AMD Readies ?Thuban? Six-Core Desktop Processor

[Idontcare]

AMD Readies ?Thuban? Six-Core Desktop Processor

Advanced Micro Devices is preparing a desktop processor with six processing engines, sources familiar with the company?s plans revealed. The new central processing units (CPUs) will not be available this year, but are likely to boost performance of AMD?s desktop platforms sometime in 2010.

AMD?s processor code-named Thuban is the company?s first desktop processor with six processing engines. The microprocessors will be compatible with socket AM3 infrastructure and will have integrated dual-channel PC3-10600 (DDR3 1333MHz) memory controller. It is very likely that Thuban processors will retain AMD Phenom II brand name as well as design of the code-named Istanbul chips for servers, thus, will feature 3MB L2 cache (512KB per core) and 6MB of L3 cache. The chips will be made using 45nm SOI fabrication process.

http://www.xbitlabs.com/news/c...Desktop_Processor.html

Keeping in mind this is coming from the guys who recently had to print a retraction of their "ZOMG AMD got SMT coming soon!" article, but this one stands to reason given the existance of 6-core Istanbul already.

Thuban...sounds like "thumpen" to me, as in AMD is ready to give i7 a good Thuban with 6cores going against i7's 4C/8T configuration :laugh:

Question of course is what kind of TDP per socket this thing is going to command, as that will limit clockspeeds. 6 cores at 4GHz would be nice, if she tops out at 3GHz then not so much.

 

[ZOXXO]

I suppose this could do until the introduction of mangy cores.:laugh:

 

[ilkhan]

AMD did just release a 40W istanbul, makes this more likely if their yields are that good. Pricing will be...interesting, to say the least. And it'll probably compete well with bloomfield.

 

[VirtualLarry]

Wow, I wonder if this could steal the gaming crown away from the i7 (as if the i7 had it in the first place, I suppose, since many benchmarks put the PhII x4 just slightly ahead of the i7 for gaming).

All I know is, Go AMD!. I want one.

 

[dguy6789]

There's waaaaaaay more benchmarks that put the i7 as faster than the Phenom 2 in every single task including gaming. Not to mention I don't think there is even one gaming benchmark where the Phenom 2 is faster than the Yorkfield Core 2 Quad. That said, I personally would rather have a hex core Phenom II than an i7.

 

[LoneNinja]

Originally posted by: VirtualLarry
Wow, I wonder if this could steal the gaming crown away from the i7 (as if the i7 had it in the first place, I suppose, since many benchmarks put the PhII x4 just slightly ahead of the i7 for gaming).

All I know is, Go AMD!. I want one.

Odds are the Phenom II X4 will do better in gaming than this X6 processor will do. It's very likely that 6 cores will reduce the clock speed and overclocking capabilities when compared to 4. Instead of a fast dual core is better than a quad for gaming, it'll just become a fast quad is better than a hexcore for gaming.

It does seem that this processor won't feature a DDR2 controller, looks as though AMD plans to remove socket AM2+ compatibility next year, at least for the new high end parts.

 

[drizek]

These aren't going to help much for gaming, but they should be a lot of fun for other things. This basically allows AMD to stay competetive at the high end, and obviously their yields are good. I bet these would come in at under $400 easy. 125W is going to be almost a certainty, but who cares as long as the power savings features are properly implemented. If they clock/volt down the way they should, at idle they will only use a few watts more than current CPUs, and at load they will offer better performance/watt, so you're OK either way. Moreover, if they can get them to function correctly in Unganged mode then when running single threaded apps you won't have much of a power use penalty over a dual/quad core.

The great thing about 6 cores with AMD is that they can now compete with nehalem on all fronts. In single threaded apps, clock for clock nehalem was never too far ahead. It is only in multithreaded apps where HT and the other stuff pull Intel into a huge lead. Now with 6 cores, it is going to be in these multithreaded apps that these cpus will show their strength.

I know this is really non-realistic, but even at 3ghz these would be interesting. 6x3ghz=18ghz. 4x4ghz=16ghz. So in theory at least, these should have a bit more power than a 4ghz quad core Phenom. With perfect scaling, it should be neck and neck with a 4ghz i7, right?

I may well get one of these things if they come in at under $200(eventually). It would be a decent upgrade to my X3 and it will extend the life of my computer all the way to Bulldozer.

Edit: I reread it and yes, the article does make it sound like DDR2 is going to get the boot. That is unfortunate, as my motherboard serves all my needs. I just want to throw in a USB 3 controller card in a couple months and it should be perfectly up to speed.

 

[exar333]

Will this be as fast per clock or overclock as well as a 6-core Westmere?

 

[drizek]

No, but it also won't have a 4 figure price tag.

 

[eternalone]

Couldnt they just keep a DDR2 controller on there, that way its AM2+ compatible. That would really help people in these hard economic times in my opinion.

 

[ilkhan]

Originally posted by: drizek
I know this is really non-realistic, but even at 3ghz these would be interesting. 6x3ghz=18ghz. 4x4ghz=16ghz. So in theory at least, these should have a bit more power than a 4ghz quad core Phenom. With perfect scaling, it should be neck and neck with a 4ghz i7, right?

Comparing clock speeds that way is a really, REALLY bad idea and isn't at all accurate.
Its only applicable if each chips is from the same series and has equal cache, bus speeds, uncore speeds, etc. And even then it doesn't take into consideration the hyper-threading advantage the i7s have.
But yes, for tasks that scale perfectly, and ignoring the architectural differences, a hex core is 50% faster in certain tasks at the same clock speeds.

If nothing else this might give Intel a reason to release cheap gulftowns earlier than current roadmaps show.

 

[hans007]

Originally posted by: eternalone
Couldnt they just keep a DDR2 controller on there, that way its AM2+ compatible. That would really help people in these hard economic times in my opinion.

well the istanbul processors still have ddr2 support. maybe that means yes.

 

[Cookie Monster]

I thought the AM3 based processors have an IMC that supports both DDR2 (being AM2+ backward compatible) and DDR3 (AM3) unless they've changed that with this chip.

So basically they are tweaking instanbul to be socket AM3 instead of socket F (1207 pins)? Guessing yields are pretty good atm, but now the final question remains. What is it going to be clocked at?

 

[Yukmouth]

6 cores?? Sounds smart, @ default voltage's these Phenom's run extremely cool so the headroom is there.

Hopefully that part drops price's on the Quad 965s, I want 4ghz already!

 

[TurtleCrusher]

Things are going to be real interesting over the next few months.

/lawn chair
/popcorn

I cant wait to see how this influences the market. AMD is going to tout their "first 6-core consumer CPU" in Intel's face, yet Intel's i7 performs just like an octo core part.

 

[dguy6789]

The i7 is far, far way away from achieving the performance of an 8 core processor. It's only marginally faster than the 4 core Yorkfield. An 8 core Yorkfield or Phenom II would totally blow away the current i7's 4 core 8 thread setup.

 

[markodude]

Originally posted by: dguy6789
The i7 is far, far way away from achieving the performance of an 8 core processor. It's only marginally faster than the 4 core Yorkfield. An 8 core Yorkfield or Phenom II would totally blow away the current i7's 4 core 8 thread setup.

Not strictly true, an i7 overclocked to 5ghz is VERY close to a last gen skulltrail (dual quad) setup in terms of raw MFlops - I have done the tests........also to say its only marginally faster than Yorkfield is a bit off, what you have to remember is only certain apps reap the benefits of huge memory bandwidth increases and multi-threading.

Some people may be playing with a 6-core 32nm Gulftown Intel CPU at 4.4 on air, and may not think Intel have much to worry about.

 

[IntelUser2000]

Originally posted by: dguy6789
The i7 is far, far way away from achieving the performance of an 8 core processor. It's only marginally faster than the 4 core Yorkfield. An 8 core Yorkfield or Phenom II would totally blow away the current i7's 4 core 8 thread setup.

In apps that are well threaded(8 threads), its pretty close. That's why in servers where high number of threads are common, i7 based Xeons are heralded as the next best thing. Bloomfield based Xeons with 2 physical processors sometimes even beat 4 Dunningtons, which are high end MP CPU based on Penryn uarch and has 6 cores each.

 

[Cookie Monster]

Dunningtons are quite heavily bottlenecked by the FSB. Istanbuls im sure also give it a run for its money (correct me if Im wrong).

 

[MODEL3]

Nevertheless, based on currently available information, AMD Thuban is due only in Q3 2010

Damn!
I thought AMD would launch them in Q2 along with leo.

The microprocessors will be compatible with socket AM3 infrastructure and will have integrated dual-channel PC3-10600 (DDR3 1333MHz) memory controller

Since Xbitlabs is saying Q3 2010 shouldn't the memory controller support DDR3 1600?

I mean even low end IGP chipset like the 785G (nearly same speed with a Q1 2008 780G) support PC3-10600 (DDR3 1333)!
785G launched now (Q3 2009)
Shouldn't after a year in Q3 2010 the speed to be 1600?
I thought that every 3 years the speed doubled (800/1066/1333/1600)

Moreover, as future video games that rely on DirectX 11 start to arrive, the advantages provided by six-core Thuban and Istanbul processors will be even more apparent.

If 2010 DX11 games can show performance difference if they move from 4 to 6 cores that would be great, but i don't think that will be the case! (i think that even in 2011 we will not see great performance difference if we move from 4 to 6 cores, i mean for the majority of the games)

 

[taltamir]

Originally posted by: VirtualLarry
Wow, I wonder if this could steal the gaming crown away from the i7 (as if the i7 had it in the first place, I suppose, since many benchmarks put the PhII x4 just slightly ahead of the i7 for gaming).

All I know is, Go AMD!. I want one.

not anytime soon... games scaling will be limited for a while a yet, and the turbo boost of intel is a big weapon.

 

[Idontcare]

Originally posted by: ilkhan

Originally posted by: drizek
I know this is really non-realistic, but even at 3ghz these would be interesting. 6x3ghz=18ghz. 4x4ghz=16ghz. So in theory at least, these should have a bit more power than a 4ghz quad core Phenom. With perfect scaling, it should be neck and neck with a 4ghz i7, right?

Comparing clock speeds that way is a really, REALLY bad idea and isn't at all accurate.
Its only applicable if each chips is from the same series and has equal cache, bus speeds, uncore speeds, etc. And even then it doesn't take into consideration the hyper-threading advantage the i7s have.
But yes, for tasks that scale perfectly, and ignoring the architectural differences, a hex core is 50% faster in certain tasks at the same clock speeds.

If nothing else this might give Intel a reason to release cheap gulftowns earlier than current roadmaps show.

I agree it is a not-so-great metric for comparing processor capability but it is not entirely without merit either if we work thru the math a bit.

Clockspeed combined with IPC speaks to computing throughput for peak performance, TFLOPs for example is usually reference with respect to peak clockspeed x IPC.

So while comparing core x GHz to another core x GHz is not really a metric anyone would take the time to talk about, comparing core x GHz x IPC to core x GHz x IPC is a metric that does effectively get talked about a lot (spec_rate is basically this, as is cinebench) and for otherwise equivalent/identical architectures (such as Deneb vs. Istanbul) the IPC part is going to be the same for both chips, reducing the relevant differences between the two competing products down to a simple cores vs. clockspeed equation.

I'm only posting this ilkhan because when I read drizek's post I had the same "egads!" gut response I get the feeling you had upon reading the Ghz x core comparison, but then I rationalized it to myself as I just did above in this post and I decided yeah in this case its not such a terrible way to reduce the salient product differences here to zero-order.

Originally posted by: MODEL3

Nevertheless, based on currently available information, AMD Thuban is due only in Q3 2010

Damn!
I thought AMD would launch them in Q2 along with leo.

The microprocessors will be compatible with socket AM3 infrastructure and will have integrated dual-channel PC3-10600 (DDR3 1333MHz) memory controller

Since Xbitlabs is saying Q3 2010 shouldn't the memory controller support DDR3 1600?

Q3 2010? Hmmm...that's getting dangerously close to when we'd expect a 32nm release timeline...Could the story here really be that Thuban is 32nm 6-core Phenom III?

 

[ilkhan]

I thought that istanbul introduced a couple of things to help with IPC (snoop filter IIRC). Not a huge difference for sure, but something.
Yeah, I did the "egads!" thing, but I've done the comparison myself. its a simplistic but sometimes worthy method of comparison (I keep think a gulftown at 4Ghz would be like 6-7x faster at encoding than my poor Q8200 at 2Ghz).

The 32nm musing is interesting though. Could be their first 32nm release and the last chip before BD. :shrug:

 

[drizek]

I was just thinking, maybe they dropped DDR2 support because they were finally seeing real benefits to DDR3. That would imply that it is running at faster clocks than the DDR2 Opterons.

I'm only posting this ilkhan because when I read drizek's post I had the same "egads!" gut response I get the feeling you had upon reading the Ghz x core comparison

"I'm going to get flamed for this" was what was going on in my head when I was writing it. I reread it to make sure that I had "probably", "in theory", "perfect", etc. in there before I posted. The phenom-phenom should be close to a direct comparison. Even the L3 shouldn't be an issue since from the reviews I've seen, the 8xx series don't really take much of a hit compared to the 9xxs. The i7 it is obviously more complicated than that, but just the simple fact that it has 6 physical cores helps to counter nehalems SMTP advantage.

 

[Idontcare]

Originally posted by: drizek
I was just thinking, maybe they dropped DDR2 support because they were finally seeing real benefits to DDR3. That would imply that it is running at faster clocks than the DDR2 Opterons.

The semi-inebriated conspiracy nut inside me tends to wonder if there isn't some manner of mobo-support ecosystem "spread the wealth" effects going on here.

Last thing an AM2+ mobo maker wants to hear from AMD is "guess what guys!? This is great news, for our customers I mean, but we are going to release this uber cored chip in 2010 and its not going to cost our existing customers a dime in terms of mobo upgrades because we are counting on you guys to write BIOS updates, free of charge of course, so our customers can just pop these new chips into the mobo's you already sold them in 2008! Isn't that exciting! The future is Fusion!".

On the flipside given the issues AMD has had with mobo makers (not) creating BIOSes that properly implement CnQ and compatibility for Phenom I would not be surprised if AMD elected to avoid allowing this to happen again. (which would only further tarnish AMD's reputation)

Originally posted by: drizek

I'm only posting this ilkhan because when I read drizek's post I had the same "egads!" gut response I get the feeling you had upon reading the Ghz x core comparison

"I'm going to get flamed for this" was what was going on in my head when I was writing it. I reread it to make sure that I had "probably", "in theory", "perfect", etc. in there before I posted. The phenom-phenom should be close to a direct comparison. Even the L3 shouldn't be an issue since from the reviews I've seen, the 8xx series don't really take much of a hit compared to the 9xxs. The i7 it is obviously more complicated than that, but just the simple fact that it has 6 physical cores helps to counter nehalems SMTP advantage.

I'd say given the lack of responses to it, outside mine and ilkhan and we are just trying to be helpful on expounding on this metric for the greater audience, that you had yourself totally covered and well caveated.

Its unfortunate that you've been so battle-hardened to know you have to be wary of these things, at the same time it is kind of nice that folks take the time to think it thru as you did because that means the forums are not filled with posts full of ill-conceived metrics for comparing processors.

If this is the worst we've come to then this ain't half bad in my opinion.