OK, so first, this thread is all wrong because AMD is not making any claims on desktop performance. I can't speak to the validity of the original slide (or any subsequent charts floating around.) The slide is in AMD's template, but because I didn't make it, I can't vouch for the authenticity. It also could have been an older slide based on performance projections (template dates automatically update with today's date, so don't go off the date on the slide...)
We have said that a 16-core will have 50% more throughput than a 12-core (top bin to top bin) but that is a throughput number, not a speed number. Trying to figure out clock speed on client workloads is like trying to determine the top speed of a corvette based on the towing capacity of a suburban because they are using the same engine.
As to why client benchmarks show up at launch - it's all about the supply chain. In the server business customers make decisions today about the products they will be deploying a year from now. If you have to deploy a database server in Q1 2012 you are going to buy the best product in Q1 2012, you don't delay the project a quarter because something faster will be available. In the grand scheme of things launching your IT project on time is more important than any performance boost. So you can give performance estimates.
On the client side, if you release benchmarks, people say "don't buy now, there is something better coming in a quarter." When you are a consumer and spending your money, you're willing to wait. That messes up the supply chain for our OEMs. That causes stalled sales and returns. Think of the buyer's remorse when the typical consumer buys a PC with a 30-day return policy and then hears something great is coming around the corner on the same day he credit card bill shows up. Pack it up, return it and wait.
The reality is that less than 5% of the market is buying top bin parts (in a good quarter), but that other 95% tends to think about "timing" purchases, and that messes up the supply chain. Let's say out of the 5% you might be able to swing half over to your camp with some benchmarks early. Is it worth the 2.5% market share gain to put the 95% of this quarter's revenue at risk? Surely not.
I know you are saying "well intel did it with SB". Well, that was their choice. I can also say that just because intel does it does not make it a good business practice (and I have a long list of other things there...) Notice that they did not actually release the data, it was a leak, and the performance gain claimed in the leak did not really match the actuals as well. Maybe they leaked, maybe they didn't, but I don't think it was a good idea.
I can assure you that if there is any performance data that comes from AMD prior to launch it will come through my blog on AMD.com, it will not be leaked because I don't believe in leaks.
We have said that a 16-core will have 50% more throughput than a 12-core (top bin to top bin) but that is a throughput number, not a speed number. Trying to figure out clock speed on client workloads is like trying to determine the top speed of a corvette based on the towing capacity of a suburban because they are using the same engine.
As to why client benchmarks show up at launch - it's all about the supply chain. In the server business customers make decisions today about the products they will be deploying a year from now. If you have to deploy a database server in Q1 2012 you are going to buy the best product in Q1 2012, you don't delay the project a quarter because something faster will be available. In the grand scheme of things launching your IT project on time is more important than any performance boost. So you can give performance estimates.
On the client side, if you release benchmarks, people say "don't buy now, there is something better coming in a quarter." When you are a consumer and spending your money, you're willing to wait. That messes up the supply chain for our OEMs. That causes stalled sales and returns. Think of the buyer's remorse when the typical consumer buys a PC with a 30-day return policy and then hears something great is coming around the corner on the same day he credit card bill shows up. Pack it up, return it and wait.
The reality is that less than 5% of the market is buying top bin parts (in a good quarter), but that other 95% tends to think about "timing" purchases, and that messes up the supply chain. Let's say out of the 5% you might be able to swing half over to your camp with some benchmarks early. Is it worth the 2.5% market share gain to put the 95% of this quarter's revenue at risk? Surely not.
I know you are saying "well intel did it with SB". Well, that was their choice. I can also say that just because intel does it does not make it a good business practice (and I have a long list of other things there...) Notice that they did not actually release the data, it was a leak, and the performance gain claimed in the leak did not really match the actuals as well. Maybe they leaked, maybe they didn't, but I don't think it was a good idea.
I can assure you that if there is any performance data that comes from AMD prior to launch it will come through my blog on AMD.com, it will not be leaked because I don't believe in leaks.
