Quartus Performance - What Processor, Memory, Disk?

--- Quote Start ---

I'm about to get a new PC, and want to make sure that Quartus will run smoothly and fast. I'm managing increasingly large designs and I hate to wait, especially when working late :-)

I've searched on the web and on the Altera Website and, apart from recommending against using Windows 7, I haven't found much out there. If anyone has advice, or knows where there is some, I would greatly appreciate being pointed to it.

Altera -- it would really be great if someone could create a few reference designs of different sizes and post the compile performance on various processor and o/s configurations. I have to think that someone at Altera is doing this work and looking at the results. If both the results AND the designs are posted, it would allow customers to be able to run their own tests so that they can judge how their machine is performing.

(at the same time, i understand that the x-company might take these same designs and use them for comparative purposes, but i still think this would have a lot of value for customers. and, it's not like a little competition is a bad thing, is it? :-) )

I guess the obvious advice here is to get the fastest processor with the most memory, but it's kind of hard to figure out what that means today. There are a variety of multi-core solutions, and there are huge differences in floating point performance between processors, especially now that the new i-series Intel processors are out. Still, if those parts don't get me anything, I'd prefer to spend my money somewhere else. There's nothing else I do that eats processing cycles like Quartus, so I'd like to tailor the machine to Quartus

I'm curious whether anyone has any advice on the relative performance of Quartus running on:

• different processors (single, multiple cores, multi-threaded cores, esp AMD vs Intel)

• memory configurations

• operating systems

• graphics cards (doubt this makes a difference but, hey, wouldn't it be great if a few hundred GPUs were working to speed up the processing?)

Thanks in advance!

--- Quote End ---

Hi,

I'm not a "real" PC expert and I don't have run a benchmark with different PC or PC configurations.

My larges Design is a StratixIV (EP4SE530) with 80% Utitlization and moderate Timing requirements. With my PC a full compile takes ~ 3h. The Peak Memory Usage is around 7GByte. Especially the RTL view or technology view needs a lot of Memory. I would expect that the largest StratixIV or the announced StratixV requires 16Gbyte Memory.

For the Analysis,Synthesis, Fitter and Timing Analysis Quartus offers a ( in my point of view) very good Multi-Processor Support ( in average up to 3 - 4) .

Maybe this infos helps a little bit.

Kind regards

GPK

There are differences between Quartus Web Edition and subscription Edition :

multiple cores cpus are useless for quartus web edition.

I think high level cache memory (l1, l2...) on cpus can help very much.

For RAM : I suggest to use only qdr-dram. The more, the better. But you need a good motherboard to employ those memory.

For OS : no idea between 64bit and 32bit edition of windows.

Vista was just a transition between XP and Seven.

My new PC , CPU is I7 ,when compilation, can only utilize one core~~:rolleyes:

Who can tell me how to fix it?

Quartus 9.1 SP2

Have a look in quartus > assignment settings > ...

In my previous post, I meant that with quartus web edition, you can only use one core.

--- Quote Start ---

For the Analysis,Synthesis, Fitter and Timing Analysis Quartus offers a ( in my point of view) very good Multi-Processor Support ( in average up to 3 - 4) .

--- Quote End ---

My experience has been that the multi-processor support of Quartus is useless. Whenever I have enabled it, it does not seem to use my extra cores any substantial amount. And then to top it off, when I timed a complete compile, it ended up taking the same amount of time even though it used more processor resources.

I keep thinking that the effectiveness of the multi-processor support must depend heavily on the design being compiled, otherwise Altera should be embarrassed to advertise it as a feature. I have only ever tried it with relatively small designs that only take 5 - 10 minutes to compile.

I agree with mmTsuchi, I think having the fastest memory and cache possible is a big benefit. From my experience with the designs I have worked with, this is more important than extra processor cores.

--- Quote Start ---

My experience has been that the multi-processor support of Quartus is useless. Whenever I have enabled it, it does not seem to use my extra cores any substantial amount. And then to top it off, when I timed a complete compile, it ended up taking the same amount of time even though it used more processor resources.

I keep thinking that the effectiveness of the multi-processor support must depend heavily on the design being compiled, otherwise Altera should be embarrassed to advertise it as a feature. I have only ever tried it with relatively small designs that only take 5 - 10 minutes to compile.

I agree with mmTsuchi, I think having the fastest memory and cache possible is a big benefit. From my experience with the designs I have worked with, this is more important than extra processor cores.

--- Quote End ---

Hi,

in order to show my experience with multi-processor support I ran a small benchmark with

one large StratixIV design:

All available CPU's allowed:

Analysis & Synthesis 00:31:15 3.3 2413 MB

Partition Merge 00:02:03 1.0 1356 MB

Fitter 02:04:15 3.1 3746 MB

TimeQuest 00:14:28 1.5 2678 MB

Assembler 00:06:40 1.0 1933 MB

02:58:41

Only 1 CPU allowed

Analysis & Synthesis 01:03:35 1.0 2401 MB

Partition Merge 00:02:12 1.0 1053 MB

Fitter 02:40:47 1.0 3618 MB

TimeQuest 00:15:34 1.0 1934 MB

Assembler 00:05:58 1.0 1934 MB

04:08:06

Kind regards

GPK

In the bench mark when 4 cpu's were used the time decreased about 25%, so the number of cores is not the total answer. The experience that shows more and faster memory follows logically, because a core cannot process if it is waiting for memory. Many years ago Gene Amdahl said that a multi-processor would increase performance about 40%. That was for a 2-way and it looks like it is true for a 4-way. The next thing to consider is whether the data is in memory or not. The cpu has to wait to get data from memory, and it has to be brought in from disk first. For large files the block size on disk is 4KB or larger so the average data rate is high. But look at the file sizes in Quartus .. probably 1KB average. Each file hast to be read individually so a large disk cache is very important.

Beware all the hype about multi-core.

Kind regards, Karl

I think what works best is quite dependent on your design.

Personally, I have never seen the multi processor option in Quartus help. However, some of the results posted in this thread indicate it can be a big help.

Has anyone here ever done testing with 32 bit vs 64 bit?

I also tested this with the same design I mentioned earlier. I found that using 64 bit took slightly longer to compile. In this case, I would guess if your design can compile with less than about 2 gigs of RAM usage, 32 bit is probably a little faster. For larger designs, 64 bit may be a big advantage.

One thing I have not tried is compiling under Windows vs Linux. Does anyone have any experience with this?

For operating system, it doesn't matter much for Quartus - Windows and Linux have similar compile times.

Bigger caches definitely help a lot.

One thing to keep in mind when you are benchmarking multi-core compilation: Quartus is parallelizing the time-intensive algorithms, but not as much for the constant overhead of a compile. If you benchmark a short compile (eg. 5 or 10 minutes), you won't see much parallel benefit since those 5 or 10 minutes are dominated by constant overhead.

On average, we measure ~20% compile time reduction when using a multi-core machine - your exact performance will depend on your design. For example, if your design is hard to route, it will spend more time in the algorithms and so parallel will be more important.