Rumour: Bulldozer 50% Faster than Core i7 and Phenom II.

[Topweasel]

Pffft... So with 22nm SB-E/Ci7s coming out the end of this year - AMD will be relegated to good bang/buck category as usual instead of producing 'enthusiast' chips.

Why. That's not even enthusiast pricing. AMD can still have a performance platform without having selling a minimum $600 CPU with a $250-$300 Motherboard. The market for the SB-E is relatively small and exists purely give Intel some balance and a location to offload surplus server chips.

 

[inf64]

Pffft... So with 22nm SB-E/Ci7s coming out the end of this year - AMD will be relegated to good bang/buck category as usual instead of producing 'enthusiast' chips.

1.SB-E is not 22nm CPU.It's 8-core(native;6 core via disabled 2 cores) 32nm 400mm^2 behemoth with slightly upped clock speed for 4 core part(3.6GHz) and slightly downgraded clock speed for 6 core part(3.3Ghz). Versus SB 2600K it will offer better performance only in applications that scale very good over 8 threads-this means mostly rendering and video encoding. Versus 990x it will offer just a minor performance increase,think 10-15% tops in MT applications.

2. Ivy Bridge,the first 22nm intel part,is officially delayed to March or April 2012,just in time for Trinity APU(800GFLOPs of potential FP power- 600SP 6850-like GPU based on VLIW4 ,most likely with 3ch. memory controller and 4 Enhanced Bulldozer cores). So not an "easy" fight as it is Vs Llano's K10 cores.iGPU will be no battle,Trinity,maybe even Llano,will simply outclass IB's GPU.

 

[Lonbjerg]

1.SB-E is not 22nm CPU.It's 8-core(native;6 core via disabled 2 cores) 32nm 400mm^2 behemoth with slightly upped clock speed for 4 core part(3.6GHz) and slightly downgraded clock speed for 6 core part(3.3Ghz). Versus SB 2600K it will offer better performance only in applications that scale very good over 8 threads-this means mostly rendering and video encoding. Versus 990x it will offer just a minor performance increase,think 10-15% tops in MT applications.

2. Ivy Bridge,the first 22nm intel part,is officially delayed to March or April 2012,just in time for Trinity APU(800GFLOPs of potential FP power- 600SP 6850-like GPU based on VLIW4 ,most likely with 3ch. memory controller and 4 Enhanced Bulldozer cores). So not an "easy" fight as it is Vs Llano's K10 cores.iGPU will be no battle,Trinity,maybe even Llano,will simply outclass IB's GPU.

Talking about "outclassing" in IGP/APU/*other useless crap* made me ROFL, thank you for that.

 

[Tuna-Fish]

Key word is almost. Supposedly, Bulldozer ver. 1 is going to be physically compatible to AM3, even though AM3+ adds pins. If Bulldozer ver. 2 is using yet another socket, why bother with AM3+ at all, if those pins are never going to be used?

The primary difference seems to be that AM3+ has a beefier pins to supply power to the chip.

This makes sense -- as vcore goes down, and wattage stays stable, you need more pins to survive the increased current.

 

[inf64]

Talking about "outclassing" in IGP/APU/*other useless crap* made me ROFL, thank you for that.

Funny thing that intel is updating their iGPU with IB.If it is "useless crap",they should just stick with old GMA and be "fine". Oh wait,they will sell zero IB APUs if they do that.With Nehalem-like per core performance(Trinity),who cares for CPU performance anymore? It is so fast that any regular/average Joe will notice zero difference between two products in conventional x86 workloads.In games or GPGPU optimized software on the other hand,they will easily notice 2x or 3x better performance.No brainier for smart consumer.

 

[Ajay]

1.SB-E is not 22nm CPU.It's 8-core(native;6 core via disabled 2 cores) 32nm 400mm^2 behemoth with slightly upped clock speed for 4 core part(3.6GHz) and slightly downgraded clock speed for 6 core part(3.3Ghz). Versus SB 2600K it will offer better performance only in applications that scale very good over 8 threads-this means mostly rendering and video encoding. Versus 990x it will offer just a minor performance increase,think 10-15% tops in MT applications.

My Bad, forgot IB was the first Intel chip on 22nm. D'oh!

I thought that SB was 10-15% faster than Nehalem in single threaded apps. Assuming no improvements in multi-threaded apps, I guess that performance estimate makes sense. Curious though, that Intel is dropping the top speed bin for 6 core chips.

 

[Lonbjerg]

Funny thing that intel is updating their iGPU with IB.If it is "useless crap",they should just stick with old GMA and be "fine". Oh wait,they will sell zero IB APUs if they do that.With Nehalem-like per core performance(Trinity),who cares for CPU performance anymore? It is so fast that any regular/average Joe will notice zero difference between two products in conventional x86 workloads.In games or GPGPU optimized software on the other hand,they will easily notice 2x or 3x better performance.No brainier for smart consumer.

Funny thing, GPU's don't stand still either.

Neither does games.

The world isn't static and thus "APU" is just PR hype for "IGP"...and it will always be a joke compard to a dedicated GPU.

Seriously it can't even beat a GT240...lay off the coolaid.

 

[inf64]

My Bad, forgot IB was the first Intel chip on 22nm. D'oh!

I thought that SB was 10-15% faster than Nehalem in single threaded apps. Assuming no improvements in multi-threaded apps, I guess that performance estimate makes sense. Curious though, that Intel is dropping the top speed bin for 6 core chips.

It does offer some IPC improvement Vs Nehalem,but the point is it runs at lower clock speed to begin with : 990x @ 3.46Ghz Vs top end SB-E 6C @ 3.3Ghz. So some of the improvement is negated by clock difference and some is negated by poor support for multiple cores in today's desktop environment.50% more cores,on desktop, bring around 16% more performance in the case of 980x vs i7-975(same clock but 50% more cache and 50% cores). Since IPC improvement of SB-E is already figured in SB results(vs 1st gen i7),we can expect around 16% better results with 3% correction(3.4Ghz/3.3Ghz) versus 2600K,or in numbers : "211.1pts" x1.16x0.97~=237pts. For those who may argue SB-E has more cache than SB(2600K),that's true but remember that 16% increase Westemere6C sees versus i7 975 is counting in 50% core increase AND 50% (L3)cache increase,so it is also figured in the calculation. SB-E will indeed have even more than 50% of L3 cache on board,but we have no idea at what clock the cache will run nor what kind of effect it will have.It is effectively 25% more cache per core than what Wetsmere got versus Nehalem,so I think it may result in like ~5% better results at same clock("248pts" in above linked chart of an average desktop performance).

So in the end SB-E @ 3.3Ghz can be around 16-18% faster than SB 2600K,or 12% faster than Westmere 6C @ 3.46Ghz.

Bulldozer on the other hand is designed to offer 30-50% more performance in the same TDP envelope as Thuban 6C design @ 2:22 time mark.That is probably a range of improvements,lowest being in single thread,highest being in MT workloads. Apply the geometric mean of that (~39%) and you should land above Westmere 6C @ 3.3Ghz,at around 228pts in linked chart. That is slightly below SB-E.

Funny thing, GPU's don't stand still either.

Neither does games.

The world isn't static and thus "APU" is just PR hype for "IGP"...and it will always be a joke compard to a dedicated GPU.

Seriously it can't even beat a GT240...lay off the coolaid.

More facts less fanboism please.
Read the updated chart AT posted(with 1600 and 1866MHz memory). Also read the page 3 of the preview and see how Sumo GPU(inside Llano) with DDR3 1866Mhz memory does versus desktop 5570 with gddr5. It comes very close,despite the fact that GPU core inside Llano works at 600Mhz,or 8% lower.On average,Llano's GPU paired with 1866Mhz DDR3 is 18% slower in AT benchmarks than discrete 5570.Factor in 8% GPU clock difference and you get 9% difference which can be attributed to gddr5 on discrete part.That is almost exactly the level of discrete card performance but without dedicated RAM or fancy huge L3 cache. Also note that 5570 is barely faster than GT 240 too,does it make a "fail" discrete GPU too? No,of course not.

 

[GaiaHunter]

More facts less fanboism please.
Read the updated chart AT posted(with 1600 and 1866MHz memory). Also read the page 3 of the preview and see how Sumo GPU(inside Llano) with DDR3 1866Mhz memory does versus desktop 5570 with gddr5. It comes very close,despite the fact that GPU core inside Llano works at 600Mhz,or 8% lower.On average,Llano's GPU paired with 1866Mhz DDR3 is 18% slower in AT benchmarks than discrete 5570.Factor in 8% GPU clock difference and you get 9% difference which can be attributed to gddr5 on discrete part.That is almost exactly the level of discrete card performance but without dedicated RAM or fancy huge L3 cache. Also note that 5570 is barely faster than GT 240 too,does it make a "fail" discrete GPU too? No,of course not.

2 corrections:

1st) AT 5570 is running at 750 MHz (unless it is a typo) - it isn't a stock 5570, so actually it is 25% lower.

2nd) 5570 generally use DDR3.

 

[inf64]

2 corrections:

1st) AT 5570 is running at 750 MHz (unless it is a typo) - it isn't a stock 5570, so actually it is 25% lower.

2nd) 5570 generally use DDR3.

Thanks ,I missed the specs of 5570 in his review so I assumed it was default clocked(650Mhz) part with gddr5.
With this in mind,Llano performs actually on par with 5570 w/DDR3 memory.Quite an achievement for a shared IMC part.

 

[Ajay]

Bulldozer on the other hand is designed to offer 30-50% more performance in the same TDP envelope as Thuban 6C design @ 2:22 time mark.That is probably a range of improvements,lowest being in single thread,highest being in MT workloads. Apply the geometric mean of that (~39%) and you should land above Westmere 6C @ 3.3Ghz,at around 228pts in linked chart. That is slightly below SB-E.

Well, I certainly hope the new stepping delivers that level of performance - and hopefully overclocks well. If so, then I take back what I said. The video makes it clear that AMD intends to make this a high performance part at a decent price. As a folder, it will be very interesting to see what kind of price/performance an overclocked bulldozer delivers in SMP F@H. Sure, an overclocked Gulftown kicks @ss - but at a steep price. I won't be interested in increasing my folding power till 2H12 (I'm buying a new home at the moment), but maybe Komodo will be the way to go.

At least my hopes are back up that we will have interesting options in the next couple of years.

 

[Lonbjerg]

More facts less fanboism please.

Wake me when it actully is something else than a renamed IGP that wants to be a GPU when it grows up... :thumbsdown:

 

[Idontcare]

Bulldozer on the other hand is designed to offer 30-50% more performance in the same TDP envelope as Thuban 6C design @ 2:22 time mark.That is probably a range of improvements,lowest being in single thread,highest being in MT workloads.

Wouldn't we expect 30-50% more performance within the same TDP footprint for even a 32nm Thuban?

Going from 45nm-based doped-poly gates and nitrided gate-ox to 32nm-based HKMG and shrinking even the same design (putting the gained thermal budget towards amping up the clockspeeds and/or boosting core-count) should enable a Stars core CPU a 30-50% performance boost IMO.

(since we aren't talking about IPC, just performance/watt)

 

[piesquared]

1.SB-E is not 22nm CPU.It's 8-core(native;6 core via disabled 2 cores) 32nm 400mm^2 behemoth with slightly upped clock speed for 4 core part(3.6GHz) and slightly downgraded clock speed for 6 core part(3.3Ghz). Versus SB 2600K it will offer better performance only in applications that scale very good over 8 threads-this means mostly rendering and video encoding. Versus 990x it will offer just a minor performance increase,think 10-15% tops in MT applications.

2. Ivy Bridge,the first 22nm intel part,is officially delayed to March or April 2012,just in time for Trinity APU(800GFLOPs of potential FP power- 600SP 6850-like GPU based on VLIW4 ,most likely with 3ch. memory controller and 4 Enhanced Bulldozer cores). So not an "easy" fight as it is Vs Llano's K10 cores.iGPU will be no battle,Trinity,maybe even Llano,will simply outclass IB's GPU.

Not doubting you on the bolded part, in fact you are probably right. I just have one doubt because intel says they only make chips with all the cores working.

 

[Topweasel]

Not doubting you on the bolded part, in fact you are probably right. I just have one doubt because intel says they only make chips with all the cores working.

They won't remark defective chips after the Pentium bug. But that doesn't mean that they won't disable features and a cores to fit a given profile at a targeted price. Ton's of examples of them disabling cache's for Celerons through their life and back in the day with the Pentium D they had the Prescott cores linked in the die and literally broke off a defective CPU and sold the good one as a single core option.

 

[piesquared]

They won't remark defective chips after the Pentium bug. But that doesn't mean that they won't disable features and a cores to fit a given profile at a targeted price. Ton's of examples of them disabling cache's for Celerons through their life and back in the day with the Pentium D they had the Prescott cores linked in the die and literally broke off a defective CPU and sold the good one as a single core option.

Hmm ok. It's strange then that tortelini said, when refering to AMD's X3, that they made chips with all the cores working.

 

[Genx87]

Funny thing that intel is updating their iGPU with IB.If it is "useless crap",they should just stick with old GMA and be "fine". Oh wait,they will sell zero IB APUs if they do that.With Nehalem-like per core performance(Trinity),who cares for CPU performance anymore? It is so fast that any regular/average Joe will notice zero difference between two products in conventional x86 workloads.In games or GPGPU optimized software on the other hand,they will easily notice 2x or 3x better performance.No brainier for smart consumer.

If CPU means nothing why is AMD bringing out BD? If GPU means so much why not have a discrete GPU that offers far more performance? We just making up things out of thin air?

 

[Genx87]

More facts less fanboism please.
Read the updated chart AT posted(with 1600 and 1866MHz memory). Also read the page 3 of the preview and see how Sumo GPU(inside Llano) with DDR3 1866Mhz memory does versus desktop 5570 with gddr5. It comes very close,despite the fact that GPU core inside Llano works at 600Mhz,or 8% lower.On average,Llano's GPU paired with 1866Mhz DDR3 is 18% slower in AT benchmarks than discrete 5570.Factor in 8% GPU clock difference and you get 9% difference which can be attributed to gddr5 on discrete part.That is almost exactly the level of discrete card performance but without dedicated RAM or fancy huge L3 cache. Also note that 5570 is barely faster than GT 240 too,does it make a "fail" discrete GPU too? No,of course not.

So you believe OEMs will ship Llano based platforms with higher end ram? Why? This is a budget platform designed to up integrated graphics performance. And Trinity will be out when? 2013? That will give us 2 more generations of GPU's to cram down to the low end.

 

[Mopetar]

I believe Trinity is schedule for 2012. It's using the same 32nm process, but changing out the Stars cores for Bulldozer cores, so I shouldn't expect it to be extensively delayed.

 

[piesquared]

Yes, AMD has said about 100 times at least that Trinity is scheduled for 2012. As it's on the same 32nm process that Llano has done the legwork for, and since they've had silicon back from the fabs for a month, and they've demoed it at AFDS, mid 2012 looks to be the latest it will launch. Apparantely Anand made a remark in his article stating Trinity 2012-2013, no surprise that he would say that, but everyone else under the sun including AMD is saying 2012.

 

[OCGuy]

Yes, AMD has said about 100 times at least that Trinity is scheduled for 2012. As it's on the same 32nm process that Llano has done the legwork for, and since they've had silicon back from the fabs for a month, and they've demoed it at AFDS, mid 2012 looks to be the latest it will launch. Apparantely Anand made a remark in his article stating Trinity 2012-2013, no surprise that he would say that, but everyone else under the sun including AMD is saying 2012.

People actually trust AMD for launch targets now? Haha.

I would trust a magic 8-ball before I trusted anything AMD said about CPU roadmaps.

 

[RobertPters77]

People actually trust AMD for launch targets now? Haha.

I would trust a magic 8-ball before I trusted anything AMD said about CPU roadmaps.

Sadly true.

 

[podspi]

Well, it SEEMS like if they actual working samples of Trinity, so as far as we can tell it is at least as far along as IB ()

 

[Topweasel]

Hmm ok. It's strange then that tortelini said, when refering to AMD's X3, that they made chips with all the cores working.

Its just word play. He didn't say that they never had a unit sold with transistors not in use. They just apply a pass/fail policy to the CPU's. If it fails for any reason its disposed of. But that has and won't stop them from disabling functionality to drive sales of higher cost chips. Hell look at HT, every single CPU they make has the circuitry for that yet only a few have the functionality enabled.

 

[Morg.]

Funny thing, GPU's don't stand still either.

Neither does games.

The world isn't static and thus "APU" is just PR hype for "IGP"...and it will always be a joke compard to a dedicated GPU.

Seriously it can't even beat a GT240...lay off the coolaid.

You seem to misunderstand one fundamental aspect of the market :
Mostly everyone is not like you.
People play on crappy consoles that are almost as powerful as the e350 Zacate.
Not everyone finds it fun to spend 600+ bucks on a PC.
The APU / SoC approach is leading all the handheld/tablet market, and it's reaching into the notebook market for now, including nettops.

What makes it impossible for you to imagine that most people will stop buying big expensive towers when they don't even use them ?

And while Llano is not "the shit", it's obviously making games like Crysis and Metro playable on an SoC ... which even if you and I don't quite care, is awesome.