brisbane dual core 65nm

[nn_step]

Serge stages are useful in MODERATION.
Look at Northwood, wonderful design 20 stages
Prescott, hot and slow 31 stages.
Conroe 14
K8 12
K7 10

[Rickster_64]

Are they extending the number of stages on K8L? Probably been discussed 10,000 times but......

[nn_step]

well that entirely depends, on if AMD wants to add more clock speed

[MAS]

Are they extending the number of stages on K8L?

onlly AMD khows
i think no, at least for intreger blocks
maybe for fpu they will extend a little

[Serge84]

Serge stages are useful in MODERATION.
Look at Northwood, wonderful design 20 stages
Prescott, hot and slow 31 stages.
Conroe 14
K8 12
K7 10

Netburst had 31 stages and could OC like hell to clock to almost 8ghz. It was slow because of its cycles per clock where low. K8 only used 2 64-bit cycles, Conore uses 2 128bit cycles. K8L will do the same. It only depends how many stages K8L has. K8L is not K8 so it is unknown to me how many stages K8L has, I hear 14 to 16 stages. The max on 12 stages was 4.2ghz for K8's on the 939. Lets hope AM2 will get the same atleast once again with K8L.

The FPU has doubled in size btw.

[nn_step]

Serge perhaps the next logical step isn't more integer speed, rather a duality.
Since the main design targets for both AMD and Intel are Latency vs throughput.
What if, and this is entirely hypothetical: That there are two cores at different speeds, in the same processor.
One specialized for low latency, be 7-10 stages, and clock to say 2.5Ghz
Another specialized for Throughput, have lots of stages and say a 5Ghz clock speed.
Now the easiest of the two to program for is the lowest latency one.
And the Throughput one would have the highest performance.
Now by doing this idea you get a few special benefits:
1) your Throughput processor no longer needs to be superscalar. Meaning you are going to lose about 70% of the logic.(think 30mm^2 on 90nm)
2) Your Latency processor no longer needs to focus on throughput. Meaning IPC improvements are the highest goal. (Once again shedding alot of die space)
3) you can add different numbers of both to make a proper mix of performance for any enviroment.
Now with DX10, the low Latency version would be best for games.
And with the long pipeline and huge throughput that processor would also be killer for graphics and encoding.
In theory the most perfect hybrid processor. But this is complete speculation and may or may not have anything to do with K8L

[Motiv]

In theory the most perfect hybrid processor. But this is complete speculation and may or may not have anything to do with K8L

From the recent quotes out of AMD central, they may have something similar.

It's almost as tho the K8L will be the superscaler CPU, while through HT links "newer" co-processors will be slotted in.

maybe high Ghz Integer based ones, rather than the old ALU type.

It makes perfect sense if you think about it. Buy 1 CPU and add on packages for the rest (Pys, High Throughput)

If you get my drift

[Turtle 1]

Considering the last 2 post. Here's something from the Intel side. Very speculative, as were the last 2 post.

http://www.theregister.co.uk/2006/12...arrabee_gpgpu/

[Thesavage]

Be interesting to see how nvidia responds to that news ( Left in the cold)

[Shintai]

Ouch!

http://www.anandtech.com/cpuchipsets...spx?i=2893&p=8

65nm is slower than 90nm. Due to L2 cache changed from 12cycles på 20cycles. Quake 4 and HL2 is the worst perhaps with 5% lower FPS.

[RPGWiZaRD]

Ouch!

http://www.anandtech.com/cpuchipsets...spx?i=2893&p=8

65nm is slower than 90nm. Due to L2 cache changed from 12cycles på 20cycles. Quake 4 and HL2 is the worst perhaps with 5% lower FPS.

Ahh so that explains it.

[MAS]

Three SS that shook the world








lamers here, lamers there, lamers everythere
testers did try to reach HT 3.0 frequency but in vain

[nn_step]

well first batch isn't exciting, lets see what is going to come

[Shintai]

Three SS that shook the world
lamers here, lamers there, lamers everythere
testers did try to reach HT 3.0 frequency but in vain

Do your home work right, and you would see they tried a lower HT aswell. Same result.

So the lamer is you, not them.

[MAS]

they tried a lower HT aswell

SS pls

[MAS]

Here http://www.techreport.com/reviews/20...m/index.x?pg=4 and here http://www.techreport.com/reviews/20...m/index.x?pg=4 i only see suspicious and probably wrong K8 OC without driver DirectHands.sys installed

doctor is waiting

[Serge84]

Reviewers can never OC worth crap anyways. No AMD is that CPU FSB limited. Mine can reach fing 380. Lets see some of that and not crappy 230FSB marks. I mean come on juice up the cpus with 1.6v 1.7v 1.8v and lets see what happens to the extreme.

They didn't try a low enough multiplier and suck at OCing to begin with. Its more of a challange to get 350FSB then it is 230FSB is too ameture easy. And I dou't its the cpus/mobos fault. They prob only messed with it for 5 mins to get that speed. My cpu can do 2.7ghz without even changing a bloody thing upto 275FSB it starts to need changing.

[Shintai]

SS pls

Why dont you read what Anand wrote in the comments. I know you have a really hard time accepting it, and the 5% lower performance due to L2 is another kick.

Anand Lal Shimpi on: Dec 15, 2006 8:01 AMRating: 2
Sorry I should have addressed this in the review - HTT speed had no impact on my overclocking results with this particular chip; even with a HT multiplier of 4X the chip won't get into Windows at 2.99GHz. Memory wasn't a factor as it was set to the lowest speed possible in order to see how far we could push the CPU.

I have a feeling that with better air cooling close to 3GHz may be possible, but I wanted to look at the worst case scenario overclocking potential of just using a stock heatsink/fan similar to what we did in our Core 2 overclocking article upon its release.

I'm working on the Brisbane 4800+ now and will find out soon if it overclocks any better.

Take care,
Anand

http://www.anandtech.com/talkarticle...889&frmWhere=2

Maybe you should try show us your superiour skills!

[Equil|briuM]

Im not so botherd with Brisbane, 4x4 and all the rest.

Got myself a great deal on a 5200+ X2 and @ 3Ghz 24/7, it will last me till K8L easily.

Tbh, and this goes for Kentsfield.

AMD & Intel need to stop :banana::banana::banana:et-assing about on who can get ou the first product and just release the CPUs that matter.

Clovertown - True Quad Core.

K8L - What ever it will be - but something different instead of die shrinks or CPU coupling which has been around for years.

I think AMD are great.

A small firm in comparison to Intel keeping and for what was quite some time beating Intel in the market - I think they have more than enough to make a BIG come back.

Yeah, Conroe is a breakthrough - a breakthrough for Intel, but not a real technological breakthrough as after all - it only works out to be around 25-50% over AMD's current 3-4 year old K8 arcitecture.

Like I said, I think Clovertown is gunna be something else and along side it K8L.

Once more this is only the beggining - just look @ the talk of Fusion CPUs.

Great Stuff.

[cky2k6]

stop being an asswad mas, your fake ss before is enough, dont make yourself look worse by arguing a moot point... i run my ht upwards of 1200, it makes no difference in stability, maybe some of the first nforce 4 boards had issues, but were past that already.

[Shintai]

Seems even Anandtech was friendly with the highres benchmarks. 65nm K8s looks like a complete disaster in gaming.

http://www.firingsquad.com/hardware/...view/page5.asp

[Epsilon84]

Seems even Anandtech was friendly with the highres benchmarks. 65nm K8s looks like a complete disaster in gaming.

http://www.firingsquad.com/hardware/...view/page5.asp

Geez, don't rub it in anymore, it's torture already for AMD fans.

[nn_step]

Seems even Anandtech was friendly with the highres benchmarks. 65nm K8s looks like a complete disaster in gaming.

http://www.firingsquad.com/hardware/...view/page5.asp

what the duce, that page will not be displayed until after you accept half a dozen tracking cookies

[Zimon]

I really can't get to grips with why AMD have started to use these half multipliers....

I suspect they are responsible for most of the performance loss reported by both Anand and Firingsquad with those 4800+ chips:
As far as I understand, they cause a lower RAM speed, and a corresponding ~4% loss in theoretical bandwidth.

Interestingly, the difference in clockspeed between 4600+ and new 4800+ chips is also 4%...
- Maybe we're just seeing differences between MHz and bandwidth sensitive apps and games...?

Seems this would explain how the 4600+ is able to beat the 4800+ in some of the Firingsquad tests?

[cky2k6]

a 65nm 4800 is 2.5ghz with 512kb cache, so not its not the same as the 90nm part. from the firing squad review, the memory divider is probably why the 4800 is slower than the 4600 at times, as the 4600 uses a normal multi.