The Return of Mhz war

[Cooler]

http://www.theinquirer.net/?article=31712

According to Inq AMD will be ramping up speeds next year on 65nm process with Stress Memorization Technology.

 

[BassBomb]

of course it was gonna happen, if they can make chips cooler than intel's 65 nm process on 90nm
think how cool their 65s turn out.. that can let them increase clockspeeds

and inq is bullshit

 

[compgeek89]

AMD may be ramping up mhz, but that doesnt mean they are claiming their CPUs are better simply because their clockspeed is higher like Intel did.

 

[Absolute0]

I totally saw this, AMD ramping up Mhz because A64 is established and after a die shrink they'll have plenty of Mhz headroom, i think we should be seeing some overclocks to 3.4 Ghz on modest cooling.

Notice the role reversal now though, AMD ramping up clockspeed while Intel makes the more efficient architecture.

 

[Furen]

Just because clock speed isn't everything it does not mean that clocks are meaningless. Performance depends on both the efficiency (IPC) and the clock speed. If either of the two is low the overall performance will be low unless the other is extremely high. With Conroe Intel will increase efficiency with a very modest increase in clock speed (compared to Yonah) while AMD will use a clock speed hike as a stop gap before its next major revision.

 

[BassBomb]

they will both be fighting over architecture not mhz war

amd is just upping because they can, not to create a mhz war.. trying to push more and more mhz to get ahead

 

[Fox5]

Originally posted by: Absolute0
I totally saw this, AMD ramping up Mhz because A64 is established and after a die shrink they'll have plenty of Mhz headroom, i think we should be seeing some overclocks to 3.4 Ghz on modest cooling.

Notice the role reversal now though, AMD ramping up clockspeed while Intel makes the more efficient architecture.

Don't expect too big of a clock increase, unless AMD's increased pipeline length too.
The fastest common 130nm was 2.4ghz, with a 2.6ghz top end chip.
The fastest 90nm was 2.8ghz with a 3ghz top end chip.
Another 400mhz would bring us up to 3.2ghz to 3.4ghz.
And even assuming they do a bit better than that (going by the percentage increase + additional process techniques) they'll maybe make 3.6ghz.
Isn't Conroe supposed to launch at 3.1ghz? That would put it about dead even in performance with AMD's 65nm.

 

[Cooler]

Top Chip on Core 2 this year will be 3.3Ghz

 

[jazzboy]

Just saw this 2nd article from the inq.

So it looks like it could be 3.4 ghz X2s before the end of the year.

Performance wise that should at least be competitve with mainstream Conroes although it will likely get beaten still by the conroe XE.

Lets just hope that this doesn't turn out to be another "prescott to reach 5ghz" or "netburst to reach 10ghz" claim.

 

[Ika]

the difference between intel and amd is, intel created an architecture (netburst) meant for MHz; amd created an architecture (hammer? whatever A64 is) based on efficiency. now they're increasing clock speed in the face of looming danger (Core 2) to keep up with Intel.

 

[JPH1121]

Don't neglect that AMD has a VERY efficient and scalable architecture...

AMD gets a 1:1 improvement in memory performance as clock speeds ramp up. Intel on the other hand gets a much lower improvement to mem performance.

AMD is STILL going to win the server market I think...at least 4-way+ servers (not suprisingly where the most $ can be made).

Increasing clock speeds on an AMD processor is a GREAT idea. Increasing clock speeds on a core platform is just a good idea.

 

[Aluvus]

Originally posted by: guoziming
the difference between intel and amd is, intel created an architecture (netburst) meant for MHz; amd created an architecture (hammer? whatever A64 is) based on efficiency. now they're increasing clock speed in the face of looming danger (Core 2) to keep up with Intel.

The official name is K8, though it's often been called Hammer or AMD64 or other such things.

 

[imported_UncivilizedAMD]

http://www.hkepc.com/bbs/news.php?tid=599540

according to this site, these will be what is released q1 07. Highest chip is 3 ghz fx-64 which still trails in ghz to Conroe's proposed 3.33 xe.

 

[Fox5]

Originally posted by: Aluvus

Originally posted by: guoziming
the difference between intel and amd is, intel created an architecture (netburst) meant for MHz; amd created an architecture (hammer? whatever A64 is) based on efficiency. now they're increasing clock speed in the face of looming danger (Core 2) to keep up with Intel.

The official name is K8, though it's often been called Hammer or AMD64 or other such things.

I think Hammer was the name of the initial Opteron core.
AMD64 is the instruction set maybe?

 

[Hard Ball]

Originally posted by: Fox5

Originally posted by: Aluvus

Originally posted by: guoziming
the difference between intel and amd is, intel created an architecture (netburst) meant for MHz; amd created an architecture (hammer? whatever A64 is) based on efficiency. now they're increasing clock speed in the face of looming danger (Core 2) to keep up with Intel.

The official name is K8, though it's often been called Hammer or AMD64 or other such things.

I think Hammer was the name of the initial Opteron core.
AMD64 is the instruction set maybe?

"Hammer" was the codename for the uarchitecture during a certain period of 2001-2002;
You are right, AMD64 is the official name of the x86_64 ISA
The initial Opteron core is code named "Sledgehammer"
The initial Athlon64 core is code named "Clawhammer"

 

[coldpower27]

Brsibane isn't going to have a significant impact this year if at all.

We'll see what price AMD charges for these in comparison to Intel. As well as what performance level they provide. 2.8GHZ on the 90nm node is likely to be the highest mainstream grade they can achieve. Not to mention that isn't likely to even beat the E6600 model of Conroe in performance.

 

[dexvx]

Originally posted by: JPH1121
AMD is STILL going to win the server market I think...at least 4-way+ servers (not suprisingly where the most $ can be made).

Hurricane

Current Xeon's on Hurricane can match current Opterons in 8-32 way. Since Conroe is more or less a drop in replacement, it should scale just as well.

 

[JPH1121]

I took a quick looksie but couldn't find anything that showed any benchmarks or anything along those lines comparing the Opteron and Xeon processors.

Not saying it isn't true, but from everything I've read, the Opteron is the king of the server world (at least lower cost, general purpose servers where the Xeon would be a consideration) as cpu quantity increases to anything past 8 cores.

 

[dexvx]

Originally posted by: JPH1121
I took a quick looksie but couldn't find anything that showed any benchmarks or anything along those lines comparing the Opteron and Xeon processors.

Not saying it isn't true, but from everything I've read, the Opteron is the king of the server world (at least lower cost, general purpose servers where the Xeon would be a consideration) as cpu quantity increases to anything past 8 cores.

You won't find many benchmarks (at least from 3rd party sites, SPEC or TPC is the best you'll get) because they are systems costing over $20K. I'm not saying Xeon/Hurricane is power friendly, it definitely is not. However, Paxville Xeons coupled with the Hurricane chipset performs on par with Opterons in 8 or higher way. With multiple memory channels and FSB's the Hurricane basically relieves the FSB constraints on the Xeon, and since I'm speculating that Conroe is more or less a drop in replacement (like the 975, it just took a VRM modification), it should scale just as well.

 

[Accord99]

Originally posted by: JPH1121
I took a quick looksie but couldn't find anything that showed any benchmarks or anything along those lines comparing the Opteron and Xeon processors.

One reason why is Opteron's max out at 8S, and even then scalability is poor beyond 4S due to cache coherency traffic consuming too much HTT bandwidth and CPUs being too many hops away resulting in increased memory latency. That's why HP, IBM nor Sun has produced a >4S Opteron system.

 

[Fox5]

Originally posted by: dexvx

Originally posted by: JPH1121
I took a quick looksie but couldn't find anything that showed any benchmarks or anything along those lines comparing the Opteron and Xeon processors.

Not saying it isn't true, but from everything I've read, the Opteron is the king of the server world (at least lower cost, general purpose servers where the Xeon would be a consideration) as cpu quantity increases to anything past 8 cores.

You won't find many benchmarks (at least from 3rd party sites, SPEC or TPC is the best you'll get) because they are systems costing over $20K. I'm not saying Xeon/Hurricane is power friendly, it definitely is not. However, Paxville Xeons coupled with the Hurricane chipset performs on par with Opterons in 8 or higher way. With multiple memory channels and FSB's the Hurricane basically relieves the FSB constraints on the Xeon, and since I'm speculating that Conroe is more or less a drop in replacement (like the 975, it just took a VRM modification), it should scale just as well.

Isn't hurricane considerably more expensive?
And what if the processors are working on something that relies on them getting data from each other, like say rendering a 3d scene, as opposed to just handling a sever workload?

 

[alex123]

Performance wise that should at least be competitve with mainstream Conroes although it will likely get beaten still by the conroe XE.

And what about PRICE wise? Would it be competitive with Conroe?

My only problem with AMD X2 is their price... However, 3 days ago I paid $330 CAD for X2 3800 rather then paying $145 CAD for PD 805... I am glad that Conroe launch would leave AMD no choice but cut 939 prices dramatically

 

[alex123]

This article is June 2005, but it talks about Hurricane chipset and "next several generations of Intel chip technology"... So there is a good chance that IBM will be able to tweak their Hurricane chipset for upcoming Woodcrest chips


...
The current xSeries 366 and xSeries 460 machines will support future Intel processors, by the way. "We have created an architecture that can run the next several generations of Intel chip technology," Mr Bretzmann boasts.

IBM is also not interested in using the Opteron processors from Advanced Micro Devices in the servers that support the Hurricane chipset, and as we reported earlier this year, Big Blue is not going to support the Itanium processor with Hurricane, either
...

 

[dexvx]

Originally posted by: Fox5

Originally posted by: dexvx

Originally posted by: JPH1121
I took a quick looksie but couldn't find anything that showed any benchmarks or anything along those lines comparing the Opteron and Xeon processors.

Not saying it isn't true, but from everything I've read, the Opteron is the king of the server world (at least lower cost, general purpose servers where the Xeon would be a consideration) as cpu quantity increases to anything past 8 cores.

You won't find many benchmarks (at least from 3rd party sites, SPEC or TPC is the best you'll get) because they are systems costing over $20K. I'm not saying Xeon/Hurricane is power friendly, it definitely is not. However, Paxville Xeons coupled with the Hurricane chipset performs on par with Opterons in 8 or higher way. With multiple memory channels and FSB's the Hurricane basically relieves the FSB constraints on the Xeon, and since I'm speculating that Conroe is more or less a drop in replacement (like the 975, it just took a VRM modification), it should scale just as well.

Isn't hurricane considerably more expensive?
And what if the processors are working on something that relies on them getting data from each other, like say rendering a 3d scene, as opposed to just handling a sever workload?

The initial cost isn't that much higher.

AFAIK, Opterons with more than 8way would lag worse with something that requires a lot of inter-CPU communications (although I'm not sure rendering is something like that). Has to do with cache coherency traffic (because they don't have a decent snoop now), as Accord has said.

 

[Viditor]

Originally posted by: dexvx

Originally posted by: Fox5

Originally posted by: dexvx

Originally posted by: JPH1121
I took a quick looksie but couldn't find anything that showed any benchmarks or anything along those lines comparing the Opteron and Xeon processors.

Not saying it isn't true, but from everything I've read, the Opteron is the king of the server world (at least lower cost, general purpose servers where the Xeon would be a consideration) as cpu quantity increases to anything past 8 cores.

You won't find many benchmarks (at least from 3rd party sites, SPEC or TPC is the best you'll get) because they are systems costing over $20K. I'm not saying Xeon/Hurricane is power friendly, it definitely is not. However, Paxville Xeons coupled with the Hurricane chipset performs on par with Opterons in 8 or higher way. With multiple memory channels and FSB's the Hurricane basically relieves the FSB constraints on the Xeon, and since I'm speculating that Conroe is more or less a drop in replacement (like the 975, it just took a VRM modification), it should scale just as well.

Isn't hurricane considerably more expensive?
And what if the processors are working on something that relies on them getting data from each other, like say rendering a 3d scene, as opposed to just handling a sever workload?

The initial cost isn't that much higher.

AFAIK, Opterons with more than 8way would lag worse with something that requires a lot of inter-CPU communications (although I'm not sure rendering is something like that). Has to do with cache coherency traffic (because they don't have a decent snoop now), as Accord has said.

Here's the link to comparisons in tpmC...Link
As you can see, the only Hurricane system in 4P to beat the Opteron on the DL585 requires a $640k FiberChannel drive array...this really throws off the comparison as TPC is a highly transaction-based benchmark.
As to cost, an x-series 460 IBM with 2 Xeon 3.33 (single core) and 2GB DDR2 is about $20,000.
An HP DL585 with 2 (single core) 2.2 GHz Opterons and 2GB memory is ~$9,400.

To understand the problems with 8-way Opterons, you must remember the structure of them...
1. Each CPU has it's own system memory (up to 4 slots usually per CPU).
2. Other chips can access any memory through the cHT connections to that chip. For the 2xx Opterons, there are 2 cHT connections, and the 8xx have 3 cHT connections.
3. In a 4P configuration, the most any connection to another chip's Ram will have to go through is 2 cHT connectors...this is still extremely low latency. However, in an 8P config, the max number is 3 connections, and the occurence of having to go through 2+ cHT connections is greatly increased. While this doesn't really effect the cache coherency that much, it does slow down the data rate for read/write to some of the system ram. It can also cause some minor scheduling conflicts for the cHT.

While there are certainly some white box 8P Opteron servers out there, the major OEMs have found that it's far more cost effective to cluster the Opteron than it is to put all of the CPUs on a single system...however, one of the changes due out for the upcoming Opteron+ systems at the end of this year is an increase in cHT connections. This should allow for 16P Opteron systems without the memory latency.