backtop

Print E-mail del.icio.us 53 comment(s) - last by Smiting Eye.. on Apr 10 at 8:34 PM
AMD's newest roadmap reveals many more Phenom processors for 2008, but tops off at 2.6 GHz

AMD's latest corporate guidance reveals a flurry of changes to the company's line of desktop processors. Most importantly, the new roadmaps indicate the arrival of AMD's higher-end Phenom processors.

AMD's first update to its performance segment will take place later in Q2 2008; the company will unveil two new Phenom processors: the AMD Phenom 9750 and the Phenom 9850. The Phenom 9850 features a 125 Watt thermal envelope and runs at 2.5 GHz, while the Phenom 9750 runs at 2.4 GHz and features a 125 Watt TDP. Around this time another configuration of the Phenom 9750 will be released, but this processor will feature a lower 95W TDP. 

In Q3 2008 AMD will release its AMD Phenom 9950 clocked at 2.6 GHz.  The Phenom 9950 is one of AMD's last high-end 65nm chips and features 140W TDP, the highest of any Phenom processor. Roadmaps indicate that there will be another processor released after the AMD Phenom 9950 though this is likely a lower TDP version of the same frequency. 

Next month AMD plans to launch two new chips for its upper mainstream market, the AMD Phenom 9550 and Phenom 9650. AMD guidance says the 9650 runs at 2.3GHz, whereas the 9550 runs at a slightly lower 2.2 GHz.

Halfway through Q2 2008 the company will release three more Toliman triple-core processors for its lower-mainstream segment: the AMD Phenom 8450, Phenom 8650 and Phenom 8750. They feature an operating frequency of 2.1GHz, 2.3GHz and 2.4GHz, respectively. All three chips feature a 95W TDP.

AMD guidance suggests all of these processors are based on the B3 stepping.  There are no more scheduled revisions to the 65nm architecture. 

AMD finally makes the jump to 45nm halfway into Q4 2008 with its Deneb and Propus-based chips.  Both feature quad-core architectures and 2MB of L2 cache. However, the major difference between the two is that Deneb features an L3 cache pool shared amongst the processor cores, Propus does not. Deneb will be targeted at AMD's performance segment, while Propus will be for AMD's mainstream chips.

Both of these 45nm processors are described as 95W TDP processors. 

As indicated in December, the company will not announce DDR3 support with the initial Deneb and Propus processors.  The initial 45nm Phenom processors will stick with DDR2-1066 until early 2009, where another revision will include DDR3 support. 
 

Comments     Threshold


This article is over a month old, voting and posting comments is disabled
45nm
By Desslok on 3/21/2008 12:44:03 PM , Rating: 2
quote:
AMD finally makes the jump to 45nm halfway into Q4 2008


Things are not looking to bright for AMD, sppeds topping out at 2.6 and 45nm not coming till late 08. I am hoping for the server side of things were the large TLB can really do some good hopefully. We need AMD to stay in the game and keep Intel feet to the fire.




RE: 45nm
By FITCamaro on 3/21/2008 12:52:26 PM , Rating: 3
I'm just surprised they aren't able to get any more speed out of the 65 nm chips.


RE: 45nm
By eye smite on 3/21/2008 12:56:11 PM , Rating: 3
I'm betting it's just going to take time. Like the original athlon x2's socket 939 that topped out at 2.4 ghz. They gor those up to 3.2 ghz and I've seen forums with cpu-z screenshots showing the 5000+ black edition stable at 3.7 ghz. I'm sure they'll mature the product at a somewhat similar and slow rate in Phenom.


RE: 45nm
By Mitch101 on 3/21/2008 2:16:45 PM , Rating: 2
Yea sadly last year AMD roadmap was roadkill.

One would think 3ghz is possible since AMD demoed them and I'm sure they didn't have 45nm samples at the time. The machine did have a lot of fans. Really they shouldn't bother with 65nm any more since 45nm is close.

I would put all engineering on a smooth transition to 45nm above everything else and get it out in time for the back to school sales.


RE: 45nm
By MagnumMan on 3/21/2008 3:03:02 PM , Rating: 2
I've used two Phenoms now and the most I can muster out of either with water cooling is 2.75GHz. You get to a point where upping the voltage on the processor doesn't help. I think there's a major design issue on the current 65nm process. I think going to 45nm now, and not scheduling any more revisions for the 65nm is a big gamble. If they were to focus all their effort on the 45nm parts and get them out early, say in Q3, it would be worth the gamble. There is a lot of time between now and September. I'm still waiting for them to offer rebates to early adopters of their flawed B2 CPUs. I know it will never happen, but we deserve it. Hopefully they do not just shrink the die and call it a day...


RE: 45nm
By ImSpartacus on 3/22/2008 11:13:34 AM , Rating: 1
Yeah, it's just a bad architecture. I hope the 45nm process brings some heavy revisions for overclockers. It's not fair that a last generation Q6600 can overclock well over the high end's stock speeds and current gen Phenoms can't even hit their [figurative] high end speeds.

At least they're catching up on chip process. AMD stayed with the 65nm's for a whole year while Intel has been working magic with the process for some time now.


RE: 45nm
By Darkskypoet on 3/23/2008 1:47:51 AM , Rating: 5
Its not a bad architecture... really, think for a second. Do you know how many bloody transistors go into the 65nm package? The issue is transistor count, not architecture so much. 450m transistors at 65nm, add to that multiple power planes, and an asynchronous L3, and HT setup. Of course its not going to scale as well as a 291m transistor c2d will.

I was looking into this previously, if you take a look at the relative complexity of the two chips, you'll see quite quickly why the Phenom doesn't scale so well. If you want to assign a part of the blame to the arch go for it. But you have to realize that when one chip carries that many extra transistors, its not going to scale as well as others with fewer.

The mistake was going monolithic quad core + IMC , and then adding L3 cache all on 65nm. They would have had a much better chip initially if they would have ditched the L3 cache, run the god damn thing synchronously, and let the IMC do its job. Instead of sitting with an L3 TLB issue, and craptacular scaling, they would have come in with a leaner meaner chip with relatively similar performance (if not better performance at the higher achievable clocks) for about 2/3rds the wafer space. Especially considering Intel has had the cache density lead forever.

It honestly seems like someone had a bit too rosy a forecast for what 65nm was going to be able to achieve. Thinking r600 here...

However, AMD has, since the K8, gone server first --> then desktop... and thus the cache, cache issues, etc. If any chip on the board was made for MCMing, it was the K8, and derivatives. HT anyone? It astounded me then, and astounds me now, that AMD didn't just MCM the K8, and do an optical half node shrink (55nm) at TSMC, or Chartered with the simplified cache K8 design they've been making at 65nm for a while.

Instead, we have k8s topped at around 3.2, and phenoms that can't hit 3. Considering the die size of a k8 at 55nm (or 45nm but they wouldn't have those yet) an MCM variety would be dirt cheap, and a great competitor to the cheaper c2ds. Further, If AM2 compatible, a direct quad core drop in with a bios update.

Sigh. How many of us armchair CEOs would love to have run AMD for the past 2 years or so?

(Note: I am thinking that the multi cpu variety opterons think 8xxx, with the extra HT link would have been able to handle mating quite nicely. This does increase the die size I am sure, but I doubt by that much. Consider a direct extremely local HT-HT link between the two dies as superior to linking through NB.)


RE: 45nm
By vignyan on 3/23/2008 9:58:22 AM , Rating: 2
And all this while I thought it used clean silicon transistors! :P

I did not know it mattered how many transistors went in. The designers well knew their targets before. In that case, you want to call the p4 EE design a marvell with its very high gate count and very high clock and very high power consumption, very high incompetency and hence a very BIG flop! In fact shame on Intel for making such a product in the first place...

My fighter plane flies but not just as fast as the enemy's because its complex... Use my plane will ya? jeez... i got to stop... Sorry pal... Bad design = not meeting targets... The Q6600 is beating phenom in most benchmarks... what else is more shameful than this! i am comparing with a product launched in dec '06 w/ product launched one full year late which by all means is only fair for AMD i guess...

AMD failed us. accept it... but it will come back... i hope so...


RE: 45nm
By Darkskypoet on 3/23/2008 9:11:54 PM , Rating: 4
That's exactly right. They did fail. It's a shame too, because honestly, AMD should've beat intel to the MCM'd Quad core. The 8xxx opteron design, as I mentioned, could've been MCM'd a long time ago. Sadly, It probably would've been able to go head to head with Phenom as well. But still not beated the Q6600.

However, quite honestly, both core 2 duo and k8 / k10 owe quite a lot to the Alpha of yore. Short and wide. The model cpu, is quite a bit different from the model human... lol.

The P4 was stupid because they threw away a good architecture (P3, Pm, etc) to make it. AMD didn't go far enough in utilizing their bag of tricks. If we want to speak of design issues, the K10 should be wider then it is, and the L3 cache (as I said above) should be gone. Hell, a 256bit memory controller, broken into 4 64bit channels, wider design, and no cache. Still a smaller chip then current, and it would have been a screamer. further, as Intel moves to a tri channel memory controller, I figure AMD will have to increase Memory bandwidth when they fuse graphics into the mix.

Yes it does matter how many transistors go in, when manufacturing for a certain process. Not that the number means anything when comparing chips, except that it'll give you an idea how something might scale, and its relative complexity. But as I've said before many times; AMD was stupid to go for a monolithic core when even the mighty intel wasn't going to go for it. Not that no one could do it, but when god and its money don't, there is usually a reason. So perhaps if you only have 2 fabs, you shouldn't be so stupidly ballsy to tie your best fab up making monstrously huge dies.

I would have thought Intel with its what, 14 fabs?, would have had a better shot at making it work, because even if it had issues, it wouldn't have hurt them as badly.

However, one thing perhaps tht my ignorance is hiding from me... does anyone know if the IMC actually makes it much harder to MCM the dies? I was just thinking about that, does it put them in a position to have to use a NUMA-esque memory architecture? Or can the Coherent local HT links do it for them? Maybe someone closer to the scene can explain...


RE: 45nm
By Proteusza on 3/24/2008 8:40:22 AM , Rating: 2
Good analysis.

I do agree that AMD's greatest problem has been its poor management. The K8 gained them a lot of ground and started to narrow the gap between Intel and AMD. But then they didnt really update the architecture, and so were caught with their pants down when Intel released the C2D and even worse, the C2Q.

I did hear about a failed project that cost AMD a few months, but really, thats just poor project management.

I do wonder why AMD didnt MCM the K8's. Given a slight die shrink and a quad core, it would have really competed well with Intels chips. I guess the problem is that if they did that, there would be no reason for anyone to buy a Phenom, since the quad core K8s would offer better performance.

I think Phenom will go down as AMD's netburst architecture. I dont think they did enough to improve IPC, and the monolithic quad core, while good in theory, has done little of them.


RE: 45nm
By Von Matrices on 3/22/2008 2:05:38 AM , Rating: 2
That increase in clock speed also took two years to achieve that 800MHz, and that 3.7 GHz OC was a process shrink after the 2.4 GHz "wall". By 2010 better processors will exist than a 3.4GHz (or 3.9GHz OC'ed) Phenom.


RE: 45nm
By ninjaquick on 3/21/2008 3:44:09 PM , Rating: 3
Really even at 45nm they might not make it beyond 3.0GHz at the nominal voltage just because they have massive transistor count and the method they use to make the transistors.
AMD sticks a massive amount of technology in their chips, but use the aging SOI process.
They really need to focus more on getting tech in their CPUs that gets used my programmers, and hardware that bleeds less electricity (which was the beauty of SOI back in the day of K8). Problem is that as transistors get smaller, they get closer together, so the electrons sometimes decide to hop from one pathway to another. It kinda makes a mess.


RE: 45nm
By amanojaku on 3/21/2008 12:53:40 PM , Rating: 3
quote:
Things are not looking to bright for AMD


I don't agree. AMD is certainly not playing at the level of Intel, but it is still competitive, particularly around pricing. I agree that the CPU speeds need more clock and that 45nm is late compared to Core2, but who cares? If you already have an aging CPU any new AMD will most likely outperform it. If you're without a PC any new AMD CPU will certainly be a step up!

We haven't seen anything as significant as the original Athlon, but I'm keeping my fingers crossed: for a long time Intel hadn't seen anything like the Core/Core2 before it debuted, either.


RE: 45nm
By tastyratz on 3/21/08, Rating: 0
RE: 45nm
By batman4u on 3/21/2008 1:16:23 PM , Rating: 1
Me too.....never knew what happenned :(........... and then they massacred ATI by firing too much ATI personel and selling info to Nvidia is because ATI got behind and now neither of two companies are in Top of the game ........so sad......... and now intel wants to come in the graphics game..... i wouldnt be supprised if intel in 2 years is behind nvidia and ATI third.


RE: 45nm
By adam92682 on 3/21/2008 11:38:59 PM , Rating: 2
intel is already #1.


RE: 45nm
By Pirks on 3/23/2008 3:03:35 AM , Rating: 2
They are #1 in sales, but not in quality. They sell insane amount of poo-poo called "Intel Video". Yes, they sell A FREAKIN' LOT of it, but it doesn't make it anything but poo-poo.

Ever read anand's review of AMD 780G where they compared latest nVidia, AMD and Intel IGPs? In that review AMD kicked nVidia's #ss pretty good, and Intel's crap couldn't even FREAKIN' RENDER A GAME! I couldn't believe my eyes, the frame rate was like 2 FPS in Intel G35 IGP and the screen was like total grey... there goes your #1


RE: 45nm
By Sulphademus on 3/24/2008 3:51:35 PM , Rating: 2
They are number one in volume only.
How many of us have a Mobo with integrated graphics (intel) and then slap in an 8800GTX? That counts as a sale in both Intel's and NVidia's columns, even though the Intel isnt used.

And Intel's video seems to be good enough for Uncle Buck who just surfs eBay and plays solitaire.


RE: 45nm