Intel states in their press material that Haswell should be ~10% faster than IB.
In order to get the throughput it offers in AVX2 and FMA3,applications need to be written with those in mind.
Intel states in their press material that Haswell should be ~10% faster than IB.That's just a silly way of stating: Unless it's AVX2, the performance increase will be minimal.
haswell IGP have a good chance to beat some low-end GPUs... even some mid-end
http://semiaccurate.com/2012/09/10/crystalwell-is-very-wide-memory-for-haswell-gt3/
haswell IGP have a good chance to beat some low-end GPUs... even some mid-end
http://semiaccurate.com/2012/09/10/crystalwell-is-very-wide-memory-for-haswell-gt3/
Intel states in their press material that Haswell should be ~10% faster than IB.
In order to get the throughput it offers in AVX2 and FMA3,applications need to be written with those in mind.
I believe HW iGPU will catch up Trinity's performance.
I'm well aware, just pointing it out to our resident AVX2 superhero. Regular users and even enthusiasts won't see anything come to fruition from the newer ISAs until at least a couple of years down the line, much like every ISA that's been tacked on. We'll get a few loudmouths who will boast synthetic benchmarks figures but nothing to show for it outside of that
It probably will, but the bigger issue here is how AMD's Kaveri will fair against Haswell on the mobile end. If it's like the current HD4000 vs. 7660G then AMD will be in trouble. Intel's on-die graphics, though far from stellar in desktop are actually quite decent in mobile. AMD's APUs are a reversal of roles in that their desktop APUs perform far better than do the mobile SKUs do to a variety of factors-- clock speeds, inefficiency at the fab level, reliance on DDR3's bus for cache, etc.
If Haswell trumps Kaveri's on-die graphics in the mobile segment then AMD will have nothing going for it on either side of the CPU... outside of drivers Intel has done better but they're still far from respectable for gaming.
Come again? SB and IB are finished designs,there is zero chance of them having either FMA or AVX2.You know what, i saw a video on Intels web site with an interview of some Israeli cpu architects, Haifa team, and they said that Intels new pipeline is modular and will be gradually introduced in 2-3 generations of cpus, that means that they could implement FMA3 and AVX2 even in SB/IB gen imho.
You know what, i saw a video on Intels web site with an interview of some Israeli cpu architects, Haifa team, and they said that Intels new pipeline is modular and will be gradually introduced in 2-3 generations of cpus, that means that they could implement FMA3 and AVX2 even in SB/IB gen imho.
At the same TDP the AMD Trinity is much faster than HD4000 in Laptops. Kaveri will be made at 28nm process and I'm expecting a nice performance/wattage increase over Trinity, especially in GPU compute and DX-11 gaming due to GCN architecture.
With HW, Intel will start to sacrifice CPU performance by raising the iGPU die size and power usage. Perhaps Socket 2011 will become the dominant enthusiast platform in the near future much like 1366 used to be.
At the same TDP the AMD Trinity is much faster than HD4000 in Laptops. Kaveri will be made at 28nm process and I'm expecting a nice performance/wattage increase over Trinity, especially in GPU compute and DX-11 gaming due to GCN architecture.
AVX2 will be supported by lots of multimedia applications on the day of Haswell's launch, or not long after. Just look at how fast SSE4 was supported, and that didn't offer anywhere near double the performance like AVX2 does! Also any application that uses OpenCL can instantly benefit from Intel's next version with AVX2 support, which will no doubt coincide with Haswell's launch. Likewise TSX can be used by the operating system so it doesn't require a major rewrite of each application to start benefiting from it.That's just a silly way of stating: Unless it's AVX2, the performance increase will be minimal.
You're wrong. AVX2 is nothing like those previous ISA extensions. It's the very first x86 SIMD extension which is truly suitable for SPMD processing. This is revolutionary because it allows for straightforward auto-vectorization. Previous extensions required developers to do the vectorization themselves, and often even required assembly programming. With AVX2 you can let the compiler take care of that, and just write ordinary scalar code. That's a huge difference which will lead to a much faster adoption.I'm well aware, just pointing it out to our resident AVX2 superhero. Regular users and even enthusiasts won't see anything come to fruition from the newer ISAs until at least a couple of years down the line, much like every ISA that's been tacked on.
^^ yes, this is huge! The only slow down then becomes regression testing in SQA.You're wrong. AVX2 is nothing like those previous ISA extensions. It's the very first x86 SIMD extension which is truly suitable for SPMD processing. This is revolutionary because it allows for straightforward auto-vectorization. Previous extensions required developers to do the vectorization themselves, and often even required assembly programming. With AVX2 you can let the compiler take care of that, and just write ordinary scalar code. That's a huge difference which will lead to a much faster adoption.
^^ yes, this is huge! The only slow down then becomes regression testing in SQA.
AVX2 will be supported by lots of multimedia applications on the day of Haswell's launch, or not long after. Just look at how fast SSE4 was supported, and that didn't offer anywhere near double the performance like AVX2 does!
The server and workstation crowd will adore AVX2, I'm sure of it. The rest of us, though, won't care nor notice.
Game development should see a rapid adoption, if Intel has done it's marketing properly. If it were Nvidia marketing AVX2, we'd never hear the end of it's benefits.
You're still not getting it. AVX sorely lacks gather support, integer support, and vector-vector shifts. Its unsuited for SPMD. And on top of that Sandy Bridge is bandwidth limited. Haswell fixes that all at once, and throws in FMA, hardware transactional memory and lock elision as well!This was the same with AVX and where has that brought us?
You are sorely mistaken. Every respectable multimedia application supports every SIMD extension up to SSE 4.1, with hand-written assembly routines. That tells me something about how much they care about improving performance and the effort they're willing to take. And yes, they're skipping AVX, because that's floating-point only. But it's a given that they'll offer AVX2 support as soon as it becomes available.You're ignoring two key points here
1 - Nobody really cares much about the potential performance benefits outside of select workstation apps and server workloads, and as is the norm with those two segments, the software is optimized and recompiled on a regular basis anyway
AVX2 offers direct profits. And developers are not too lazy to simply change a compiler flag. And while initially the market for AVX2 may be small, nobody wants to be the last. There's a very strong halo effect of being the first to support it. Benchmarks between competing software can make or break sales numbers. And they don't want to risk disappointing the early adopters who invest in the latest hardware and software.2 - Software developers are lazy and won't recompile unless they can see direct profits from it. They certainly won't recompile for a very small percentage of the market
I bought a 79 cent game for my iPhone, and a 49 dollar game for my PC. So guess where the big money is...And there's an easy way to think about this -
When was the last time you paid for software or an application and on what platform was it? PC, phone, or tablet? On what platform do most users buy their applications?
I bought a 79 cent game for my iPhone, and a 49 dollar game for my PC. So guess where the big money is...
The 17W TDP chips are pretty much neck-and-neck with each other as far as graphics performance goes.
According to here though, HD 4000 in the 17W Ivy Bridge chip is almost on par with the discrete 7550M graphics in the Trinity A6-4455M(17W) laptop while significantly outperforming the HD 7500G integrated.
That was a different era and even that was slow. If you want to relate it to more modern terms then take a look at AVX and its adoption rate. You'll see that it really hasn't been that stellar despite potential throughput increase.
For you to compare AXV with AVX2 means you clearly do not understand the difference between the 2 and what the actual benefits are to the development community.
Well boys and girls the Haswell liveblog is up! http://www.anandtech.com/show/6263/intel-haswell-architecture-disclosure-live-blog
looking good so far, name of the game is power efficiency