Sign in with
Sign up | Sign in
Your question

Will HKMG make AMD superior?

Last response: in CPUs
Share
November 21, 2008 3:54:59 AM

Without HKMG we see this
http://techreport.com/discussions.x/15927
http://www.pcper.com/comments.php?nid=6455
http://forums.overclockers.com.au/showpost.php?p=968348...
http://i4memory.com/f56/amd-austin-tech-day-information...
http://www.legitreviews.com/article/836/1/
And last, and probably least ( Poor FUaD) http://www.fudzilla.com/index.php?option=com_content&ta...

Now, all this is being done without HKMG. As its been previously hyped, adding this will add another 400Mhz to its speeds while reducing power consumption tremendously, as has been shown by various members here. Ill let them show us all again, if they so wish. This could really turn things around for AMD!!!!!

More about : hkmg make amd superior

November 21, 2008 4:00:18 AM

What Im wondering is, since Intel saw a huge increase in power/temp and even clockspeeds by using it, a much higher clocked Phenom would simply outdo any IPC leads Intel has, wouldnt it?
November 21, 2008 4:09:51 AM

So it has high clocks... who cares! what about benchmarks? How well will it preform in x264? Core i7 blows everything away at the moment.
Related resources
November 21, 2008 4:15:04 AM

My question is, as we can generally see a 400Mhz jump going from conroe to penryn in clocks and coees, can we expect that also when AMD brings this? Or, was it due entirely from node shrink? If it was HKMG, then a 4.4 Phenom/Barcy would beat a 3.2 i7, and thats what we know, as to Deneb, one can assume itll only be better in IPC or at least as good
November 21, 2008 4:21:12 AM

if a 4.4ghz phenom can out preform a 3.2ghz i7, thats great...

Not very impressive though :|
November 21, 2008 4:23:16 AM

To be more precise, even if theres no IPC improvements with Deneb, but with the addition of HKMG, will we see 3.8 stock Denebs competing with a similarly priced 3 Ghz i7? Then IPC wont matter
November 21, 2008 4:29:05 AM

Now, that was a good title, jaydeen!
November 21, 2008 4:32:45 AM

JAYDEEJOHN said:
To be more precise, even if theres no IPC improvements with Deneb, but with the addition of HKMG, will we see 3.8 stock Denebs competing with a similarly priced 3 Ghz i7? Then IPC wont matter


What are we going back to 2001 again?
November 21, 2008 4:37:18 AM

Why not? Speed kills right? Really, who knows, but theres potential coming from AMD we havnt seen in awhile, and theyre getting closer, and HKMG may close the gap completely
November 21, 2008 4:47:43 AM

already seen it. Not talking server and BW by the way. DT only here, since obviously thats where AMD aimed this time around
November 21, 2008 4:48:45 AM

From the techradar article:
Quote:
In other words, as good as AMD's new Shanghai chip is, it appears that Nehalem EP will slap it around with a wet fish.
November 21, 2008 4:51:08 AM

^^ read above. Intels aimed at server this round, and by the looks of the ocing and clocks, DT for AMD
November 21, 2008 4:56:23 AM

What I'm trying to get at is just adding new material to the process isn't going to increase performance all that much. When Intel did it, they also tweaked a little here and there, shortened data pipelines and tweaked the instruction set and so forth. Now look at the performance for Nehalem which is basically more tweaks and additions to the processor other then adding the IMC to the CPU. Just because you use different materials doesn't mean your going to improve results.
November 21, 2008 4:59:39 AM

It still can not compete in x264 and other rendering apps
November 21, 2008 5:00:56 AM

The server part isnt even out yet, and wont be for how long? Denebs right around the ciorner, and its where and what Im talking about
November 21, 2008 5:03:09 AM

Everyone says it cant compete, yet we dont have the numbers, only the clocks and the oc potential. Adding HKMG is only going to help this, and whats been leaked about power, itll help make decisions in both DT and server, regardless of numbers
November 21, 2008 5:06:34 AM

Of course, then again, I was a one of few that dared to possibly think a 4Ghz Phenom may be possible too. Then again, add the costs, lots of considerations, not just buy this, it has bigger numbers. Performance is king, and maybe i7 will beat Shanghai, I dont doubt it will. But in these times, at what price? And what continuing costs?
a c 126 à CPUs
a b À AMD
November 21, 2008 5:19:56 AM

I think you are getting too excited. Changing the process will not increase the IPC. tweaking it, such as Deneb, will.

It may allow for higher clocks. But TBH its possible the low clocks of K10 are because of its arch, not the process. Or it could be both. Who knows.

Problem is you are getting too excited about it without anything to go on. Wait until they start it up and have at least something to go on so we can make a proper judgement.
November 21, 2008 5:35:03 AM

Good answer jimmy, and ty
November 21, 2008 5:40:06 AM

When HKMG was first introduced, I wish wed had such thoughts
November 21, 2008 5:43:16 AM

As I've said before, I'll wait and see. This could be a return to something like the Athlon64 to P4 days. Or it could be a flop like the Phenom I with the TLB issues (amongst other issues). However I said it once, and I'll say it again I WANT REAL WORLD, not synthetic benchmarks! I could care less about 1-2% more FPS in a 30sec stretch of game set on a loop. What about encoding, or Decoding? Floating Point? And most importantly WHAT TOOK SO LONG!
November 21, 2008 5:51:01 AM

LOL to the last part
a c 126 à CPUs
a b À AMD
November 21, 2008 6:06:37 AM

JAYDEEJOHN said:
When HKMG was first introduced, I wish wed had such thoughts


I'm not sure about anyone else here but I already knew what it was going to do. I read a lot about it because it is a big change. The transistor has been the same for the longest time, this shook it up.

I knew it would allow for 20% better switching performance (not overall performance that could have been less or mroe) while also using 20% less power.

The clocks would have been hard to descern from that but they turned out pretty on the spot, about 20% higher than a equivalent 65nm no HK/MG CPU in OCing and power consumption.
November 21, 2008 6:11:41 AM

20% is really nice. And thats 20% in power as well?
November 21, 2008 6:14:12 AM

Im trying to somewhat figure this all out for gpus as well as cpus. TSMC will eventually use it, at 28nm, which is about 35% of current space used at 65nm, minus the power consumption, meaning somewhat higher clocks, at those nodes, who knows?
November 21, 2008 6:17:26 AM

Itd be like a tri sli G280 consuming 100 watts in mid 2010 for a single card, and thats rounding up, with conservative guesstimations included. As for AMD, maybe a 10-15% clock increase? With maybe 15-20 watts lower TDP, or actual power usage?
a c 126 à CPUs
a b À AMD
November 21, 2008 6:29:29 AM

Not sure with AMD. I ahven't seen much from IBM or what materials they plan to use, if its Hafnium it may be just as good as Intels. That and the quality of it may not be as good since they are rushing the design of HK/MG out while Intel has been working on this for quite a long time.

And yes its 20% less power draw. In fact if you look at a QZ9650 when its idle the CPU itself only uses less than 10w of power compared to a equivalent QX6850 that uses more closely to 20w of power idle.

Maybe for AMD we will see that. but we have to see what IBM gives us on the info of it. Intel released all this info back in about June of 07. So hopefully IBM will have the info.

Then again I still hear rumor that AMD is going to wait till 32nm to release HK/MG.
November 21, 2008 6:45:54 AM

Not sure. Ive heard both. Later on 45, and at 32. Who knows?
November 21, 2008 6:59:18 AM

It doesn't matter. Intel will still slap them around like the second banana they are.
November 21, 2008 7:02:17 AM

With what?
November 21, 2008 7:07:39 AM

Unfortunately, we have been misinformed by those whore misinformed, and told us this could never happen in the first place. Now that its happened, whats the next tact? Insults, Acts of supposed superiority? This wasnt sposed to happen, remember? This couldnt happen at all. I remember, and I remember who was making those mistatements, are may still be doing so. Why cant you just let it go? I guess some people arent happy that their next Intel cpu will most likely be cheaper because of things like this.
November 21, 2008 7:12:00 AM

Do you think Intel has just been standing still with process tweaks? HK/MG is trivial to Intel at this point. The next process revision will bring with it one of the greatest breakthroughs in fabrication history.

November 21, 2008 7:18:26 AM

You mean? You cant possibly mean THAT? NO, not THAT. Ohhhh noeeeesssss, just like we NEED HKMG on 45nm, not THAT too? All is lost.... cmon, get off it
November 21, 2008 7:32:13 AM

You really should take some reading comprehension classes.
November 21, 2008 7:41:23 AM

Didnt I mention a certain degree of superiority? Youre trying to show it now. Id suggest you read my links first. See whats changed here. And then, somehow be happy about it, as the vast majority of poeple hearing and seeing this is. Cause its here, may as well enjoy it
November 21, 2008 8:42:57 AM

A higher clock speed would ruin the power saving advantage of that process to a certain degree. Don't forget that Performance per Watt is what is keeping AMD alive and clock speeds can only be improved so far before the additional power consumption takes its toll. On the desktop market that could make a slight difference and help AMD come back to the mid-range market (at least in the 4 core arena), but i highly doubt that.
AMD is in a bad position. They can't really make too many changes to improve IPC because that might require a new socket that is incompatible with the older ones which in turn would lead to AMD loosing their upgrade customers. If they concentrate on backwards compatibility they can change only so much and it is a little more difficult - add that to the fact of their limited budget and you can clearly see where things are going.
I think they will stick to the server segment and use the new process to improve the power efficiency and only slightly increase the clock speed.

The topic title reminded me of Thunderman. Weird.
November 21, 2008 8:53:57 AM

So youre thoughts are more tailored to lower power than higher clocks. have you read this? http://www.eetimes.eu/semi/showArticle.jhtml;jsessionid=0JFPMZHPOV0J2QSNDLRSKH0CJUNN2JVN?articleID=212002481&printable=true

And say, after the HKMG implementation, youre still saying theyd stay at present clocks, and just reap the power savings?

Ummm this post is a response to my other previous post, where it would seem eithe HKMG is a great thing, or its dog meat, it just depends on whos using it I guess heheh
November 21, 2008 12:58:39 PM

Will HKMG make AMD superior to what? Everyone knows your chips beat the competition's 2-year-old offering, which is good if you just want to sell everything at bargain-basement. I'm asking what you think will be the relative timing of 45nm-HKMG for AMD and 32nm (HKMG) for Intel, especially as AMD just tweaked 45nm heavily for non-HKMG. Everyone seems to forget timing. It's the basis of premiums that let you secure your next round of R&D.

I, too, saw a lot of naysaying about 45nm for AMD, but the naysayers probably did not see the implications of not making any changes to the core. Holding the architecture very steady means that you can put maximum effort into improving your process. Combined with the general opinion that 65nm was AMD's flop, the substantial gains seen at 45nm should be expected, and are necessary to stay in the running. That 15-20% overall performance improvement (7-10% cache, 7-10% clocks) some were predicting over Barcelona would have seen them exiting the race.

You mentioned that i7 was aimed at servers, not desktops. That is only by circumstance. By catching up to AMD on the uncore (IMC, QPI, caching), the whole socket had to be changed, which makes recent upgrades expensive. Were the rest of the platform cost excluded - e.g., for long-term upgrades - the i7 would be extremely competitive on DT. It is visibly faster than Penryn or K10, more power efficient, and safer to overclock. As a downside, it is much more complex and appears to overclock a bit lower than the Penryns, but not by enough to drag overall performance down.
November 21, 2008 2:25:58 PM

OK folks, heres what Ive found. Not sure about HKMG's use at 45nm, but heres what is now old news with HKMG :
27.7 Implementation and Optimization of Asymmetric Transistors in Advanced SOI CMOS Technologies for High Performance Microprocessors, J. Hoentschel, A. Wei, M. Wiatr, A. Gehring, T. Scheiper, R. Mulfinger, T. Feudel, T. Lingner, A. Poock, S. Muehle, C. Krueger, T. Hermann*, W. Klix*, R. Stenzel*, R. Stephan, P. Huebler, T. Kammler, P. Shi, M. Raab, D. Greenlaw, M. Horstmann, AMD Fab 36 LLC & Co. KG, *University of Applied Science Dresden

Sub-40nm Lgate asymmetric halo and source/drain extension transistors have been integrated into leading-edge 65nm and 45nm PD-SOI CMOS technologies. The asymmetric NMOS and PMOS saturation drive currents improve up to 12% and 10%, respectively, resulting in performance of NIDSAT=1354μA/µm and PIDSAT=857μA/μm. Product-level implementation show a speed benefit of 12%.

From here http://www.his.com/~iedm/program/sessions/s27.html

Now these numbers are basically the same as Intels, and this is on SOI, so its assumed close to same performance benefits, as the drive current will come up quite a bit using HKMG
November 21, 2008 2:30:33 PM

So, and someone who REALLY knows correct me here, that this may or may not directly translate over to the chip of a healthy 12% gain at same power usage in speeds, or if not total gain, a nice portion of those speeds without more power.

So, in essense, we may see a 3.4Ghz Deneb using HKMG at 125 TDP
November 21, 2008 5:43:12 PM

Thanks for providing this information, jaydeejohn.

Hmm..ramping up clocks. Historically AMD has gone for smarter not faster, designs so I hope that eventually the focus will be on the core logic and not just how fast it runs. Remember the boasting that went on, when people were able to crank up the clock speed of their Pentium 4's to seemingly ridiculous speeds? Remember what happened to those systems the moment Core 2 Duo came out..they weren't necessarily redundant, but their position as king of the MHz castle was practically worthless in terms of real-world performance.

I am pretty confident that AMD is laying the foundation on which more powerful products are based, it's just a matter of time. As mentioned, let's take into account the fact that AMD has nowhere near the resources of Intel, and has had to bear the brunt of the recent global economic crisis. Yet AMD still manages to produce some excellent products, including GPU's that can run with the best of the pack, as well as upgradeable server and super-computing hardware offering phenomenal bandwidth using tried and tested hardware protocols such as Hyper-Transport Technology..a technology that is based on an open architecture.

And of course, without AMD you would most likely not see Core 2 Duo..so competition is good news for the consumer and those who benefit from generally faster speeds and lower costs :) 
November 21, 2008 7:51:34 PM

And possibly this round, their ocing will win out as well. So far, its looking like it, as from what Im hearing a i7 did 5.7, but didnt actually get into windows, while a Deneb hit 6.3 . And this is all on pre release chips, and the AM3s are said to go a lil faster yet, then of course, theres the matter of HKMG, and what thatll bring. I understand the physical limitations, but others claimed they knew them as well, and without HKMG, at these lower nodes, clocks like these would be unattainable.

What I find most compelling here is, AMDs channels are already stouter, if you will, than Intels, adding gate first HKMG is going to create some very nice power savings while adding to overall speeds
November 21, 2008 9:08:22 PM

jaydee do you work for AMD?
November 21, 2008 9:20:50 PM

Huh? No, just exciting times ahead, when all I thought was coming was i7, and its so so performance in gaming and from what it looks like now, oceeing may not be as good as we thought it was, or Denebs looking alot better for it
November 21, 2008 9:39:35 PM

speculation my friend. i hope its allll true. i truely do but at least wait for some credible sources. your setting yourself up for disappointment if its just slightly better than what they already offer.
i7 not overclocking very good? how do you figure that? the i920 already in peoples hands and overclocked from 2.66 to 4Ghz on air. i would call that pretty good but you expected more?
November 21, 2008 9:59:26 PM

JAYDEEJOHN said:
Huh? No, just exciting times ahead, when all I thought was coming was i7, and its so so performance in gaming and from what it looks like now, oceeing may not be as good as we thought it was, or Denebs looking alot better for it


jdj, but like core i7 is already out, and the performance for gaming, well I'm, not 100% how much better it is then the yorkfield's at comparative clockspeeds are [i'd assume not much, but I haven't really paid attention]. But the I7 already look pretty solid and if you don't mind spending ~$200 for some DDR3 6GB kits[ about ~$110 for DDR3 3GB kit's], or even the $220 for the MSi platinum x58 motherboard [up to $400 for the most "extreme"], I don't really see any point in which AMD can catch up. I'm not saying it can't happen, b/c I would like nothing better than to build an AMD high powered rig, but it just doesn't seem likely. :pfff: 

As much as Thunderman seems crazy to us all, I think his motives are understandable... :pt1cable: 

AMD should be about innovation and not stagnating the market w/ slightly higher clocked phenoms, like Intel was doing w/ the old net burst's. God where is the old AMD that was every enthusiasts friend, w/ their beautiful K8's... I really do miss the old AMD. :( 

But anyways if I were to build a new system today that core i7 920 w/ the asus P6T x58 seems like a real nice combination. :whistle: 
November 21, 2008 10:04:14 PM

i have the itch to build a system but want to wait until Deneb is out and see some side by sides with the i7. right now i cant help but be skeptical because almost all the news on Deneb is coming from AMD then sprinkled down through 2nd parties without them touching them but giving lip service. if the claims are half truths they got something good coming but i will gladly wait to reserve judgment.
November 21, 2008 10:34:55 PM

Adding IPC and having a 30+% oc isnt stagnating anything. Its early I know, and I was somewhat dissappointed by i7, and thrilled by Deneb. Ive already explained why before, but Ill do it again. The improvements in DT from last gen to this gen, in Deneb we see huge improvements, while in i7, its meh, so so, and thats coming from a gamers POV

See? Do you see what you write? I was looking for more from i7 as a gaming cpu, and was underwhelmed. Now, after all this info, and theres some here that dont believe it, well, you need to read more. I think if it were i7 you would have by now, and know how it will go, but this attitude, if this doesnt fire up a oceer, or a gamer, or someone who just wants competition back, then itll hit you eventually. This chip is real, it does as everyones said. It just may be a better overclocker than i7. From initial stock clocks, Im pretty sure the difference will be more. Im not making the chip better by posting what Ive posted, as itll do whats been shown, so no, I dont think Ill be disappointed, as alot of oceers want this chip, and want it bad
!