• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Thinking about next CPU purchase.

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Rifter said:
If i have to wait for haswell-E to get 6 core from haswell i'll just get Ivy-E late this year and say screw haswell.
Click to expand...
Haswell's AVX2 support doubles the throughput per core, so no need to wait for a 6-core.

The only point in getting a 6-core is to increase the throughput of parallel software by 50% in the best case. But AVX2 achieves 100% in a much more power efficient way by doubling the width of integer vector operations, providing fused multiply-add support, and last but not least parallel memory accesses.
 
I hate to praise something just cause I already have it, but the 3930K is not a small bump over nehalem. Its a pretty big deal.
 
BenchPress said:
Haswell's AVX2 support doubles the throughput per core, so no need to wait for a 6-core.

The only point in getting a 6-core is to increase the throughput of parallel software by 50% in the best case. But AVX2 achieves 100% in a much more power efficient way by doubling the width of integer vector operations, providing fused multiply-add support, and last but not least parallel memory accesses.
Click to expand...

I have read this as well, but will the software need to be recompiled to use this? for example its taking a long time for software to adapt to multi core CPU's, can we expect a similer timeline for software to be adapted to AVX2?
 
moonbogg said:
Yeah thats true. He mentioned he wanted to upgrade for no reason though to something with 6 cores, so SB-E would be the perfect useless upgrade, but not quite as useless as THIS:

http://www.youtube.com/watch?v=aqAUmgE3WyM&feature=player_embedded
Click to expand...

If it comes down to SB-E and a 970 being my only options i would go the 970 route for now and wait out broadwell. However it seems likely i will wait for haswell and settle for a 4 core and hope AVX2 takes off alot faster than i think it will.
 
Rifter said:
I have read this as well, but will the software need to be recompiled to use this? for example its taking a long time for software to adapt to multi core CPU's, can we expect a similer timeline for software to be adapted to AVX2?
Click to expand...
Yes it requires software to be recompiled. But it's not comparable to multi-threading. Making use of multiple threads requires very big changes to the software, and optimizing it properly can take a long time and takes great expertise. Hence the multi-core adoption has been slow. In contrast, AVX2 can be automatically used by the compiler to vectorize code loops. Thanks to the addition of parallel memory accesses (gather instructions) and several other vector equivalents of scalar operations this becomes straightforward. So the adoption of AVX2 will be much faster than multi-threading.

That said, Haswell also adds TSX technology, which makes multi-threading both easier and more efficient.
 
BenchPress said:
Yes it requires software to be recompiled. But it's not comparable to multi-threading. Making use of multiple threads requires very big changes to the software, and optimizing it properly can take a long time and takes great expertise. Hence the multi-core adoption has been slow. In contrast, AVX2 can be automatically used by the compiler to vectorize code loops. Thanks to the addition of parallel memory accesses (gather instructions) and several other vector equivalents of scalar operations this becomes straightforward. So the adoption of AVX2 will be much faster than multi-threading.

That said, Haswell also adds TSX technology, which makes multi-threading both easier and more efficient.
Click to expand...

Well this is good to know, so i guess as long as the software you are using is still being supported and updated then it should be an easy fix to add AVX2 support. Next question, does windows 7 support AVX2? Or will there be a operating system upgrade/patch required for the use of AVX2?
 
Rifter said:
Well this is good to know, so i guess as long as the software you are using is still being supported and updated then it should be an easy fix to add AVX2 support. Next question, does windows 7 support AVX2? Or will there be a operating system upgrade/patch required for the use of AVX2?
Click to expand...
No new O.S. patch will be required. AVX extended the vector registers to 256-bit, and AVX2 reuses those same registers. The patch was part of Windows 7 SP1 last year.
 
Based on what I've been reading, just about anything can use AVX2, not just video. The exciting part is many times the performance.
 
Rifter - What do you even do with your rig that you would need all these cores/threads? I have a very high suspicion that you wouldn't even use all the cores if you had them. I highly suggest waiting for haswell and getting that. If you want to waste money though upgrade to SB-E.
 
2is said:
Based on what I've been reading, just about anything can use AVX2, not just video.
Click to expand...
Just about anything that uses SSE2 (or which used MMX before that) can be easily rewritten (or sometimes just recompiled) to use AVX2. As with SSE2, it only works on things that can be worked with in parallel.
 
Ken g6 said:
Just about anything that uses SSE2 (or which used MMX before that) can be easily rewritten (or sometimes just recompiled) to use AVX2. As with SSE2, it only works on things that can be worked with in parallel.
Click to expand...
AVX2 is useful for more than what previously used SSE(2). The problem with all previous x86 vector extensions is that they lack instructions for parallel memory accesses and independent shifting. This meant that any scalar code loops containing dependent memory load operations or shifts could not be automatically vectorized and were difficult or impossible to successfully vectorize manually.

AVX2's addition of gather and vector-vector shift support fixes these issues and allows the CPU to use the same massively parallel programming model of the GPU, and be even more widely applicable.
 
I doubt there will be an IB-E. IB was about the GPU, the CPU side only saw a modest increase in preformance. If IB-E was made then Intel could either decrease TDP and keep the clock speeds the same with slightly higher performance or they could keep the TDP they had and increase clockspeed quite a bit.

They could also add more cores in this fictional IB-E. Overall it wouldn't be much of a performance boost unless they really improved something.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top