Bulldozer has 33% less ALUs than K10

[cheesehead]

I'm quite happy to see that AMD's products are competitive for the server market. However, for everything I use a computer for, they don't compete well with anything beyond Intel's low-end products. I would imagine that the consumer/desktop market is not quite as lucrative as server/enterprise sales, but even so, it would be nice to have some real competition to Intel's product line. Especially for laptops - as far as I can tell, the only way to get the new core i5 mobile processors comparable in terms of power use to the C2Ds they replaced is to slow them down to the point where there's no reason to upgrade.

Or you could produce some high-end ARM products and make all us Linux fanboys really, really happy.

 

[Accord99]

There are two different design philosophies, big cores with HT or lots of smaller cores. Lots of smaller cores scale better under heavy loads. Bigger cores with HT is better if you have more cores than threads (i.e. single threaded apps.)

You forgot one more, lots of smaller cores with multi-threading represented by the Sun Niagara processors.

In late 2007, a single 1.417 GHz Sun T2 CPU with 8 cores and 8 threads per core achieved 78.5 in SPECint_rate2006. By comparison, the contemporary Opteron 2356 quad-core at 2.3GHz scores 52.3-52.4.

Now that's what I call a throughput oriented CPU but then, where is Sun today?

This is not a troll, it is a simple fact of life. Both of these processors are giving you 12 total threads, the AMD is giving you slightly higher performance. The Intel is 42% more expensive.

Or its Intel feeling that AMD is not equivalent in terms of performance and/or features... Or it's AMD feeling the same way, thus the pricing difference.

But you can't say that it is more cost-effective if the same level of performance is 42% more expensive. That just doesn't work.

Which is not what I said. I said it's clearly the most cost-effective way to increase throughput in terms of power use and die size, which is why it's found on high end CPUs like the Power series all the way down to console CPUs and Atoms.

 

[Idontcare]

You make price/performance comparisons with the apps that are relevant to you.

I don't care if they compare virtual core count to bulldozer module count, or cores vs cores, or threads vs threads...it all comes down to what performance at what cost (and cost can include platform costs as well as power consumption costs if those metrics are of relevance to the consumer as well).

I'm sorry, but I have to say this paragraph questions the intelligence of your post.

My having posted a comment regarding the use of price/performance comparisons calls into question the intelligence of my post?

 

[JFAMD]

Intel can charge whatever they want. pricing isn't really based so much on how much it cost them to produce a CPU, it is more based on historical pricing and how much they can get out of people. (AMD does the same thing, the reason they release similar performance at lower prices is because they have to give people a good reason to switch over.)

Intel's Top-o-the-line cpu costs just as much to produce as the bottom-o-the-line cpu (using the same architecture).

The point I was trying to make was that people are arguing about something being more cost effective, but that is not borne out in the price that they are charging.

If something was less expensive, you can argue that the manufacturing was more cost effective, and the evidence is the lower price. When the price is higher it is more difficult to argue that the manufacturing is more cost effective. It may be, but it is less clear.

Even if you are an intel fan, you should want to pay less for your processors if they cost less to manufacture.

 

[Vesku]

My having posted a comment regarding the use of price/performance comparisons calls into question the intelligence of my post?

Yeah, I'm not sure what they are going on about there, IDC. Seems to me that most people and organizations go with total package price/performance + support etc. Price is no object performance purchasing is definitely an outlier. Although brand perception goes into the performance side of the equation essentially, offsetting price.

 

[Cogman]

The point I was trying to make was that people are arguing about something being more cost effective, but that is not borne out in the price that they are charging.

If something was less expensive, you can argue that the manufacturing was more cost effective, and the evidence is the lower price. When the price is higher it is more difficult to argue that the manufacturing is more cost effective. It may be, but it is less clear.

Even if you are an intel fan, you should want to pay less for your processors if they cost less to manufacture.

The cost of manufacturing does not translate into the the price of a cpu. I believe I read somewhere that it was around $10 to physically manufacture a single cpu.

This should be evident by the fact that over the years, the transistor count has been increasing exponentially, yet, the cost of CPUs has been steady, if not in a decline.

R&D and tool purchasing/creation are the two most costly parts of CPU development, and it still doesn't translate into the price of a cpu.

Intel and AMD sell their cpus at or near ideal price margins. IE, they try and price them as high as people are willing to pay without scaring away too many people. Intel could halve all their prices, but it would double their sales in the long run (which is what they care about). If however, they believed that half pricing would result in more then 2x sales + enough to cover the extra overhead of making more CPUs (which isn't that big of portion of the cost. modern bottom-of-the-line cpus sell around $30, which indicates roughly the total overhead), then you can bet they would half it in a heart beat.

The process that makes intels $1000 CPU is the same one that produces their $200-$300 cpu. They come from the same wafer, they are just binned and cut according to what they can handle/what is demanded. My point is that the prices of the CPU does not reflect the manufacturing cost in the least.

 

[scooterlibby]

I'm sorry, but I have to say this paragraph questions the intelligence of your post.

My having posted a comment regarding the use of price/performance comparisons calls into question the intelligence of my post?

Hmm based off of a careful and literal interpretation of the poster's wording, I believe they mean that the paragraph (written by you) submits queries to your post (as a whole) regarding your post's intelligence. This, also, does not make sense.

 

[Idontcare]

The cost of manufacturing does not translate into the the price of a cpu. I believe I read somewhere that it was around $10 to physically manufacture a single cpu.

http://forums.anandtech.com/showpost.php?p=28800897&postcount=38

 

[Cogman]

http://forums.anandtech.com/showpost.php?p=28800897&postcount=38

Thanks, That wasn't were I read it (it was a while ago) but it was still an interesting read that pretty much confirms what I was saying.

 

[JFAMD]

The cost of manufacturing does not translate into the the price of a cpu. I believe I read somewhere that it was around $10 to physically manufacture a single cpu.

Server processors are significantly more than $10 to make.

 

[Cogman]

Server processors are significantly more than $10 to make.

I suggest you read IDC's link above. the process for making a server processor vs desktop processor isn't significantly different. (other then perhaps stricter binning)

 

[Markfw]

I suggest you read IDC's link above. the process for making a server processor vs desktop processor isn't significantly different. (other then perhaps stricter binning)

Not going to argue with anybody on this, except 95% of the cost of any processor(Intel or AMD), is R&D. THEN comes the binning.

 

[Cogman]

Not going to argue with anybody on this, except 95% of the cost of any processor(Intel or AMD), is R&D. THEN comes the binning.

my point being that the actual process of creating a server CPU vs a desktop CPU isn't so vastly different that the ~$10 manufacturing cost wouldn't carry over.

 

[Cogman]

Server processors are significantly more than $10 to make.

Just to clarify, I'm talking about the actual physical processing of making the chip, not the R&D that goes into it.

 

[Idontcare]

Just to clarify, I'm talking about the actual physical processing of making the chip, not the R&D that goes into it.

Yep, there's production cost (the topic of the post I linked), then there's R&D cost (which is very small for the chips currently in production), and then there's SG&A.

For Q1 2010 AMD's R&D expense was $323m (20% of gross revenue), SG&A was $219m (14% of gross revenue).

For Q1 2010 Intel's R&D expense was $1,564m (15% of gross revenue), SG&A was $1,514m (15% of gross revenue).

 

[Schmide]

Lets not forget Intellectual Property Purchasing and Licensing, most of the value of any product is beyond the physical and logistical.

 

[GaiaHunter]

The pricing is nearly entirely determined by its competitive position. AMD is not pricing Phenom II X6's at $300 for top SKU because they want to or "they want to be nice to the customers" they know that's what makes the chip competitive with the i7's.

Thuban=346mm2
Nehalem=263mm2

You are PERFECTLY aware that bigger die sizes generally result in greater defects and greater costs.

Though its questionable whether SMT on Nehalem is cost effective compared to other MT solutions since cores don't take too much of the die, the truth is SMT is one of the most effective ways of increasing performance.

Yeah but what if you can add another physical thread in the same die size package?

Imagine if in the same 346mm^2 you had 12 physical cores opposed to 4c/8t or 6c/12?

Do people care if ATi has 1600 shaders (or 320 if you prefer that metric) vs 480 of NVIDIA? Do even people care about clock speeds of NVIDIA architectures vs clock speeds of ATi architectures?

Of course not. They care about price, performance, consumption, features, noise.

AMD has several problems vs Intel.

One of them is that their cores are less efficient on a clock per clock base on single threads.

And you can forget HT - if the Phenom II family had more performance on a single thread than i7, people would go after the Phenom II unless they really had an use for the logical cores and it would still be a trade.

I bet most people buy i7 over Phenom II because of single thread performance, not because of HT.

A second AMD problem is, as you've, put die size - not only AMD processors are slower on a clock per clock basis, they are bigger.

It makes complete sense to reduce the die size.

What if you can have the power of 4 cores in the package size of a current dual core? And 8 cores in the package size of a current quad core?

Maybe in the future we will have an AMD octo-core fighting an Intel quad-core in absolute performance. But what if the octo-core is around the same speed in single threads (and threads 1-4) and faster on threads 5-8, at around the same die size of the Intel offering?

Will then Intel be able to control the price just because it has a 6/8 core with HT? Sure the 6/8 core 12/16 threads will be faster in threads 9-16, but how many desktop users need that many threads? And in the business market you can just drop multiple processors anyway.

This discussion is a bit pointless as we don't have the relevant data about BD and SB: we don't know single thread performance, we don't know die size, we don't know power usage, we don't know OC headroom and we don't know price (which will be dependant on many of the previous variables).

Maybe we will need a new metric to compare different company products or maybe we will simply have to do as you do in GPUs - you just compare by performance, price/performance and performance/watt.

 

[JFAMD]

There is additional testing/validation, there are packaging differences, there are other differences that I can't get into.

Talking about the pure silicon difference being the same is only talking about one component of the cost.

Just like saying wine and grape juice are both made from grapes, but they should cost the same.

 

[Cogman]

There is additional testing/validation, there are packaging differences, there are other differences that I can't get into.

Talking about the pure silicon difference being the same is only talking about one component of the cost.

Just like saying wine and grape juice are both made from grapes, but they should cost the same.

That is, however, the component that you were listing for cost effectiveness and lower prices. The testing, validation, and packaging differences are all after the fact of CPU production. putting 1000 ALUs on the chip vs putting 100 ALUs on the chip isn't going to significantly affect the cost of the CPU, just like putting hyper threading on a CPU vs not isn't going to significantly affect the cost of the CPU. Desktop or server.

 

[Idontcare]

There is additional testing/validation, there are packaging differences, there are other differences that I can't get into.

Might those differences account for, to a certain extent, the pricing disparity between the AMD and Intel server processors you listed in your post above?

Intel doesn't want a Barcelona B2 stepping fiasco either...might be costing them a bit more to reduce the risk of such things and as such there is a benefit transferred to their customers by way of risk reduction.

 

[Phynaz]

Might those differences account for, to a certain extent, the pricing disparity between the AMD and Intel server processors you listed in your post above?

Intel doesn't want a Barcelona B2 stepping fiasco either...might be costing them a bit more to reduce the risk of such things and as such there is a benefit transferred to their customers by way of risk reduction.

Ouch!

That wasn't directed to me and it still smarted.

 

[JFAMD]

Might those differences account for, to a certain extent, the pricing disparity between the AMD and Intel server processors you listed in your post above?

Intel doesn't want a Barcelona B2 stepping fiasco either...might be costing them a bit more to reduce the risk of such things and as such there is a benefit transferred to their customers by way of risk reduction.

The point I was making was relative to our client products. I know our costs, I don't know intel's costs. I am willing to bet that our costs for server processors are in line with Intel (process node to process node). They will have some savings based on volume.

And in case you were wondering, Intel had a TLB issue at the exact same time as AMD, they just chose to ship the processor with the errata. So in terms of "transferring risk to customers," we didn't transfer risk because we didn't ship. However, if you want to accuse anyone of "transferring risk", it's the other guys.

 

[Edgy]

So all the BD tech previews/discussion suggests AMD may have new tech in BD to compete against HT. That's great.

Can AMD's BD version of Phenom X4 compete against Intel's SB version of Ci7 Quad in other areas - areas where AMD's not very competitive now - such as IPC, single threaded performance, and higher potential for overclock on decent air cooling etc.,?

OP's concern is with merit even if the cause for the concern - the article and its predicted count of less ALU per "core" - may not be the ideal metric based on which we can reliably forecast BD's IPC performance in comparison to PH II or Intel's SB.

After reading through BD info on Anand, I'm forced to admit that if AMD relies on eventual load-sharing of FP operations with on-die GPU core, then BD may be, at least initially, designed without much single threaded performance or IPC improvements over PII let alone Intel's SB..

At least until a version of BD with GPU on die comes along which can prove that FP resources on GPU can effectively "improve" the overall IPC and single threaded performance seamlessly to render the architecture competitive.

 

[IntelUser2000]

My having posted a comment regarding the use of price/performance comparisons calls into question the intelligence of my post?

I could have been more direct, but it would be pushing it. I admit it did sound ambiguous, which was a result of that.

It's not about price/performance. It's about manufacturing cost. And the post totally ignores it, which is fine in terms of the consumer, NOT to the company. While it is true the manufacturing costs seem very little, few % difference in overall profit could well turn net profits into black to red. The ASPs on Intel processors are less than $100, and AMD is even less than that.

And while we are at it, we aren't talking about CPUs with die sizes that are well within what the manufacturers call it "sweet spot", 263mm2 and 346mm2 are WELL beyond that.