Sunday, June 2nd 2013

FinalWire AIDA64 v3.00 Released

FinalWire Ltd. today announced the immediate availability of AIDA64 Extreme Edition 3.00 software, a streamlined diagnostic and benchmarking tool for home users; and the immediate availability of AIDA64 Business Edition 3.00 software, an essential network management solution for small and medium scale enterprises.

The new AIDA64 release introduces completely rewritten memory bandwidth and memory latency benchmarks, optimized for AVX2, AVX and SSE capable AMD, Intel and VIA processors. The latest AIDA64 update also implements optimized 64-bit benchmarks for AMD "Kabini" and Intel "Haswell" processors, and supports the latest AMD Radeon and nVIDIA GeForce graphics accelerators.
DOWNLOAD: AIDA64 v3.00 Installer, AIDA64 v3.00 ZIP

New features & improvements
  • Multi-threaded cache and memory bandwidth benchmarks with AVX2, AVX and SSE optimizations
  • Block-random cache and memory latency benchmark
  • Optimized 64-bit benchmarks for AMD "Kabini" and "Temash" APUs
  • AVX2 and FMA optimized 64-bit benchmarks for Intel "Haswell" processors
  • Preliminary support for L4 cache of Intel "Crystal Well" processors
  • Crucial M500, OCZ Vertex 450 SSD support
  • GPU details for AMD Radeon HD 7990 "Malta" and nVIDIA GeForce 700 Series
New cache and memory benchmarks
The completely rewritten cache and memory bandwidth benchmarks now utilize AVX2, AVX and SSE instructions of the latest AMD, Intel and VIA processors; and use up to 32 threads for read, write and copy bandwidth measurement. The brand new cache and memory latency benchmark now uses a block-random method to eliminate the effect of hardware prefetchers, and to minimize TLB misses. For legacy processors all benchmarks and the System Stability Test are available in 32-bit version as well.

Optimized 64-bit benchmarks for AMD "Kabini" and "Temash" APUs
AIDA64 CPUID Panel, Cache & Memory Benchmark panel, System Stability Test, and all memory and processor benchmarks are fully optimized for AMD "Kabini" and "Temash"desktop, mobile and tablet APUs, utilizing AVX, SSE4.1 and AES-NI instructions. Detailed chipset information for AMD "Yangtze" integrated Fusion Controller Hub. Improved support for the upcoming AMD "Kaveri" and "Beema" APUs.

Next-generation Intel processors and chipsets support
AVX2 and FMA optimized 64-bit benchmarks for the upcoming Intel "Haswell" desktop and mobile processors. Extended chipset information for Intel "Lynx Point" and "Lynx Point-LP" PCHs, and for the integrated memory controller of Intel "Haswell" processors. AIDA64 benchmarks and System Stability Test utilize Advanced Vector Extensions 2 (AVX2), Fused Multiply-Add (FMA) instructions, and AES-NI hardware acceleration of the new Intel processors. Improved support for Intel "Avoton" and "Valleyview" SoC, "Haswell-E" and "Ivy Bridge-E" CPU. Preliminary L4 cache bandwidth and latency benchmarks for the upcoming Intel "Crystal Well" processor. Intel Centrino Mobile Technology 9th Generation (Shark Bay-MB) compliancy test.

Support for the latest hardware technologies
Graphics processor, OpenGL and GPGPU details, temperature and cooling fan monitoring for the latest GPUs: AMD Radeon HD 7730, Radeon HD 7990 "Malta"; nVIDIA GeForce GT 705M, GeForce GT 735M, GeForce GT 740M, GeForce GT 745M, GeForce GTX 760M, GeForce GTX 765M, GeForce GTX 770, GeForce GTX 770M, GeForce GTX 780, GeForce GTX 780M, GeForce GTX Titan LE, Tesla K20, Tesla K20X. Improved support for Crucial M500 and OCZ Vertex 450 SSDs.
Add your own comment

13 Comments on FinalWire AIDA64 v3.00 Released

#1
buggalugs
wow, my read/write/copy memory benchmarks went up 4000-5000 MB/s on my 3770K ......but my latency is 20% worse. Weird
Posted on Reply
#2
Fiery
FinalWire / AIDA64 Developer
by: buggalugs
wow, my read/write/copy memory benchmarks went up 4000-5000 MB/s on my 3770K ......but my latency is 20% worse. Weird
I'm quoting from the AIDA64 forum, since I'm not sure if I'm allowed to post a link to there.
We're rolling out a new major update to AIDA64 on June 03, 2013. It will feature the usual improvements to support the latest and greatest hardware technologies, such as GPU details for AMD Radeon HD 7990 "Malta" and nVIDIA GeForce 700 Series, and optimized benchmarks for AMD Kabini/Temash and Intel Haswell. On Haswell AIDA64 v3.00 will utilize all new instruction set extensions, so the benchmarks and the System Stability Test will also use the appropriate AVX2, FMA and BMI2 instructions.

But more importantly, we've replaced the outdated set of memory benchmarks with brand new ones. The new bandwidth (read, write copy) benchmarks now use multiple threads to squeeze out every last bit of performance from the caches and the memory modules. On modern multi-core processors, using the old single-threaded benchmarks you couldn't see the actual memory bandwidth, but only the memory bandwidth available for single-threaded applications. With the new benchmarks you will however get considerably higher scores, much closer to the theoretical memory bandwidth available. It is especially true for 3-channel and 4-channel memory configurations, such as Intel X58 and X79 based high-end desktop systems; and also for NUMA-enabled multi-socket systems, such as 2- and 4-way AMD Opteron and Intel Xeon based servers and workstations. For example:

Core i7-3960X with X79 chipset and 4-channel DDR3-1600:
- AIDA64 v2.85 Memory Read: 16825 MB/s [ old ]
- AIDA64 v3.00 Memory Read: 45640 MB/s [ new ]

We've also implemented multi-threaded cache benchmarks, that now include support for the latest AVX and AVX2 instruction set extensions. Thanks to using all CPU cores for the cache benchmarks, you will get dramatically different cache bandwidth scores than with the old benchmarks. For example:

Core i7-3960X with X79 chipset and 4-channel DDR3-1600:
- AIDA64 v2.85 L1 Cache Read: 121.8 GB/s [ old ]
- AIDA64 v3.00 L1 Cache Read: 674.7 GB/s [ new ]

And thanks to the 2x widened L1 cache bandwidth of Intel Haswell, on these new processors using the new cache benchmarks of AIDA64 v3.00 you will get unusually high scores. Using Haswell, with a hint of overclock it's quite easy to cross the 1 TB/s mark for the L1 cache ;)

We've also replaced the old cache and memory latency benchmark with a brand new one that uses a different approach, recommended by processor architecture engineers. The old memory latency benchmark used the classic forward-linear solution, so it "walked" the memory continuously, in forward direction. Unfortunately that classic approach was sometimes over-optimized by "too smart" memory controllers, that led to unrealistically low latency scores. It was a constant fight for us to get around those over-optimizations, to make sure AIDA64 provides stable and reliable latency results. With the new latency benchmark we've switched to a block-random solution, that keeps "jumping" to random addresses inside a memory block for a period of time, and then skips to a new block and continues "jumping" to random places inside there as well. With this new solution memory controllers cannot find a pattern anymore in the latency measurement, and so they cannot over-optimize the benchmark. The block-random approach however means that latency results will be higher, and since the scores are in nanosec, it means the results will be worse than what you got used to. For example:

Core i7-3960X with X79 chipset and 4-channel DDR3-1600:
- AIDA64 v2.85 Memory Latency: 55.9 ns [ old ]
- AIDA64 v3.00 Memory Latency: 67.5 ns [ new ]

AIDA64 v3.00 also supports benchmarking the eDRAM L4 cache of the upcoming Intel Crystal Well processors.

Less radical change, but still deserves a note: we've also revamped the framework around the CPU ZLib benchmark, so you can see a few percent gain in performance using AIDA64 v3.00. For example:

Core i7-3960X with X79 chipset and 4-channel DDR3-1600:
- AIDA64 v2.85 CPU ZLib: 418.3 MB/s [ old ]
- AIDA64 v3.00 CPU ZLib: 444.5 MB/s [ new ]

We hope you'll find the new AIDA64 release useful.
Regards,
Fiery
Posted on Reply
#3
buggalugs
by: Fiery
I'm quoting from the AIDA64 forum, since I'm not sure if I'm allowed to post a link to there.



Regards,
Fiery
Ok thanks for the info. This is a huge change in performance benchmarks especially for triple/quad channel memory setups.

All this time reviewers have been telling us memory performance of triple/quad setups isn't much better than dual channel when in reality it is. Most review sites use AIDA for memory benchmarks, I don't know why you wouldn't update the memory benchmarks earlier if there is such a huge difference like this.

It could change people's purchasing decisions.
Posted on Reply
#4
Fiery
FinalWire / AIDA64 Developer
by: buggalugs
Ok thanks for the info. This is a huge change in performance benchmarks especially for triple/quad channel memory setups.

All this time reviewers have been telling us memory performance of triple/quad setups isn't much better than dual channel when in reality it is. Most review sites use AIDA for memory benchmarks, I don't know why you wouldn't update the memory benchmarks earlier if there is such a huge difference like this.

It could change people's purchasing decisions.
Well... First of all, memory performance isn't that much of a factor than what synthetic numbers would lead us to believe. Going from dual-channel to quad-channel is by no means 100% performance increase in the "real world", even when you add a few more CPU cores. Second, it took us over 18 months to develop and validate the multi-threaded cache and memory benchmarks. There's a reason why there's not much tools you can use for complete multi-threaded cache and memory benchmarking purposes: it's a tough job to properly measure cache and memory bandwidth when you need to utilize more than 1 processor thread :(
Posted on Reply
#5
Ed_1
by: Fiery
Well... First of all, memory performance isn't that much of a factor than what synthetic numbers would lead us to believe. Going from dual-channel to quad-channel is by no means 100% performance increase in the "real world", even when you add a few more CPU cores. Second, it took us over 18 months to develop and validate the multi-threaded cache and memory benchmarks. There's a reason why there's not much tools you can use for complete multi-threaded cache and memory benchmarking purposes: it's a tough job to properly measure cache and memory bandwidth when you need to utilize more than 1 processor thread :(
IS there any chance of adding more load to your stability tests . I see many times MB makers like Asus recommend not using prime95 or Linx (Intel burn test) etc .

But I see many reports that just using AIDA64 (I assume older version than 3.0 ) it can be total stable in AIDA64 but not elsewhere when OC .
You can see HardOCP review of 4770k for example

Maybe you could add a extreme level ,which would keep what you have but add a bit more load .
Just a thought .
Posted on Reply
#6
buggalugs
by: Fiery
Well... First of all, memory performance isn't that much of a factor than what synthetic numbers would lead us to believe. Going from dual-channel to quad-channel is by no means 100% performance increase in the "real world", even when you add a few more CPU cores. Second, it took us over 18 months to develop and validate the multi-threaded cache and memory benchmarks. There's a reason why there's not much tools you can use for complete multi-threaded cache and memory benchmarking purposes: it's a tough job to properly measure cache and memory bandwidth when you need to utilize more than 1 processor thread :(
Ok That makes sense. Sorry if I came across a little ungrateful. I love AIDA, been using it for years....back when it was called Everest.

It touched a nerve with me because memory performance was one of the main reasons I was unhappy with my X79 quad channel system that I spent $1000 on. My Z77 dual channel system performed 25% better on memory benchmarks at the same clocks.....now it turns out the benchmarks weren't really accurate.

Not blaming AIDA, this should have been made clear by reviewers or Intel should have provided a tool or explanation or something.

Basically we're saying single threaded memory benchmarks that are all over the internet in many reviews going back for the last few years, are pretty much useless and inaccurate.

I do appreciate it is hard to do, and it took a long time, its better late than never I guess.....
Posted on Reply
#7
Fiery
FinalWire / AIDA64 Developer
by: Ed_1
IS there any chance of adding more load to your stability tests . I see many times MB makers like Asus recommend not using prime95 or Linx (Intel burn test) etc .

But I see many reports that just using AIDA64 (I assume older version than 3.0 ) it can be total stable in AIDA64 but not elsewhere when OC .
You can see HardOCP review of 4770k for example

Maybe you could add a extreme level ,which would keep what you have but add a bit more load .
Just a thought .
If you enable only the FPU and GPU subtests, the stress on the system is very high. Using a power draw meter device, you can see that it draws more power from the system than using other stress testing solutions. But of course, stability is not like math, so you cannot make a single test that can be used to reveal any sorts of instability. Sometimes Prime95 works better than AIDA64 to get to a BSoD or lockup, but on other systems AIDA64 finds the issue first.

BTW, on the HardOCP review... I don't mean to knock an otherwise great site, a great author and a great review, but what the author there explained about AIDA64 is just silly. Why would anyone consider an overheating (throttling) system stable? Why would you even want to get to a BSoD or lockup when your system is overheating? First, you should take care of properly cooling the overclocked system, and only after then you get to finding stability issues. The AIDA64 System Stability Test clearly shows the overheating state. Putting that to a review with huge screen shots, and then knocking AIDA64 System Stability Test is just not something I'd take too seriously :ohwell:
Posted on Reply
#8
Fiery
FinalWire / AIDA64 Developer
by: buggalugs
It touched a nerve with me because memory performance was one of the main reasons I was unhappy with my X79 quad channel system that I spent $1000 on. My Z77 dual channel system performed 25% better on memory benchmarks at the same clocks.....now it turns out the benchmarks weren't really accurate.
It was quite accurate. It accurately reflected the memory performance you could get when using only 1 or 2 processor threads. Typically you get that by browsing, or playing classic 3D games, or running any applications that isn't multi-threaded.

And BTW, on the X79 system, try to fall back to a 2-channel configuration by removing half of the memory modules. Then use a few benchmarks (not just AIDA64, but PCMark, some game timedemo, etc), to see how much performance would you lose. You won't lose that many percentage overall -- I'd say between 5 and 10%, but in most cases even less than that. And that's where relying on pure memory bandwidth scores turns out to be the wrong approach ;)
Not blaming AIDA, this should have been made clear by reviewers or Intel should have provided a tool or explanation or something.
Please don't blame Intel :) They reached to us before they introduced X79 and Sandy Bridge-E, but back then we were simply not ready to roll out the multi-threaded memory bandwidth benchmarks. We here at FinalWire are a sort of guys who still believe you need to release something only when it is ready. We don't publish something that we're not 100% certain about. We had to keep the old, single-threaded memory bandwidth benchmarks while we were developing the new ones. By releasing unfinished benchmarks that provide unreliable (albeit considerably higher) scores would just upset many of you guys ;)
Basically we're saying single threaded memory benchmarks that are all over the internet in many reviews going back for the last few years, are pretty much useless and inaccurate.
No, not at all. They are still accurate, although not quite relevant on 3 or 4-channel memory configurations. Would you consider any benchmarks running on Haswell and not having AVX2+FMA optimizations useless and inaccurate? (that would cover 99% of the benchmarks, except for AIDA64 v3.00 of course :cool: ) Would you consider SuperPI an inaccurate benchmark just because it's single-threaded? No and no, if you ask me.
Posted on Reply
#9
Ed_1
by: Fiery
If you enable only the FPU and GPU subtests, the stress on the system is very high. Using a power draw meter device, you can see that it draws more power from the system than using other stress testing solutions. But of course, stability is not like math, so you cannot make a single test that can be used to reveal any sorts of instability. Sometimes Prime95 works better than AIDA64 to get to a BSoD or lockup, but on other systems AIDA64 finds the issue first.

BTW, on the HardOCP review... I don't mean to knock an otherwise great site, a great author and a great review, but what the author there explained about AIDA64 is just silly. Why would anyone consider an overheating (throttling) system stable? Why would you even want to get to a BSoD or lockup when your system is overheating? First, you should take care of properly cooling the overclocked system, and only after then you get to finding stability issues. The AIDA64 System Stability Test clearly shows the overheating state. Putting that to a review with huge screen shots, and then knocking AIDA64 System Stability Test is just not something I'd take too seriously :ohwell:
Ok, that fair enough . I do realize that no one test whether it be prime95, LinX, 3dmark etc
is going to be totally stability if passed . you do need to run whatever apps/games to see if they are stable .

I guess what rubs me a little is AIDA64 is payware and if i understand it it not cut down after trail time period is up (like 3dmark ) . So I haven't tried it as I don't install trail stuff if its useless after that point . Maybe only exception would be photoshop or something .

Anyway, maybe you keep adding tests over time that would broaden the test base and maybe catch more things .
thanks for reply ,feedback .
Posted on Reply
#10
Fiery
FinalWire / AIDA64 Developer
by: Ed_1
Anyway, maybe you keep adding tests over time that would broaden the test base and maybe catch more things.
We definitely want to do that. We'll also come up with a redesigned System Stability Test layout that helps to better understand how stress testing works. We'll soon publish video tutorials as well, to guide users on verifying system stability with AIDA64.
Posted on Reply
#11
Fiery
FinalWire / AIDA64 Developer
by: Fiery
BTW, on the HardOCP review... I don't mean to knock an otherwise great site, a great author and a great review, but what the author there explained about AIDA64 is just silly. Why would anyone consider an overheating (throttling) system stable? Why would you even want to get to a BSoD or lockup when your system is overheating? First, you should take care of properly cooling the overclocked system, and only after then you get to finding stability issues. The AIDA64 System Stability Test clearly shows the overheating state. Putting that to a review with huge screen shots, and then knocking AIDA64 System Stability Test is just not something I'd take too seriously :ohwell:
One more issue about the HardOCP case, them using AIDA64 System Stability Test: I'm afraid they must have used an outdated AIDA64 version that didn't support Haswell, and so it didn't use FMA code in the System Stability Test. Without using that new instruction set extension, it's not possible to fully load Haswell's execution units. Truth be told, you cannot blame HardOCP on that particular issue, since AIDA64 v3.00 was released after they've written the review on Haswell.
Posted on Reply
#12
Ed_1
by: Fiery
We definitely want to do that. We'll also come up with a redesigned System Stability Test layout that helps to better understand how stress testing works. We'll soon publish video tutorials as well, to guide users on verifying system stability with AIDA64.
that would be very good on both cases , many app don't have good documentation and video is best to show how to use it as you get visual tutorial .

Yes, to bad you released version 3.0 few days after reviews came out .

I am old school I basically use prime95 (newest version) and then test apps . I do agree LinX is not very realistic load , only thing it good for mainly is to see how thermal load goes (how good HS etc )
Posted on Reply
#13
TheHunter
woot my Q9450 finally broke 10Gb/s mark ^^

Posted on Reply
Add your own comment