Fuddo: AMD 45nm K10.5 scheduled to launch in 1H 2009

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar


Agreed on all points. Everyone is optimistic about AMDs 45nm but I have too many doubts about SOI as it will be spread too thin thus they will have to clock it lower to help stop too much leakage. DDR3 will be best for Nehalem and probably be around the DDR3 1800-2000 by then if not faster.



Yes and no. I don't see Intel delaying Nehalem as they have it in testing. Nehalem will be their jewel in the crown. Not that Core2 wasn't but this will put Intel on par in every segment so AMD can't brag about the "Incredible" performance. It will be a level playing feild all around and will give AMD something to worry about. So I see servers about late Q2/early Q3 2008 and the EE Nehalem Q4(probably November) and then the desktop side late Q1 2009. But I am kinda crazy.




That is to be seen. The fact is AMD has yet to show any samples and it is to be released what Q3 08? They should bhave in lab test showing now and in mid to late Q2 08 ES samples shipped out. But then again they will probably host another "in house only" event somewhere showing the best of the best chips off and get ES samples out a week after release. Then we will find that the ES and retail versions don't even do what it should.

But thats a bad scenario. We could say AMD has working 45nm Phenoms and is ready to roll out. But thats just giving them too much credit after the entire fiasco in 2007. Just makes you not believe in the company.
 
Kind of old news.
http://news.softpedia.com/news/AMD-039-s-Tri-Cores-Tolimans-and-Regors-Ready-Until-2009-72743.shtml
This is from last year, and says almost exactly the same thing.
I must have dreamed that AMD demoed 45nm parts at Cebit, except http://www.theinquirer.net/gb/inquirer/news/2008/03/05/amd-shows-45nm, or if you dont believe them http://www.engadget.com/2008/03/04/amd-demonstrates-45nm-quad-core-processors-at-cebit/
I have also heard that samples have been shipped.
Jimmy what does
Agreed on all points. Everyone is optimistic about AMDs 45nm but I have too many doubts about SOI as it will be spread too thin thus they will have to clock it lower to help stop too much leakage
Click to expand...
mean? Perhaps you should let Soitek in on it, since they supply AMD with the soi wafers?
 


Why is everyone accusing AMD of lying? Intel's done quite a bit of that in the past and I never saw you guys jump on their case, nor on Nvidia's fudging of Crysis demos etc.

This article needs to be seen in light of the fact that AMD has announced the first 45nm Denebs and Shanghais will be SOI and the second generation of 45nm Denebs and Shanghais will have a new HK process currently being developed by IBM. That is what I believe is meant by 10.5 here. Same core, same C stepping as SOI 45nm, but new process that will allow for clocks higher than 3.0 or 3.2 (the expected high end for SOI Denebs).

Besides, the 45nm triple cores will arrive after the quad cores, just like with 65nm. Whether AMD will start to disable perfectly good cores for their newly created market, or whether they have one defective core that has to be binned as triple (which seems to be the case with the 65nm) remains to be seen.



Jimmy, I'd give SOI the benefit of the doubt. Just because Intel didn't find a workaround doesn't mean it's the end of the world for AMD. I trust IBM engineering on this. They may have realized that SOI won't go on forever, but they seem to be confident it will work for AMD with lower clocked Phenom's (i.e. lower than HK/MK but still quite higher than 65nm).

Wouldn't it be a trip if more pipelines worked for Phenom without degrading performance? Like Intel was on to something in the absolutely worst generation where Northwood pipelines were long enough to begin with. I'm not sure what the Phenom pipelines are, but if they're like the X2's 14 or 16 (is my memory serving me there?), then there's room for going higher to get better clocks without running into a Prescott like brick wall.
 
look the Phenom would be on par with the Q6600 if they decied to have the North Bridge set to 2.2-2.4GHz instead of 1.8GHz. The north bridge is how fast the L3 cache runs. Even if the Phenom is at 3GHz the L3 cache will only run at a blazing 1.8GHz instead of using the 3GHz CPU clock speed.
 



What is that analysis based upon???
Almost all of the articles are based upon Engineering Samples running @2.0 not 1.8ghz and the Phenom loses by about 20%.

So increasing ONLY L3 Cache by 10% is going to improve total performance by 20%???
Even Increasing the CPU by 20% will not yield a 20% gain due to scaling.
 


It's based on first hand experience with my own Phenom 9600BE. With the IMC/NB running at 1.8ghz it limits the L3 cache to 1.8ghz, as well as the HT multi is locked to at or below the NB/IMC multi. What happens with the IMC/NB being locked at 1.8 while the processor is running at a higher clock rate, is that the L1, L2 and cores are forced to wait for data to be sent from the much slower L3 cache. As far as I can tell, it almost equates out to an average speed between the IMC and CPU cores. For example 3ghz/1.8ghz= ~2.4ghz performance or around that area, may be a bit higher in some things that don't hit cache a lot. If I can convince MSI to put out the bios with the things that I and many others want in it, I'll be able to resume testing.

Another interesting this is, IMC/NB speed effects how the core speeds performance scales, and vice versa. Besides, according to the ES tests, 2.4ghz by 2.0ghz Phenom only lags behind a Q6600 by about 9% not 20%. It does effect performance in gaming, otherwise my system wouldn't be scoring higher than THGs Phenom systems in the last phenom reviews.
 

Boy was that a small enough sample to compile an 'average better than Phenom' statement.
If I just pick out memory bandwidth tests, does that mean a Phenom is x% faster than a Q6600?
If you're going to quote benchmarks, at least link to a site with a variety on and post the average by all means...
I'm not bashing, just pointing out that was hardly a useful response... :)
 
http://www.tomshardware.com/2008/02/05/amd_phenom_9600_black_edition/page13.html

tabelle_001.png


Stock Phenom 9600 is about 13.5% slower than a Q6600 across a 'wide variety of benchmarks'. 😉
 
The 9750 should be out in May. I'm just not sure if I'll get it or wait for 45nm Deneb. Yes, I know that Deneb will compete against Penryn, if not Nehalem, but it should improve things quite a bit on the AMD side.

What I like about sticking with AMD right now is that a high quality AM2+ board will still have 45nm Phenom capability. So I could go B3 9750 in May on a Crossfire AM2+ board, and upgrade to Deneb in December, moving the B3 to a 780G board for a second PC upgrade.

How many boards that support Q6600 will also support Nehalem this time next year? How many that will support Penryn will? Many Intel enthusiasts here are used to switching out boards for their Intel builds every time they upgrade their CPU, while at the same time kvetching about the motherboard companies lack of support for 690 upgrades to Phenom, but when all is said and done; a 780G board and a B3 is the perfect budget upgrade to quad core this spring because it will be supported when Deneb arrives.
 


I agree Luke. But the last one was purely gaming and from some other benchies I have seen Phenoms tend to hold the system back a bit. Maybe with a higher NB but thats probably still luck of the draw as some may be able to clock it higher.

As for edyen(or whatever) Intel admitted that SOI looked promising when AMD first started using it and was going down that path. But you are telling me that Intel, with probably one of the largest R&D's of all the technology companies, couldn't find a way to get it to work and probably did everything and they couldn't but AMD with a limited R&D could?

Sorry but that seems unlikely. I still am holding my judgement on SOI @ 45nm until I see it. I have a bad feeling only due to the fact that AMD is trying to get a HK/MG even for their 45nm. Doesn't that strike you as strange? That they will release 45nm on SOI but soon after will have 45nm HK/MG? There is a reason and its probably cuz 45nm SOI is not going to be that great.
 


No, I'm not saying that AMD with a limited R&D could. I'm saying IBM can. That's the difference. I'd put IBM's R&D capability up against Intel any day of the week. Besides, I'm hoping that a Prescott style bump in pipelines will work for Phenom. What is the Phenom pipeline? 12 or 16? Don't you think it has the potential to be bumped up at 45nm without loss of performance that hampered Prescott in it's 90nm generation?

Still, it's not perfect, which is why I might wait for a Deneb with HK/MG instead of SOI. I don't think AMD needs to beat Nehalem in the enthusiast market, just provide an upgrade path from 65nm to 45nm that actually boosts performance across the board. Even a 780G motherboard today should allow that, as there will be 45nm quad and triple cores out (i.e. AM3's a plus but not necessary). This also shouldn't rely upon the motherboard manufacturer's who flubbed support for the 690G and V chipsets.



If 45nm SOI 3.0 gives a boost in thermals and performance over B3, and if it does well holiday season in OEM PC's then that's probably why AMD is going that route. They were probably locked into SOI for the first Denebs, but will go HK/MG for higher clocked Denebs plus Swift.

Okay, I'm arguing the same line that Intel fans use when they say that Intel had to come out with Prescott and Smithfield, then improve them slightly with Cedar Mill and Pressler, because they didn't have a new process and architecture. They still sold in OEM PC's (OEM rebate strong arm tactics aside). IMHO, Deneb SOI won't match Nehalem, but won't be bad for the market it's aimed at. It should be a decent improvement over 65nm B3's.

I still like the fact that I can get a 770 or 790 AM2+ board with a B3 Phenom 9750 in May and then upgrade to Deneb later; or I could just go 780G if I don't want to give CrossfireX a try next year. Unlike the 690G motherboard fiasco, the AM2+ boards should support Deneb. As is, I'm getting a Phenom 8xxx series B3 triple core for our ASUS 690G board anyways, but I doubt that would support a Deneb down the line.
 


I'll admit I've done my fair amount of Phenom bashing lately, mainly because I'm an overclocker and I'm frustrated that the best QC from AMD can't beat Intel's 12 month old budget offering, but at stock speeds it's really not that bad a CPU I guess, considering it's a bit cheaper than the Q6600 as well.

Things get nasty when you start comparing a Q6600 @ 3.6GHz to a Phenom 9600 BE @ 2.7GHz though... it barely matches a stock Q6600 at that speed. That's my main gripe with current Phenoms. :kaola:
 


Actually I provided the Link.
Also, The person I was responding to SPECIFICALLY said games showed little difference.
Hence the focus on games.

 


Going by those numbers from the Xbit review, its nice to know that my Phenom Rig with an HD3870 can blast the hell out of the rig they set up there.
 


Jimmysmitty,
I wonder if Intel set their sites on HK, rather than SOI, knowing that they would have a greater performing part? They let AMD have one generation of lead when AMD had SOI, only to come out with HK in Core 2, and take the lead back. While AMD has a much more limited buget for R&D, their relationship with IBM, and IBM's advanced process and tech, will certainly benefit AMD. I think that we'll see Intel and AMD chips competative again on the high end, with similar tech within a year or so.
 

Very true, but I have to admit I didn't read the article fully.
Having just had a click through it, it was a fairly appalling amount of 'benchmarks'...
The Tom's table had a far better spread on it, I thought. :)
 

Yeah I know exactly what you mean :)
I don't OC and am quite happy to buy AMD so a Phenom would be quite nice.
I'm in no hurry to swap out my 6000+ though, my system rocks all current (including the holy grail, Crysis! :lol: ) so I wouldn't really see any benefit in games. In CS3 yes, but I'm more than patient enough to wait for a higher-clocked B3.
I like the sound of a 2.6Ghz or higher Phenom as an upgrade! :)

And yeah jimmymsmitty, I think you're bang on with the NB. The OCers on here who have clocked the Phenom say performance is drstically improved in bumping the NB/IMC. I guess at clock speed it would be damned speedy! :)
 


Unless you know the exact level, game settings and/or timedemos used by Xbitlabs in their testing, I don't see how you can make that call. You may get higher framerates, but are you testing in exactly the same scenarios as the review? I highly doubt it.

I think it's commonly accepted that the Q6600 is faster clock for clock, so those results don't come across as surprising to me. Yes, we all know you can increase the NB speed on a Phenom, you've said it enough times already 😉, but that is not the default configuration, and is thus considered a performance tweak - a neat one I'll admit, but one can also tweak a Q6600 by running it at 6 x 400 instead of 9 x 266, and the increase in FSB and memory speeds will also increase performance in a similar fashion without even overclocking the CPU itself.
 


Well yeah, but I'm also talking about my OC specs compared to their OC specs, at the same core clock speed, on the same kind of Processor. CPU score in 3dmark06, at 2.6 by 2.4 I'm getting 3804 after a 3 run average, they get 3593 at the same core clock. And my total score is pretty close to theirs with a 3870 instead of an 8800gtx. Thats just one example.

I have Orange box so I could probably run those benchies if I get ahold of the same save game or time demo, granted it wouldn't help much since they're using a faster performing card. Getting ready to buy UT3 the next time I get paid. As far as most of their encoding benchies go, those are effected by memory latency, which is effected by IMC speed. You can tell the effect that the IMC/L3 speed has on games by looking at the results with the TLB fix enabled.

The lowered IMC/NB on the retail processors was because AMD had a kneejerk reaction to TLB 254/298 and over throttled it. It was originally supposed to run at or within 200Mhz of the cores.

And as far as bringing that up a lot, I'm sorry about that. But every time I see someone link to one of those Phenom OCing threads, articles that doesn't take into account the full abilities of the processor it gets me going. Especially when said things have the ability to impact overall performance. Everyone just wants to mess with the core multi, which doesn't have the same effect on the k10 as it does on the Athlon64/X2. Athlon 64 based the IMC runs at core speed, which is also what causes the uneven divider problem with memory, easily done since Athlons have no L3 cache. With the Phenom, L3 speed is directly linked to NB/IMC speed, and the faster the L3 runs, the less the rest of the processor has to wait on it. The IMC thing is also why it makes it notoriously difficult to OC an Athlon X2 with any headroom. Core clocks could probably fly on an athlon x2 if it weren't for that. I've noticed that very few of these review sites understand k10 much at all.
 

No, what Intel said was that soi did little for them.
Soi is only good if your problem stems from capacitance on the handle side of the wafer. Intel uses a lower v-core, so that capacitance is much lower.
Doesn't that strike you as strange? That they will release 45nm on SOI but soon after will have 45nm HK/MG?
Click to expand...
First off, you seem to think that HK/MG will be without soi. I think not. Soitek has only just begun a new wafer manufacturing plant, to make soi wafers for AMD. I doubt they would do that if AMD is going to stop using soi wafers before that plant is even finished. On the other hand, it does seem strange that they will try to do 45nm without NiSi. I'm guessing this is the advantage of immersion lithography. The HK is not that big a thing as they will be using a median high K oxide, by icreasing the nitride level in the gate oxide.
 


I'm tired of AMD promises.
When they say "H2 2008" we should expect "the last few weeks of December 2008 with token quantity and maybe defective parts". They're never honest.
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom