The Sandy Bridge Preview (Anand)

[cegras]

Copy pasta, since I couldn't have said it better myself:

So the 3.1 GHz Sandy Bridge is 23% faster than a 2.8 GHz Nehalem. So roughly 10-15% faster clock for clock as Anand says in the teaser intro on the main page. Also, you get 10% less power. Not bad. However, that means that Bulldozer only has to slightly perform better than Nehalem to match Sandy Bridge clock-for-clock and live up to the 10-100W power envelope.

I change my vote on the weekly poll to match Sandy Bridge performance.

Be careful when comparing clock-for-clock. In the comments to the article, there was this exchange:

------------------------------
"I take it turbo was also disabled on the rest of the parts used to compare, right?"

"Turbo was enabled on everything else - SB performance should be higher for final parts.

Take care,
Anand"
------------------------------

The Nehalems are tricky: given half-way decent cooling (which I'm sure AnandTech has), they will Turbo even when all 4 cores are active. Thus, the supposedly 2.8 GHz Nehalem actually runs at a minimum of 2.93 GHz and goes up to 3.33 GHz in single or dual threaded tasks. A lot depends on what Turbo scheme Intel has chosen for Sandy Bridge.

[IDANPE]

So the 3.1 GHz Sandy Bridge is 23% faster than a 2.8 GHz Nehalem.

but there is no 2.8 GHZ Nehalem in this test..

760 is 3.06-3.33GHZ
880 is 3.2-3.7GHZ

[Jowy Atreides]

Yeah, 20% IPC increases just fall from the trees everyday... totally worthless! Meanwhile, Phenom II owners will... what? wait until Q4 of next year to buy a new BD platform for a Zambezi that will be significantly behind the competition? That's right, BD is officially NOT Am3 compatible:

http://www.xbitlabs.com/news/cpu/dis...forms_AMD.html

What a shocking revelation. That I posted about last week.

BD will bring a bigger jump in perf than nehalem to sb.

Intel is about to get athlon'd

[Hornet331]

880 is 3,33ghz (2 speedbins )

And if you look at Singelthreaded scores, the increase in performance is quite respectable... a 3,1ghz cpu beats a 3,73ghz cpu (ST)? Or in MT apps a 3,1ghz cpu beats a 3,33ghz cpu, and most of the time we are in the double digit range...

So offering more performance then a 880 and consuming 20% less energy while doing it.

[Booj]

I want to see 3dmark 01 & 05 on SB

[Hornet331]

What a shocking revelation. That I posted about last week.

BD will bring a bigger jump in perf than nehalem to sb.

Intel is about to get athlon'd

You know that amd has deliver a performance boost of 35-40% over deneb (per core) to be in the range of SB, and thats with the current numbers, without turbo.

[ridney]

this is great news and a good performing chip + gpu, interesting times ahead. i'm still excited on what ontario/llano brings to the table though

[BatteryOperated]

Intel is about to get athlon'd

In about 14 months. 'nuff said.

I am predicting 1q 2012 release. Seriously.

[IDANPE]

880 is 3,33ghz (2 speedbins )

not 1 ?

1/1/4/5
http://en.wikipedia.org/wiki/List_of...icroprocessors

[Chumbucket843]

Haha well, it looks like you're in for a world of hurt also. btw, how are those 32nm issues at intel coming along? Surely they've got it all worked out by now. No? Oh.

2 failures i must fix.

have you even read natarajan's paper? no. do you even understand the first thing about what makes a process good?

for one thing it is quite concerning when at iedm 2009 there were many papers on new 32nm processes and none of them were from global foundries. intel has the best gate and metal pitch, the fastest drive current, almost 2x faster pmos performance, and the fastest SRAM. seeing that they recently posted record profits they must be doing something right with their god awful manufacturing and architecture.

Yeah, 20% IPC increases just fall from the trees everyday... totally worthless! Meanwhile, Phenom II owners will... what? wait until Q4 of next year to buy a new BD platform for a Zambezi that will be significantly behind the competition? That's right, BD is officially NOT Am3 compatible:

fail. IPC is a misleading performance metric. you must have never multithreaded code or understand how it works i take it.

[meanmoe]

This may be a ridiculous, but could someone please explain what "HT on" means on an i5 part stated to have 4C/4T?

[Olivon]

This may be a ridiculous, but could someone please explain what "HT on" means on an i5 part stated to have 4C/4T?

In order to help Intel’s partners test HT functionality however, the i5 2400s being sampled right now have Hyper Threading enabled. For the purposes of our test I’ve run with HT both enabled (to give you an idea of higher end SB parts) and disabled (to give you an idea of i5 2400 performance).

On page : The test

[meanmoe]

On page : The test

thank you. somehow I missed that.

[flippin_waffles]

What a shocking revelation. That I posted about last week.

BD will bring a bigger jump in perf than nehalem to sb.

Intel is about to get athlon'd

Yep, or how about "KO'd".

[JumpingJack]

Yep, or how about "KO'd".

I am not sure how this is gonna happen, other than just brute force more cores (modules).

AMD's presentation was 33% more cores for 50% more throughput over MC, amortizing 50% over the core count does not give them the IPC improvement you guys are thinking it will.... the math just doesn't work.

Now, if AMD matched module count to Intel physical core count, it may turn out they are in good shape... but that would be one massive CPU in terms of die size.

[trinibwoy]

BD will bring a bigger jump in perf than nehalem to sb.

That's easy to do when your starting point is lower. But what does that mean for BD vs SB? In Anand's WoW test SB was 50%+ faster than a higher clocked Thuban and if Turbo was in fact disabled then it's only going to get worse. So are you saying that we should expect 50-60% performance improvements from BD too?

[trn]

I'm loving the performance per watt; not so much loving the new chipset locked 100bclk scheme although the K chips help... but it still sucks. Mid-end chips are Meh! and integrated graphics and transcoding silicone on a desktop can suck it! save that crap for laptops. Wish they would go ahead and step up their game with a Hexa/octo core launch of Sandy Bridge instead of 2c/4c launch.

[hurleybird]

You know that amd has deliver a performance boost of 35-40% over deneb (per core) to be in the range of SB, and thats with the current numbers, without turbo.

You're forgetting that AMD's plan isn't to compete core per core with Intel, but thread per thread. One Bulldozer module competes with one hyperthreaded core. Not saying that it isn't a long shot for AMD to equal/beat SB, but BD is a new kind of chip that challenges the way we typically think about performance.

[kl0012]

If I remember correctly, an Intel-FTC settlement forces intel to provide an interface for its cpus in next 6 years. That would be a big opportunity for nvidia to design chipset especially for overclockers.

[JumpingJack]

If I remember correctly, an Intel-FTC settlement forces intel to provide an interface for its cpus in next 6 years. That would be a big opportunity for nvidia to design chipset especially for overclockers.

A PCIe interface... nVidia is pretty much out of the chipset business.

[savantu]

Haha well, it looks like you're in for a world of hurt also. btw, how are those 32nm issues at intel coming along? Surely they've got it all worked out by now. No? Oh.

Might share with us your insights about Intel's 32nm problems ? I'm sure Intel's management would love to have such problems ( which lead to 60%+ gross margins ) all the time.

[570091D]

Intel is about to get athlon'd

sig'd

If I remember correctly, an Intel-FTC settlement forces intel to provide an interface for its cpus in next 6 years. That would be a big opportunity for nvidia to design chipset especially for overclockers.

the clock generator is now on-die with SB so intel can lock it down and no-one can touch it. the free performance upgrade was fun while it lasted.

[terrace215]

BD will bring a bigger jump in perf than nehalem to sb.

Considering how far behind Nehalem Phenom II single-thread IPC is, it now looks doubtful that they will even narrow the gap by much.

Prediction: As a client processor, BD is basically dead. The core performance that the turboless SB sample showed at Anandtech is just too strong. Server & HPC it should do ok, modulo GloFo's 32nm process continuing to flounder.

[slaveondope]

Considering how far behind Nehalem Phenom II single-thread IPC is, it now looks doubtful that they will even narrow the gap by much.

On the Llano side most likely yes, on the BullDozer side its pointless to speculate at this point. BD is completely different from Phenom II as you know. Its anyones guess what and when it will bring to the show.

[FlanK3r]

guys, its no corect good full test of SB, but "pro Intel" marketing preview...No atack to SB, it will good CPU, but now its only Intels reaction at Bulldozer technology dates from Hotchips