conclusions, not mine. Anyone who can actually read can look through the ISMI presentation and see what AMD's conclusions are. AMD concludes that shorter cycle time will save money. I would have no way of verifying this.</I><BR/><BR/>I do believe I said that reduced cycle time was good and provided documentation showing exactly <I>why</I>. I have no trouble verifying that at all. I also said that if AMD was just waking up to this fact, they are way behind the curve. <BR/><BR/>This is intuitively obvious, by the way and I'm sure AMD does understand this and has been working on this all along. The presentation Scientia is linking to is just proposing changes that AMD would like to see the industry and equipment manufacturers adopt.<BR/><BR/>In conclusion, Scientia's 5 reasons posts on this site would be irrelevant and/or nonsense don't seem to apply to my post. He has yet to address a single issue that I have brought up. I doubt he can successful address them.InTheKnowhttps://www.blogger.com/profile/16869163385384973596noreply@blogger.com