In defense of "Bulldozer"

[Chiropteran]

You were linked to page 10. Look at the charts on pages 11-12.

Okay, fascinating. In whatever this benchmark is supposed to show, the Athlon II X4 645 is beating i7-870, i7-920, all other AMD CPUs. It doesn't really seem to be representative of any real world performance or anything any buyers care about.

In this particular case the 1100T uses less power and performance better than the FX-8150. Too bad the year old $99 Athlon II X4 645 puts them both to shame.

what are people with an 1100t going to do?

Same thing the people with i7-980x did when sandy bridge came out? It's not really that absurd for the previous generation top CPU to remain competitive with the next generation.

 

[bononos]

......

I'm not sure what I am supposed to see here. I see higher power consumption, but I don't see the worse performance- I don't see any performance figures for this test. I am not about to read through the entire review to find it either.

You're supposed to see that bulldozer is an odd cat. Slow when its supposed to be fast, colder when its supposed to put out more heat... Your bolded portion of my post actually says bulldozer is cooler than the 1100T running linX.

Any comments about the lostcircuits link? Your lofty challenge about no benchmark showing bulldozer to be slower and hotter has been met. Bulldozer was shown to be significantly slower and 20% hotter.

 

[Tanclearas]

I have to echo sentiments that defending BD is wasted effort. I am probably one of the few enthusiasts where my particular enthusiastic interest is one that BD might do well at (running virtual machines), and even I am going to take a pass on BD.

I haven't really searched out reviews where they tested BD for virtual machines, but Thuban seems like a significantly better option for me (already have an AM3+ board). A 1090T is $180 CDN versus $220 to $230 for an fx8120 or $260 to $280 for fx8150. Even factoring in those benchmarks where BD "wins", there aren't nearly enough scenarios where BD wins by enough to justify anywhere near a 40% price premium (it's actually higher than 40% for fx8150 over 1090T).

 

[Chiropteran]

there aren't nearly enough scenarios where BD wins by enough to justify anywhere near a 40% price premium (it's actually higher than 40% for fx8150 over 1090T).

You can't buy an FX-8150 for $279.99

http://www.newegg.com/Product/Produc...82E16819103960

AMD set the correct price, or possibly a price too low, for it's initial batch.

 

[Edrick]

AMD set the correct price, or possibly a price too low, for it's initial batch.

Are you trying to say that you feel that AMD should have charged more for BD?

 

[Chiropteran]

Are you trying to say that you feel that AMD should have charged more for BD?

I'm saying that if you have a product in which demand exceed supply, it would be idiotic to lower the price. Raising the price is debatable, with a higher price demand might have fallen too much. All I am saying is that the price isn't (wasn't) too high, not for a company that needs to make all the money it can.

 

[Edrick]

product in which demand exceed supply

Just because online retailers are sold out of these CPU's does not mean demand was so great. No one really knows how many each retailer received from AMD. For all we know, it could have been less than 100. AMD has done this before with their GPUs, introduce a product with very limited supply and then they are sold out for weeks. Nothing more than a marketing gimick to make people think these are selling like hot cakes.

 

[hooflung]

It's not amazing when something in relatively short supply gets sold out.



For the power it consumes, yes, it is. Like Intel, AMD should be providing more performance for the same or less power.. not more power.

The problem with your statement is that there aren't many benchmarks to validate any of the hate other the fact that at commodity computing it doesn't shine compared to current products. However, in every single credible review there is a bright shine of hope in this tech for the tech for current use in regards to performance for workstations and servers.

More cores = Better Virtual Machines.

VT-D doesn't exist on any K series Intel i5/i7 which is where all the comparisons to Sandy come from. So BD is a clear design win for AMD for those who are running Xen or VMWare ESX on a another partition.

The rub is the cost on power for a single CPU. However, you get 2-3 more pure VM boxes out of this chip vs an Intel or even x6. Meaning if you needed the extra VMs at once you would have needed another PC anyway... which would take more overall power... anyways.

So the whole hatred on the BD tech is so Juvenile. Thanks for keeping a unbiased head in here people. I am sure that extra minute it takes you to break the DMCA is well worth your time posting here.

 

[zsdersw]

The problem with your statement is that there aren't many benchmarks to validate any of the hate other the fact that at commodity computing it doesn't shine compared to current products. However, in every single credible review there is a bright shine of hope in this tech for the tech for current use in regards to performance for workstations and servers.

More cores = Better Virtual Machines.

VT-D doesn't exist on any K series Intel i5/i7 which is where all the comparisons to Sandy come from. So BD is a clear design win for AMD for those who are running Xen or VMWare ESX on a another partition.

The rub is the cost on power for a single CPU. However, you get 2-3 more pure VM boxes out of this chip vs an Intel or even x6. Meaning if you needed the extra VMs at once you would have needed another PC anyway... which would take more overall power... anyways.

So the whole hatred on the BD tech is so Juvenile. Thanks for keeping a unbiased head in here people. I am sure that extra minute it takes you to break the DMCA is well worth your time posting here.

The problem with your statement is that virtualization is not all or even most of what workstations/servers are used for.

 

[Tanclearas]

You can't buy an FX-8150 for $279.99

http://www.newegg.com/Product/Produc...82E16819103960

AMD set the correct price, or possibly a price too low, for it's initial batch.

This is one of the most ridiculous replies I've ever seen. You have smoked way too much my friend.

The problem with your statement is that there aren't many benchmarks to validate any of the hate other the fact that at commodity computing it doesn't shine compared to current products. However, in every single credible review there is a bright shine of hope in this tech for the tech for current use in regards to performance for workstations and servers.

More cores = Better Virtual Machines.

VT-D doesn't exist on any K series Intel i5/i7 which is where all the comparisons to Sandy come from. So BD is a clear design win for AMD for those who are running Xen or VMWare ESX on a another partition.

The rub is the cost on power for a single CPU. However, you get 2-3 more pure VM boxes out of this chip vs an Intel or even x6. Meaning if you needed the extra VMs at once you would have needed another PC anyway... which would take more overall power... anyways.

So the whole hatred on the BD tech is so Juvenile. Thanks for keeping a unbiased head in here people. I am sure that extra minute it takes you to break the DMCA is well worth your time posting here.

There aren't more "real" cores versus six-core parts. VT-d is available on PhII, so I have that already without having to switch to BD. There is no way you will get "2-3 more pure VM boxes" out of BD than an X6. 1 is about the best you could hope for.

 

[Chiropteran]

Just because online retailers are sold out of these CPU's does not mean demand was so great.

Irrelevant. You have two turds. There are 3 people willing to pay $1000 for each turd, and another 3000 people who would buy your turd for $10. What price do you price your turd at?

If you pick a value less than $1000, you are an idiot. If demand exceeds your supply at full price, reducing your price is a terrible error.

 

[Vesku]

You left out that the turd costs you $300 to produce. ;p

 

[Phynaz]

However, you get 2-3 more pure VM boxes out of this chip vs an Intel or even x6.

Where did you see this?

 

[3DVagabond]

Something tells me if AMD would've called the 4-module BD a quad core instead of an 8-core, a lot of this wouldn't have been an issue. For one thing, we wouldn't be comparing it to Thuban, which it is clearly inferior to given the price points.

The size and power consumption would have made it a fail as a quad core.

 

[Idontcare]

The size and power consumption would have made it a fail as a quad core.

Sadly, ironically, those are the reasons given for why bulldozer was pushed back from its intended 2008 release on 45nm.

I'm quite curious to see how it does on 28nm bulk HKMG. The half-node shrink might just give it the clockspeed boost and power-reduction bump it needs to make things interesting.

 

[podspi]

Sadly, ironically, those are the reasons given for why bulldozer was pushed back from its intended 2008 release on 45nm.

I'm quite curious to see how it does on 28nm bulk HKMG. The half-node shrink might just give it the clockspeed boost and power-reduction bump it needs to make things interesting.

Has it been confirmed that they're doing a 28nm shrink of ... Piledriver (I assume?) And do you happen to know if it will be GF or GL? (And built by GloFo or TSMC?!)

 

[Idontcare]

<- doesn't know jack squat

Has it been confirmed that they're doing a 28nm shrink of ... Piledriver (I assume?) And do you happen to know if it will be GF or GL? (And built by GloFo or TSMC?!)

I don't know any more than the rest of you. There's those AMD slides out there in the wild which refer to a 28nm refresh of the non-bobcat x86 stuff.

Given the questionable economic viability of GloFo pursuing development of high-performance SOI solely for the sake of one customer (AMD)...and given the undesirable effect of AMD being locked into using GloFo as their foundry so long as they (AMD) avoid bulk CMOS for advanced x86 designs...it only stands to reason they (AMD) would be divorcing themselves of relying on SOI-enabled nodes while simultaneously directing GloFo to not waste precious limited R&D $'s on developing a specialized SOI flavored 22nm node just for AMD's benefit.

(When I was at TI we had many such MOA's - memorandum of agreement - between ourselves and SUN...for example we had a High-K program for SUN at 90nm but they didn't want the associated cost and we didn't want the associated disruption to the rest of our CMOS integration in the fab, so we both agreed to shelve it in the meantime for 90nm and 65nm, focusing on improving the traditional poly-si xtor tech for both performance and cost reasons)

If such a divorce from SOI is coming at 22nm then there is no reason to not take advantage of the forking even sooner, and 28nm at TSMC makes a lot of sense in this regard because that is exactly where SUN (Oracle) is having their next-gen CPU's fabbed (so TSMC is already well invested in developing a higher performance subnode flavor to SUN, AMD could have given them (TSMC) even more motivation by signing up for it as well some years ago).

That said...there's a lot of potential fallout from this. Soitec's business model will be utterly destroyed if GloFo stops employing SOI in future nodes. And GloFo's own viability as a foundry would be all the more in question were AMD to field a 28nm high performance (flagship type) product from TSMC as being the successor to the existing 32nm product coming from GloFo.

 

[Edrick]

I'm quite curious to see how it does on 28nm bulk HKMG. The half-node shrink might just give it the clockspeed boost and power-reduction bump it needs to make things interesting.

By then it will not be competing with the 2600k any longer. It will be competing with IB and then the shadow of the pending Haswell release. Unless AMD can get a BD refresh @28nm out the door in the next 6-8 months, I see it as being too little too late once again.

 

[bryanW1995]

Just because online retailers are sold out of these CPU's does not mean demand was so great. No one really knows how many each retailer received from AMD. For all we know, it could have been less than 100. AMD has done this before with their GPUs, introduce a product with very limited supply and then they are sold out for weeks. Nothing more than a marketing gimick to make people think these are selling like hot cakes.

I'm all for the anti-BD rhetoric, but don't act like AMD is some sort of sneaky product rollout rebel here. NV, Intel, Via, Sis, etc etc etc have all over the years done exactly the same thing. And specifically with gpus, when has AMD done a paper launch? ATI did plenty of them, but since the "merger" we've seen everything from 2900 to 5970 come out with generally very good launches. In fact, in recent history NV has been more apt to pull this sort of stunt than DAAMIT, and even NV's "paper launches" have been more like soft launches with relatively limited supply that was usually replenished within a few weeks.

 

[Idontcare]

By then it will not be competing with the 2600k any longer. It will be competing with IB and then the shadow of the pending Haswell release. Unless AMD can get a BD refresh @28nm out the door in the next 6-8 months, I see it as being too little too late once again.

But it will have more cores too, right? IIRC the 10core chips are slated for 28nm.

5 BD+ modules on a half-node shrink could be quite a compelling product.

 

[bryanW1995]

<- doesn't know jack squat


I don't know any more than the rest of you. There's those AMD slides out there in the wild which refer to a 28nm refresh of the non-bobcat x86 stuff.

Given the questionable economic viability of GloFo pursuing development of high-performance SOI solely for the sake of one customer (AMD)...and given the undesirable effect of AMD being locked into using GloFo as their foundry so long as they (AMD) avoid bulk CMOS for advanced x86 designs...it only stands to reason they (AMD) would be divorcing themselves of relying on SOI-enabled nodes while simultaneously directing GloFo to not waste precious limited R&D $'s on developing a specialized SOI flavored 22nm node just for AMD's benefit.

(When I was at TI we had many such MOA's - memorandum of agreement - between ourselves and SUN...for example we had a High-K program for SUN at 90nm but they didn't want the associated cost and we didn't want the associated disruption to the rest of our CMOS integration in the fab, so we both agreed to shelve it in the meantime for 90nm and 65nm, focusing on improving the traditional poly-si xtor tech for both performance and cost reasons)

If such a divorce from SOI is coming at 22nm then there is no reason to not take advantage of the forking even sooner, and 28nm at TSMC makes a lot of sense in this regard because that is exactly where SUN (Oracle) is having their next-gen CPU's fabbed (so TSMC is already well invested in developing a higher performance subnode flavor to SUN, AMD could have given them (TSMC) even more motivation by signing up for it as well some years ago).

That said...there's a lot of potential fallout from this. Soitec's business model will be utterly destroyed if GloFo stops employing SOI in future nodes. And GloFo's own viability as a foundry would be all the more in question were AMD to field a 28nm high performance (flagship type) product from TSMC as being the successor to the existing 32nm product coming from GloFo.

Who cares about GloFo if they are the ones holding AMD back. I don't necessarily think that this is the case, but it's certain that they're not helping at least. How much performance/power has it cost BD b/c glofo couldn't handle gate last right now? I just read an article from Jan 11 in which bit-tech mentions that Ibm and GF have now decided that gate last is the way to go for 20/22 nm. What changed? Probably they saw how terrible BD was going to be and realized that they needed to make the change ASAP, even if it meant another 3-6 month delay.

But it will have more cores too, right? IIRC the 10core chips are slated for 28nm.

5 BD+ modules on a half-node shrink could be quite a compelling product.

It's only a half step shrink, and even that is just our (informed) speculation. BD will still be significantly more power-hungry than intel's lineup, and unless they can either get a big IPC bump, better software, or 6 ghz on air then they're still going to be a distant 2nd to intel's offerings.

 

[Edrick]

I'm all for the anti-BD rhetoric, but don't act like AMD is some sort of sneaky product rollout rebel here. NV, Intel, Via, Sis, etc etc etc have all over the years done exactly the same thing. And specifically with gpus, when has AMD done a paper launch? ATI did plenty of them, but since the "merger" we've seen everything from 2900 to 5970 come out with generally very good launches. In fact, in recent history NV has been more apt to pull this sort of stunt than DAAMIT, and even NV's "paper launches" have been more like soft launches with relatively limited supply that was usually replenished within a few weeks.

I never said other companies were not guilty of the same thing. I was just saying that no one knew how many AMD shipped initially. So saying they are sold out could mean anything from hundreds to thousands to tens of thousands. No one really knows. So we can not use that as an arguement as to how well BD is actually doing.

Keep in mind that there are a lot of people who hate Intel and will buy AMD just for that reason, even if it is not a great product. So after the initial rush of sales by these people, we will get a better idea of how well it sells to the non bias shoppers.

As for you statement about how AMD has been good with their GPU releases, I do recall when AMD and NV were fighting to launch their high end 40nm products back in late 2009, AMD was first, but availability was very scarce for the first few months. So while I would not call that a "paper launch" like NV, I would say that releases with limited availability may be more of a marketing gimick.

Also, when Intel released SB in Jan. of this year, they had no availability issues. And there were a lot of people who purchased those in the first few weeks. I am not saying Intel is perfect, but there is a clear difference between their latest CPU launch and AMDs latest.

 

[Edrick]

But it will have more cores too, right? IIRC the 10core chips are slated for 28nm. 5 BD+ modules on a half-node shrink could be quite a compelling product.

So now we will need applications that use 10 threads to fully appriciate BD? A lot of people on these forums (not saying you directly) feel that throwing more cores at the problem is a great fix. And that may be the case in the future, but not now and not in the near future.

BD needs its IPC increased first and foremost. And if that can happen with a die shrink to 28nm, then we will see it be more competitive.

Also, once applications start utilizing FMA, AVX, etc, then BD will start to show better than it does now. That is the problem when introducing a new uarch with new instructions. Especially when you are the first (FMA) and the biggest player in the market (Intel) does not even have it yet.

Personally, I think AMD using modules with only 1 shared FPU was not a great idea. I think that AMD not increasing IPC over their previous generation was a very poor idea. But I think the new instructions AMD introduced is a great idea and will pay out eventually. Unfortunately just not today.

 

[podspi]

Few things:

1) 28nm could either be at TSMC or GF, right? No reason to assume it is TSMC only. In fact, BD could be built at both, similar to how future Radeons will be, if rumors are to be believed. I wonder what the quality differences between the two would/will be...



2) Don't the smaller nodes outright require SOI at some point? I was under the impression 22nm/20nm was FD-SOI?


I can't imagine AMD would marry themselves to TSMC's 28nm instead of GF's, if only because AMD is a VIP to GF, and decidedly not a VIP to TSMC (behind Apple and nvidia at least).

Personally, I think AMD using modules with only 1 shared FPU was not a great idea. I think that AMD not increasing IPC over their previous generation was a very poor idea. But I think the new instructions AMD introduced is a great idea and will pay out eventually. Unfortunately just not today.

I think the shared FPUs are a great idea. Why over-provision when those resources could be dedicated to something else? The confusing bit is that BD is STILL huge. Imagine how large an 8 core BD would actually be! Clearly something went wrong, somewhere.

 

[Edrick]

The confusing bit is that BD is STILL huge. Imagine how large an 8 core BD would actually be! Clearly something went wrong, somewhere

Agreed. I thought AMD was marketing this new module system as die space saving.