A lot of the development tools (compilers, etc.) already support most Haswell instructions. That is a big plus this time around.
Even if Haswell is so good, it will take years for the effect to tripple down. First the software tools need to be build. Compilers, debuggers, etc. Heck, maybe we first need to settle on a good model to work with. Then people need to learn to use those tools. Then the software has to be written and deployed. I don't think this will be done in a few years. Until that time, single-thread software and performance matters.
You are completely wrong. Haswell isn't about IPC at all. I'd be surprised if they made any significant changes on that front. Instead, Haswell is about SPMD and hardware transactional memory.
It's best to stop talking about IPC in isolation altogether, because it doesn't say anything about the effective performance for applications (just like the MHz myth). AVX2 doubles the amount of work that can be done per instruction. In the case of load operations it can even do up to eight times more work per cycle. And hardware transactional memory will enable lots of applications which previously had bad multi-core scaling behavior to synchronize tasks between cores much more efficiently. It also opens the door to automated threading by the compiler.
So unless you thoroughly understand the implications of these technologies, please don't say people's expectations are too high. These features have been fully confirmed and are eagerly awaited by professionals.
Short answer: Because it also sucks for multi-threading.If single threaded doesn't matter then why is bulldozer so horrible?
This is an obvious case of software that would benefit greatly from multi-threading, yet it has been too complex to deal with race conditions and still achieve good scaling behavior, right? Then that is something that would gain immensely from TSX. It gets rid of race conditions and lowers the locking overhead as well!I used to be a software developer.
I've worked for two of the largest router vendors. One was planning to change it's routing software to a multi-threaded model. That was years ago. They still haven't done it.
The reason we currently don't have many tools is not because of a lack of trying, but because on today's hardware any type of generalized framework for multi-threaded development would abstract things and lose performance in the process. It takes fine-tuning of each individual application to achieve good multi-core performance scaling.So we seem to agree. Not much has been done regarding tools. And thus parallel programming is up to the individual programmer to deal with.
No, it won't take that many years. Unlike previous extensions, AVX2 has been announced and fully specified long before the hardware is launched. And because there's a big incentive to make use of it, middleware developers have already started to support it. In fact for all major compilers the support is almost ready! TSX was announced more recently, but keep in mind that Haswell's launch is still a year away and a lot of tools can be written in the meantime.Even if Haswell is so good, it will take years for the effect to tripple down. First the software tools need to be build. Compilers, debuggers, etc.
Absolutely. Things like threads and locks should be abstracted into something like tasks and dependencies. That's a lot easier to grasp by human logic. And just like object-oriented programming has revolutionized how think about code and data, I'm sure that by the end of the decade we'll have an entirely different view on threading.Heck, maybe we first need to settle on a good model to work with.
There's a significant difference between "single-threaded software" and "single-threaded performance". The latter is also important for multi-threaded software, and that will remain true in the future. Single-threaded performance-critical software however is a dying breed.Your first sentence was: "Who cares about single-threaded software".
I guess my English isn't good enough to understand that sentence.
People shouldn't choose between CPUs based entirely on their day 1 use of it. We buy one that is more powerful than our current needs, so it it will last multiple years. There are plenty of threads on this forum confirming that people want the best value for their money in the long term, and this sometimes means waiting for the next generation.We both agree, I am just trying trying to be realistic about day 1 performance, which is what most people seem to be preaching about.
People shouldn't choose between CPUs based entirely on their day 1 use of it. We buy one that is more powerful than our current needs, so it it will last multiple years. There are plenty of threads on this forum confirming that people want the best value for their money in the long term, and this sometimes means waiting for the next generation.
My Nehalem is plenty fast for another year. Upgrading to Ivy Bridge would be a waste of money since it's barely faster. And even though I won't get much benefit from Haswell on the day I buy it either, I'll know that just like my quad-core Nehalem used to be underutilized when I bought it, it will offer greater satisfaction over its entire lifetime.
People shouldn't choose between CPUs based entirely on their day 1 use of it.
Who cares about single-threaded software? Expecting it to be any faster is like expecting your new car to be any faster than your old one when waiting for a traffic light.
I do hope that it works out as you say but I have my doubts on that especially for technology that changes so fast. How long do you suppose it takes to propagate to 50% of the new applications sold? 2 years? 3 years? Even that is pretty damn fast considering how many applications are out there. It might be less on servers, and on development workstations, what about laptops and desktops?Yes it takes some developer effort, but it's easily worth it. Also, once applications are properly multi-threaded and vectorized, it's fairly straightforward to scale to more cores and wider vectors.
This is doing it wrong.
....Every time I have future proofed I have regretted it. With no exceptions. It never pans out. You pay too much, get too little, and are missing on critical features ... if you even get anything. (ex: my first x64 processor was a laptop one, the laptop never got drivers for x64 versions of windows forcing it to forever run in 32bits).
Have you ever considered that it might just be that you have poor judgement about what will be the most future proof for the least amount of money? Just saying. :whiste:Every time I have future proofed I have regretted it.
Yes, technology changes fast. But there's a clear direction to it! It is no coincidence that Intel doubled the SIMD width with AVX, and it will be optimal for 256-bit SPMD with AVX2. And it's extendable up to 1024-bit. It's also no coincidence that they're adding TSX technology to improve multi-core scaling and make it more manageable for developers to deal with concurrency.I do hope that it works out as you say but I have my doubts on that especially for technology that changes so fast.
Why on earth would it have to be used by 50% of the new applications? 9 out of 10 application I use daily aren't performance critical. But I do care about that remaining 10%, and their developers are taking every reasonable effort to use the latest technology. Unfortunately today's SIMD and multi-core technology is far from optimal, but that's going to change quite radically with Haswell. The applications that will take advantage of AVX2 and TSX first, are the ones that can use a speed boost the most. So you can count on seeing a performance improvement where it matters, in a relatively short time.How long do you suppose it takes to propagate to 50% of the new applications sold? 2 years? 3 years? Even that is pretty damn fast considering how many applications are out there. It might be less on servers, and on development workstations, what about laptops and desktops?
The best way for CPU technology to move forward now is to exploit vector parallelism and thread parallelism. So it's a sure bet to put your money on Haswell.
Why on earth would it have to be used by 50% of the new applications? 9 out of 10 application I use daily aren't performance critical. But I do care about that remaining 10%, and their developers are taking every reasonable effort to use the latest technology. Unfortunately today's SIMD and multi-core technology is far from optimal, but that's going to change quite radically with Haswell. The applications that will take advantage of AVX2 and TSX first, are the ones that can use a speed boost the most. So you can count on seeing a performance improvement where it matters, in a relatively short time.
Have you ever considered that it might just be that you have poor judgement about what will be the most future proof for the least amount of money? Just saying. :whiste:
So, coming from a Phenom II, should i look at Ivy, or pick up a 2500k if the price drops?
I've got a 2500k, and unless I find a killer deal on a 3570k, I'm keeping it.
If I ever saw someone shelling out $220+ to upgrade their IGP, I'd LOL.
Yes. Left is IB, right is Haswell. GT is the IGP portion.is haswell getting a significant IGP boost from IB?
Still running a water cooled 980X here. Im fine with that for a while.
Yes. Left is IB, right is Haswell. GT is the IGP portion.