Tech Report Woodcrest preview - Woodcrest 3.0 VS. Opteron 2.6

[Eug]

The part I don't understand is why you took Intel's PR claims as some sort of gospel in the first place.

 

[PetNorth]

sorry meanwhile you posted I was editing my previous post:

"not only claims, but Intel IDF benches"

Do you remember the WOWS, WOOTS claims from guys here and there?

 

[dexvx]

Originally posted by: PetNorth

Originally posted by: Eug

Originally posted by: PetNorth
Considering Intel claims 20-40% faster at same clock speed, I can't see it...

Cinebench overall Woodcrest 3.0ghz 15% faster than Opteron 2.6ghz
POV-Ray 4 threads Woodcrest 3.0ghz 6% faster than Opteron 2.6ghz
3DS Max overall Woodcrest 3.0ghz 21% faster than Opteron 2.6ghz
Windows Media Encoder Woodcrest 3.0ghz 5% faster than Opteron 2.6ghz
Xmpeg / DivX Woodcrest 3.0ghz 16% faster than Opteron 2.6ghz
Lame MT overall Woodcrest 3.0ghz 19% faster than Opteron 2.6ghz

This is about 15% overall faster (again Woodcrest 3.0ghz vs Opteron 2.6ghz)

(sintetic nonsense benches jumped)

btw 5% less power consumption at full load (it would be around 10% at same clock) isn't impressive either (specially 0.065 vs 0.090)

1) I always take the claims by the manufacturer with a teaspoon of salt.
2) 15% faster and 5% lower power is still significant. Even of AMD can up the performance by 10% and keep the power usage the same as now by July, Intel will still have the overall lead. I think that's quite an impressive turnaround for Intel, considering the comparatively poor performance/Watt ratio in their previous server chips.

again, considering Intel claims 20-40% faster at same clock speed.

What part you don't understand?

edit: And not only claims, but Intel IDF benches (remember?)

Tahts for single core.

Multi-core CPU's get degraded performance as it scales up in number. Happens with current Xeons, will happen with Woodcrest, and it even happens *gasp* to Opterons.

15% faster in quite impressive for 4P configuration when you think Intel is still using the DIB method. And this is on Alpha hardware with slow FB-DIMMs (not to mention the Opteron system has 2GB memory vs 1GB for the Woodcrest).

 

[SexyK]

The IDF benches were Conroe and not Woodcrest... so I don't see why you're bringing up those tests. They are completely different platforms.

 

[Eug]

Well, I guess I prefer to look at actual benchmarks done by independent sites.

I'm just impressed that Woodcrest is performing so well, and at a lower power usage too. If you ignore previous hype and just look at the numbers, Intel is doing very well.

 

[PetNorth]

Originally posted by: dexvx

Originally posted by: PetNorth

Originally posted by: Eug

Originally posted by: PetNorth
Considering Intel claims 20-40% faster at same clock speed, I can't see it...

Cinebench overall Woodcrest 3.0ghz 15% faster than Opteron 2.6ghz
POV-Ray 4 threads Woodcrest 3.0ghz 6% faster than Opteron 2.6ghz
3DS Max overall Woodcrest 3.0ghz 21% faster than Opteron 2.6ghz
Windows Media Encoder Woodcrest 3.0ghz 5% faster than Opteron 2.6ghz
Xmpeg / DivX Woodcrest 3.0ghz 16% faster than Opteron 2.6ghz
Lame MT overall Woodcrest 3.0ghz 19% faster than Opteron 2.6ghz

This is about 15% overall faster (again Woodcrest 3.0ghz vs Opteron 2.6ghz)

(sintetic nonsense benches jumped)

btw 5% less power consumption at full load (it would be around 10% at same clock) isn't impressive either (specially 0.065 vs 0.090)

1) I always take the claims by the manufacturer with a teaspoon of salt.
2) 15% faster and 5% lower power is still significant. Even of AMD can up the performance by 10% and keep the power usage the same as now by July, Intel will still have the overall lead. I think that's quite an impressive turnaround for Intel, considering the comparatively poor performance/Watt ratio in their previous server chips.

again, considering Intel claims 20-40% faster at same clock speed.

What part you don't understand?

edit: And not only claims, but Intel IDF benches (remember?)

Tahts for single core.

Multi-core CPU's get degraded performance as it scales up in number. Happens with current Xeons, will happen with Woodcrest, and it even happens *gasp* to Opterons.

15% faster in quite impressive for 4P configuration when you think Intel is still using the DIB method. And this is on Alpha hardware with slow FB-DIMMs (not to mention the Opteron system has 2GB memory vs 1GB for the Woodcrest).

or wathever...

 

[PetNorth]

Originally posted by: SexyK
The IDF benches were Conroe and not Woodcrest... so I don't see why you're bringing up those tests. They are completely different platforms.

Woodcrest is conroe, like conroe is merom or merom is woodcrest, like A64 is Opteron, Turion is A64 and go on

 

[PetNorth]

Originally posted by: Eug
Well, I guess I prefer to look at actual benchmarks done by independent sites.

100% agree here 😉

 

[SexyK]

Originally posted by: PetNorth

Originally posted by: SexyK
The IDF benches were Conroe and not Woodcrest... so I don't see why you're bringing up those tests. They are completely different platforms.

Woodcrest is conroe, like conroe is merom or merom is woodcrest, like A64 is Opteron, Turion is A64 and go on

Sure, the architecture is basically the same, but the platforms are very different. (Hence the different memory, core logic, etc...) Those aspects of the platform as a whole can have a tangible performance impact, especially when comparing desktop platforms and server platforms. Server platforms typically have lower overall performance because of the required error checking, stringent validation, etc...

 

[robertk2012]

Originally posted by: PetNorth

Originally posted by: SexyK
The IDF benches were Conroe and not Woodcrest... so I don't see why you're bringing up those tests. They are completely different platforms.

Woodcrest is conroe, like conroe is merom or merom is woodcrest, like A64 is Opteron, Turion is A64 and go on

This is not a single processor test. THERE ARE TWO PROCESSORS AND 4 CORES. SLOW RAM. AND a established opty platform that has the multiple processor platform about as effecient as you can get it.

 

[imported_Goo]

Where is the data base, web server branchmark?

 

[Henny]

Now Opteron is getting spanked! I love competition.

 

[Accord99]

Originally posted by: Goo
Where is the data base,

http://www.intel.com/performance/server/xeon/database.htm

TPC-C, arguably the most important DB benchmark available. The extra memory the Woodcrest system would boost scores a decent amount, but it's still very fast.

web server branchmark?

http://www.intel.com/performance/server/xeon/java.htm
http://www.intel.com/performance/server/xeon/web.htm

 

[Viditor]

Originally posted by: Questar
Cinebench Render x64 multi 12.5%
Cinebench Render x64 single 17.3%
Cinebench Render x86 multi 22.2%
Cinebench Render x86 single 25.8%
Cinebench Shading x64 15.2%
Cinebench Shading x86 11.9%
3DMax Render 22%
3DMax Metalray 21.5%
Windows Media Encoder 5.5%
DivX6 15.9%
Lame CBR MS single 27.7%
Lame CBR MS multi 26.1%
Lame CBR Intel single 28.6%
Lame CBR Intel multi 28%
Lame VBR MS single 17.2%
Lame VBR MS multi 5.2%
Lame VBR Intel single 24.1%
Lame VBR Intel multi 16.6%
Sphinx MS 23%
Sphinx Intel 24.8%
picCOLOR overall 34%

Power at idle 19%
Power at load 5%

All using 533Mhz memory instead of 667.

533 MHz FBD...there's a HUGE difference!
From Intel's sheet on FBDimms Link

"1. Provides over 3 times higher memory throughput? allowing for superior application responsiveness
2. Enables increased capacity and speed to balance capabilities of dual core processors
3. Allows for Intel® I/O Acceleration Technology to more quickly access and process data
4. Performs reads and writes simultaneously; eliminating the previous read to write blocking latency"

BTW...thanks for the math. That helps!

 

[imported_SLIM]

Yep, it's amazing a brand new unreleased product/architecture can beat a mature ie three year old opteron core :disgust:.

I'm glad to see competition come back to cpu/server platforms, but this hysteria about a brand new architecture being 10-20% faster in some benches is kind of silly. Next spring, I think AMD will give a pretty good response with their K8L revamp followed by quad cores at the end of the year. Maybe by then, Intel will have CSI going so we can see nice competition from 1 to 8 sockets.

Anyhow, I too would have liked to have seen some 3rd party webserver and database tests...

:beer: to competition, :thumbsdown: to over the top "spanked, killed, etc" comments

 

[terentenet]

Check this link
http://sharikou.blogspot.com/2006/05/intel-woodcrest-performance-claim.html

 

[TrevorRC]

Worth pointing out that Woodcrest actually has a typical TDP of 65W... mentioned somewhere on L'Inq, the yields were that good.

--Trevor

 

[TrevorRC]

Originally posted by: terentenet
Check this link
http://sharikou.blogspot.com/2006/05/intel-woodcrest-performance-claim.html

That's the same idiot who said Conroe was a fraud.

Look at all the people at XS dominating the benchmarks.

I need to PM a mod about filtering that site out. It's so AMD-Biased it's ridiculous.
--T

 

[DrMrLordX]

Seconded Trevor, though I don't know if they do any filtering of sites out here. Sharikou's blog = worthless.

 

[dug777]

what a clown 😉

 

[Viditor]

Originally posted by: TrevorRC
Worth pointing out that Woodcrest actually has a typical TDP of 65W... mentioned somewhere on L'Inq, the yields were that good.

--Trevor

A slight correction...the Woodcrest processors below 3 GHz are 65w TDP, the 3 GHz are 80w. However keep in mind that TDP is only a thermal guideline for building systems and NOT a guide as to power usage...

 

[kknd1967]

Now AT posted their test. Woodcrest finally seems to be impressive!

 

[DrMrLordX]

Yes, the AT tests don't really fall in line with the results from Techreport. Curiously enough, Woodcrest seems to suck at encryption.

 

[Kougar]

Originally posted by: zephyrprime
Look at this picture:
core pictures

OMG the Tulsa die is humongous! It's almost as big as an itanium.

I know my reply is a little late... Not sure whether you meant to insinuate that Tulsa is a "Core" uArch or not... but it's not 😉 Tulsa is basically a Presler (Dual Core, old Netburst style) that was spliced together with a shared 16mb L3 cache :Q