12/16 Cores.

[Davedree]

AMD have played with LGA for a bit, you'll eventually see them go LGA on desktop.

yeah you're right they did years ago for the fx 70-74 desktop.
Had heard rumours they are looking to go back to it for the future for desktop,

 

[Dead Dog]

http://www.overclockers.co.uk/showproduct.php?prodid=FS-267-OE&groupid=43&catid=2385&subcat=1270

Or this! I bet OcUK are just dying for someone to order one so they can build it!

lol

 

[:(Rhodri:)]

How overclockable are these chips? If you take a 6 core k-series 2011 chip and overclock it to 5GHz it will have exactly the same performance as one of those Xenons even assuming that 3.8GHz of turbo applies to all cores! Not a good investment IMO.

They're not aimed at enthusiasts really, they're for heavily multi-threaded applications like 3D rendering (I use a new i7 at work and run a renderfarm and the 12 cores decimate it in render times), core clock is important, but more cores and threads always help.

 

[joeyjojo]

They're not aimed at enthusiasts really, they're for heavily multi-threaded applications like 3D rendering (I use a new i7 at work and run a renderfarm and the 12 cores decimate it in render times), core clock is important, but more cores and threads always help.

Or as nodes in high performance clusters. We often get time on HECToR for big simulations, which is a cluster of 2816 nodes, each with two 16 core interlagos chips (total of 90,112 cores).

 

[CAT-THE-FIFTH]

They're not aimed at enthusiasts really, they're for heavily multi-threaded applications like 3D rendering (I use a new i7 at work and run a renderfarm and the 12 cores decimate it in render times), core clock is important, but more cores and threads always help.

Can I ask you a question,please if you don't mind??

Now I know modelling for 3D stuff is generally lightly threaded,but does single thread performance affect the actually rendering step after you finished modelling?? An example is lets say you get an Intel CPU with less threads but similar performance to an AMD with more threads,would there be a noticeable difference. Sorry,if it is a daft question!!

 

[JonSmith]

Rendering the rotating 3d viewports is GPU based, I did some tests with my dad where we had an Athlon x2 4850e paired with the onboard 7050 chip, a HD5450 and a HD6450. There was an increase in FPS from the 7050 to the 6450

The actual rendering process that makes the final image is almost never single threaded as it's an embarrassingly parallel problem. You simply split the image into as many regions as you have cores and render them separately. You can also render the final image on a GPU if the software allows it.

 

[CAT-THE-FIFTH]

Rendering the rotating 3d viewports is GPU based, I did some tests with my dad where we had an Athlon x2 4850e paired with the onboard 7050 chip, a HD5450 and a HD6450. There was an increase in FPS from the 7050 to the 6450

The actual rendering process that makes the final image is almost never single threaded as it's an embarrassingly parallel problem. You simply split the image into as many regions as you have cores and render them separately. You can also render the final image on a GPU if the software allows it.

Thanks!! Someone was telling me that single thread performance would limit the GPU rendering step(at least this is what I could gather). Hence something like a Core i5 3570k would be just faster overall for something like Cinema 4D and the like than the AMD 8 thread CPUs,ie,you would avoid them for such things, and even for CPU only rendering the relative slowness of the AMD CPUs during modelling would make them overall slower.

Now,I know that doing modelling would theoretically be faster on an Intel CPU as it is more lightly threaded,but I thought the AMD CPUs were at least OK for stuff like rendering.

 

[:(Rhodri:)]

Thanks!! Someone was telling me that single thread performance would limit the GPU rendering step(at least this is what I could gather). Hence something like a Core i5 3570k would be just faster overall for something like Cinema 4D and the like than the AMD 8 thread CPUs,ie,you would avoid them for such things, and even for CPU only rendering the relative slowness of the AMD CPUs during modelling would make them overall slower.

Now,I know that doing modelling would theoretically be faster on an Intel CPU as it is more lightly threaded,but I thought the AMD CPUs were at least OK for stuff like rendering.

In theory if they were similar performance the extra threads would mean you rendered faster, however Intel are much faster clock for clock and as such tend to dominate render farms in 3D work (in my experience at any rate). The fact they are also faster in single threaded applications means you see very few if any AMD CPUs in studios.

 

[CAT-THE-FIFTH]

In theory if they were similar performance the extra threads would mean you rendered faster, however Intel are much faster clock for clock and as such tend to dominate render farms in 3D work (in my experience at any rate). The fact they are also faster in single threaded applications means you see very few if any AMD CPUs in studios.

Thanks!

Someone told me,if they were to use an FX8350 in a render farm as opposed to a Core i5 3570,the lower single thread performance would bottleneck it in the render farm,as tasks immediately prior to the render would be slowed down massively despite the multi-threaded advantage(it was something to do with also using graphics cards too). Hence Core i5 3570>>>>something with more but weaker threads like an FX8350. Is this true?? Of course,I thought you would not use such CPUs in a farm,with some of the higher core count Xeons you were talking about. Looking at the reviews the FX8350 tends to do be closer to a Core i7 3770 in many 3D rendering benchmarks with multi-threaded performance,but of course with much inferior single threaded performance.

 

[:(Rhodri:)]

I'm not sure on that I'm afraid, as we don't have any AMDs in here, I can imagine that if the AMDs were slower to open the scene etc even by a number of seconds it would add up over a long render cycle, so is certainly possible.

 

[CAT-THE-FIFTH]

I'm not sure on that I'm afraid, as we don't have any AMDs in here, I can imagine that if the AMDs were slower to open the scene etc even by a number of seconds it would add up over a long render cycle, so is certainly possible.

Thanks!! So I suppose,it depends on a balance of these factors,ie,even if took a tad longer to open up,if the rendering performance was high enough it might be able compensate for this,or if the delay was not present it would compensate for low multi-threaded throughput.

Its a shame reviews don't look at these sorts of things,especially the ones who use 3ds MAX and POV Ray instead of a quick CB 11.5 benchmark.

 

[joeyjojo]

It depends entirely on the workload. As posted above, I was surprised to find out that the HPC we use uses dual 16 core 2.3 GHz interlagos chips (opteron 6276) which are essentially the server version of bulldozer. They have a TDP of just 115 W which makes their "threads per watt" extremely competitive. They're £500 each though so not really for consumers/casual work.

There are reviews, but usually for servers. Anandtech give them the full treatment but find for servers they're just a little bit too power hungry for the performance. Still they seem to be a great choice for clusters, and presumably therefore also for render farms.

 

[CAT-THE-FIFTH]

HECToR is at Edinburgh,right?? It was the fastest supercomputer in the UK at the time it was built. Titan which is 20 petaflop monster is probably going to be the fastest supercomputer in the US too,and that uses Interlagos too.

 

[joeyjojo]

Someone told me,if they were to use an FX8350 in a render farm as opposed to a Core i5 3570,the lower single thread performance would bottleneck it in the render farm,as tasks immediately prior to the render would be slowed down massively despite the multi-threaded advantage(it was something to do with also using graphics cards too). Hence Core i5 3570>>>>something with more but weaker threads like an FX8350. Is this true?? Of course,I thought you would not use such CPUs in a farm,with some of the higher core count Xeons you were talking about. Looking at the reviews the FX8350 tends to do be closer to a Core i7 3770 in many 3D rendering benchmarks with multi-threaded performance,but of course with much inferior single threaded performance.

3% faster than a 3770 in well parallelised applications (e.g. 2nd pass x264 http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested/3) for 62% of the price.

HECToR is at Edinburgh,right?? It was the fastest supercomputer in the UK at the time it was built. Titan which is 20 petaflop monster is probably going to be the fastest supercomputer in the US too,and that uses Interlagos too.

No idea still number 1 in the UK I believe, 800 TFLOPS.
edit again: yes you're right Edinburgh Uni.

 

[CAT-THE-FIFTH]

3% faster than a 3770 in well parallelised applications (e.g. 2nd pass x264 http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested/3) for 62% of the price.



No idea still number 1 in the UK I believe, 800 TFLOPS.
edit again: yes you're right Edinburgh Uni.

Under Linux,the FX8350 is not bad at all:

http://www.phoronix.com/scan.php?page=article&item=amd_fx8350_visherabdver2&num=1

Hopefully,AMD will produce a SR based successor and get it out next year.

 

[Zuban]

Thanks!! So I suppose,it depends on a balance of these factors,ie,even if took a tad longer to open up,if the rendering performance was high enough it might be able compensate for this,or if the delay was not present it would compensate for low multi-threaded throughput.

Its a shame reviews don't look at these sorts of things,especially the ones who use 3ds MAX and POV Ray instead of a quick CB 11.5 benchmark.

I've no expeirence of render farms etc as i'm just a hobbyist 3d user, but certainly before a frame is rendered there can be a lot of things that take place depening on the models and scene, subdividing meshes, moving objects etc and it will depend on the software whether all these functions are multithreaded or not, there can be a significant difference between cpu's depending on the complexity of the scene, how much you prebake and so on, its quite complex.

this is one reason why i sometimes crop up when bulldozer is reccomended for 3d rendering because overall the intel chips still offer a much better experience, its not always about the overall time either but the user experience, if your working on a model, and perform a complex boolean operation for example, then you can be sat waiting whilst one core thrashes away for ages, it doesnt help the workflow, and intels cpu's with their significantly better single threaded are still the better choice, i found big increases in speed between my 1090t @3.8ghz and my 3930k at stock for those single threaded moments, obviously with it overclocked its a massive difference, and doing things that used to lock up my amd rig were just not an issue anymore.

Piledriver though offers fantastic value for money when it comes to rendering, but 3d software is so huge it typically is made up of hundreds of parts and plugins all of which could be single/lightly threaded or multithreaded, the 3d rendering engine itself is always multithreaded but theres a lot of other bits before you get there that arent. its not something that will change that quickly either, 3d softwares been multithreaded since i started using it in the mid 90's because its always been used on multi socket machines before multicore so its not a new concept there but its just taking a very long time for things to change, so whilst that happens the intel's are still going to be the more favourable choice for this kind of thing.

incedently, the thread where i was running some tests in premier my 3930k @ 4.6ghz was around 22 minutes to render the sequence whilst a dual xenon with 16 cores was around 20 minutes, obviously there is a massive difference in price but once overclocked the single processor isnt far behind.

 

[CAT-THE-FIFTH]

I've no expeirence of render farms etc as i'm just a hobbyist 3d user, but certainly before a frame is rendered there can be a lot of things that take place depening on the models and scene, subdividing meshes, moving objects etc and it will depend on the software whether all these functions are multithreaded or not, there can be a significant difference between cpu's depending on the complexity of the scene, how much you prebake and so on, its quite complex.

this is one reason why i sometimes crop up when bulldozer is reccomended for 3d rendering because overall the intel chips still offer a much better experience, its not always about the overall time either but the user experience, if your working on a model, and perform a complex boolean operation for example, then you can be sat waiting whilst one core thrashes away for ages, it doesnt help the workflow, and intels cpu's with their significantly better single threaded are still the better choice, i found big increases in speed between my 1090t @3.8ghz and my 3930k at stock for those single threaded moments, obviously with it overclocked its a massive difference, and doing things that used to lock up my amd rig were just not an issue anymore.

Piledriver though offers fantastic value for money when it comes to rendering, but 3d software is so huge it typically is made up of hundreds of parts and plugins all of which could be single/lightly threaded or multithreaded, the 3d rendering engine itself is always multithreaded but theres a lot of other bits before you get there that arent. its not something that will change that quickly either, 3d softwares been multithreaded since i started using it in the mid 90's because its always been used on multi socket machines before multicore so its not a new concept there but its just taking a very long time for things to change, so whilst that happens the intel's are still going to be the more favourable choice for this kind of thing.

incedently, the thread where i was running some tests in premier my 3930k @ 4.6ghz was around 22 minutes to render the sequence whilst a dual xenon with 16 cores was around 20 minutes, obviously there is a massive difference in price but once overclocked the single processor isnt far behind.

Thanks again!! This thread has been very useful indeed. I just wished review methodologies,made things clearer - if they explained things like some of the people have said in this thread,for the average person looking at doing some hobbyist 3D stuff,it would inform them better.

Having said that most reviews post useless benchmarks on something simple like audio transcoding too - as most people rip from a SATA or USB DVD drive,you are most likely to hit a read speed or transfer interface bottleneck. Many of the reviews rip CDs copied onto an SSD or onto a RAMdisk which is not what 99% of people do I suspect.

 

[Muzza1875]

I wonder how these would be for folding at home.