AMD says K10 will Knock Intel dead

Is AMD Skipping K9 processors?? I know K6, K7, K8.... now K10? Please don't tell me AM2 is K9...
 
i believe K9 is the X2.
K8L was the Turions

K10 is Barcelona/Agena/kuma derivatives of Barcelona

i say they are in better position because they now have all the Tier 1 OEMs working with them, they have 2 Fabs running (1 is now at 100% 65nm, the other is churning out 90nm and converting to become Fab38), they have GPU expertise from their ATI acquisition, they are established as a valid name in the server (lucrative) arena and their alliance with IBM (according to their announcments) will bring 45nm in mid-2008 to the server chips (Shanghai and octo-core).
 
Okay, I get the whole Canine thing now... LoL. AMD Will take a Bite out of Intels candy @$$.
 
pxc said:
Per clock maybe, but AMD is launching Barcelona at low clock speeds. The advantage disappears in real product availability.
Click to expand...

Phil Hester said 50% greater IPC/watt. As Kyle originally put it:

On IPC and AMD’s Barcelona Phil Hester had this to say: “We want to stay focused on upgrade compatibility. We must realize more IPC per watt, and our next-generation architecture will show a 50% improvement of IPC per watt.” So I came away with a somewhat clouded view on IPC
Click to expand...

This is odd.

Comparing a 120w quad core to a 120w dual core the dual core uses 60w/core, the quad is 30w/core.

The dual core is at I/w(core), where "I" stands for a number of instructions per clock cycle and "w" stands for watts. So I/w x 60w = 60I/core. Total instructions = 60I/core x 2 cores = 120I.

In the same way the quad core is 1.5I/w(core) x 30w = 45I/core x 4cores = 180I, thus maintaining continuity.

This would be silly considering C2Q is already scaling 1.6 to 1.7 over C2D at the same clocks. And if "native quad core" scales better than Intel's non-native whatever, K10 would be lower IPC than K8. That would not make sense. Phil probably botched the line. Maybe Kyle made him nervous :p
 
Bao01 said:
Phil Hester said 50% greater IPC/watt. As Kyle originally put it:



This is odd.

Comparing a 120w quad core to a 120w dual core the dual core uses 60w/core, the quad is 30w/core.

The dual core is at I/w(core), where "I" stands for a number of instructions per clock cycle and "w" stands for watts. So I/w x 60w = 60I/core. Total instructions = 60I/core x 2 cores = 120I.

In the same way the quad core is 1.5I/w(core) x 30w = 45I/core x 4cores = 180I, thus maintaining continuity.

This would be silly considering C2Q is already scaling 1.6 to 1.7 over C2D at the same clocks. And if "native quad core" scales better than Intel's non-native whatever, K10 would be lower IPC than K8. That would not make sense. Phil probably botched the line. Maybe Kyle made him nervous :p
Click to expand...

K10 uses direct power management so it can power down cores, cache, and such with power plains at the voltage level determened by load on each core and what speed each core is sepratly doing. Meaning core 1 can be at 100% with clocks at 100% at 1.2v wile the other 3 cores are pretty much turned to super low power state giving off almost 0 watts wile doing no tasks saying on idle. Allowing a quad to pull off 30watts on full load with a single core on and the other 3 have 0.5v or so.

That power advantage doesn't include the 40% performance advantage over intel eather. They maintain both lower then same power watts over C2Q and still have a 40% advantage in performance. K10Q scales 3.6x greater then C2Q, and 1.8x per socket upto 8 sockets. Intel is put in a no contest realm with 1P and up platforms when they only have wolfdile and yorkfield for about a year in only desktop setups. Giving equal performance in desktop until AMD raises the clocks to put more pressure on intel. Your core 3 doesn't come out until the end of 2008 agenst a much more powerful fusion APU before that time comes.

For a new arc like K10 there is never once been a clock limit on launch ever. 1st K8's started at 1.8ghz i beleave now they are sold as dual cores desined from the start but now at 3ghz took a year before they released 2.8ghz single cores thats like what a 40% increase then original max launch speeds now we are at about 60% with 3ghz. Whats K10's limit maybe the same so about 4ghz? IBM's POWER6 and CELL will do 5ghz and AMD is with IBM so its unknown. Its 90% new and mobular! Any part can be easally changed enhanced tweaked and added or connected to the ark with out any trouble. Its a arc that can be changed on the fly without a whole new arc so it pretty much has no limit in where it could go or what it could do. This is not the old days of past cpu's this is the next step to making APU's and fusion. K10 was desined to be intigratable with anything. You can imagine how scarry that could turn out to be. Easy way to make a native octo-core. Remember its mobular. The arc isn't limited to its current desine meaning it can get addons without mattering or needing a new desine.

K10 is the ultimate cpu desine and will be around longer then K8 ever was. This thing will be used in fusion and APU's. Could be changed abillion times in future configs. Amazing I must say, they have created a monster. On die smirks they could expand onto and widen stages. Its not a standard stick with one desine for 4 years like K8 was limited 2. This is the future of cpu arcs where any part can be greatly changed without needing a total core revamp like the old days.

K11 is fusion but most of it will be K10 expantions. They will add more stages if more clock speed or cycles are needed but until they replace silicon with diomends then APU's will be based off of K10 but replacing silicon with special materials is just changing materials then it could be the same desine only much faster. Diomends arn't even limited to the 22nm berrior heat or even GHZ berriors like silicone is so I'm sure we will see this arc around past 2011. But by then it wouldn't even look like the K10 as we know it.
 
I think by "must be" they are referring to their financial situation: dire straits.

They appear to be F'd if they don't go a level up on Conroe. Too bad within a few months 'Penryn' based Intel CPUs will release and once again show AMD what the bottom of the stock market looks like.

I still have my Winchester.. I just buy what gives better performance.
 
Nothing new there. I have bets open K8whatever will perform 5% better for integer applications than K8 at the same clockspeed (let's take gcc compilation as example). Didn't see any info indicating otherwise.
 
This is the first time AMD has been "flat on their back" since before they released the Thunderbirds, so its only a matter of time before they come back on top.
 
cyclone3d said:
what I dond't get is why a lot of people seem to think that it is impossible for AMD to re-take the speed crown....

AMD had the lead for how long???? Ever since the K7 came out.... the C2D was the first time in that many years that Intel beat AMD clock/clock.

For that matter.. a K6-2 with a board that had good L2 cache blew away Intels PIII clock/clock... and the K6-3 was even faster because of the on-die L2 cache.

And Benchmarks (especially ones that cater to Intel) don't count.. you have to take real world apps/games.
Click to expand...

Because this is the first time in a long while Intel decided to actual use architectural strength rather then rely on market perception to compete.

Clock for clock doesn't matter it's the overall performance of the processor that counts.

AMD was hurting bad before the K8 came in as well, but the rejuvenation that K8 brought them won't occur in the same way with K10 as Penryn will actually be an improvement at launch over Conroe, and the only place for Intel to go is up since Penryn isn't a major overhaul, but rather a safe evolutionary path. Prescott started worse off then Northwood, and only slowly improved over time, at a rate slower then the AMD K8 did.

I don't know what your talking about so benchmarks which favor AMD don't count now right. :rolleyes: Conroe beats AMD in real world applications optimizations or not, the fact that Intel has the ability to get developers to optimize for them is simply tough luck for AMD and does count regardless.
 
Bao01 said:
Phil Hester said 50% greater IPC/watt. As Kyle originally put it:



This is odd.

Comparing a 120w quad core to a 120w dual core the dual core uses 60w/core, the quad is 30w/core.

The dual core is at I/w(core), where "I" stands for a number of instructions per clock cycle and "w" stands for watts. So I/w x 60w = 60I/core. Total instructions = 60I/core x 2 cores = 120I.

In the same way the quad core is 1.5I/w(core) x 30w = 45I/core x 4cores = 180I, thus maintaining continuity.

This would be silly considering C2Q is already scaling 1.6 to 1.7 over C2D at the same clocks. And if "native quad core" scales better than Intel's non-native whatever, K10 would be lower IPC than K8. That would not make sense. Phil probably botched the line. Maybe Kyle made him nervous :p
Click to expand...

This would make sense if you start taking the difference in clock frequencies between the 2 architectures.

A performance improvement of 150% means that overall the processor would be 2.5x Dual Core K8 at the same TDP though he did say over the next 2 years mind you. So this kinda performance improvement has a dude date of September 2008. Pretty much in line with 45nm Bareclona's.

60L/Core x2 = 120L at 120W

75L/Core x 4 = 300L at 120W

So each core would be 1.25x as powerful as a K8 Core, which would put K10 mildly ahead of the current iteration of Conroe as Conroe is more or less 1.2x K8, but not overly so. This is very rough but in line with what alot of people expect same or mildly better per clock over Conroe.

Though a 50% Improvement in IPC per Watt is hard to quantify. So 1.5x IPC at x Wattage or the same IPC at 2/3 the Wattage. But Wattage is typically also related to how high a processor is clocked at...

Something, like going from a 5200+ F2 Stepping at 89W to a 65W F3 Stepping is an improvement of 0.4x so 1.4x the IPC at Wattage x.
 
InorganicMatter said:
As with all things ATI/AMD, I will believe it when I see it.
Click to expand...

hah, you could say the same thing about Intel. As for all this talk about TDP, remember that both companies measure TDP differently.

I'm eager to see some real Barcelona results.
 
well do we count the fact that AMD has some of the northbridge on the cpu die when counting performance per watt or is it measured on a system as a whole ?
 
Digital Viper-X- said:
well do we count the fact that AMD has some of the northbridge on the cpu die when counting performance per watt
Click to expand...

Yeah, the TDP measurement is for the entire CPU. AMD measures TDP as the maximum power that a CPU can possibly draw (under the "worst possible scenario") while Intel calls TDP the maximum you are likely to see sustained.
 
Digital Viper-X- said:
well do we count the fact that AMD has some of the northbridge on the cpu die when counting performance per watt or is it measured on a system as a whole ?
Click to expand...

Total system power draw and heat measured on the processor are the real figures that count. No matter what Intel or AMD says. When Independant testers post info, then we'll all know. Both Intel and AMD have lied or have not been real honest.
 
quadnad said:
Yeah, the TDP measurement is for the entire CPU. AMD measures TDP as the maximum power that a CPU can possibly draw (under the "worst possible scenario") while Intel calls TDP the maximum you are likely to see sustained.
Click to expand...

I see it as "Max" vs. "Avg". Any engineer understands that the Max (worst-case) value is more significant than the Avg value when you design something. You ALWAYS design for worst case scenario. Of course, this was a bigger lie back in the Prescott days...
 
Mav451 said:
I see it as "Max" vs. "Avg". Any engineer understands that the Max (worst-case) value is more significant than the Avg value when you design something. You ALWAYS design for worst case scenario. Of course, this was a bigger lie back in the Prescott days...
Click to expand...

yessir. I just don't think that most people know even know that they maintain different measurements for tdp...
 
You must log in or register to reply here.
Back
Top