Opteron 170 vs 175... price / OC performance

[Tomcat76]

I'm interested in upgrading from an X2 4200+ to an Opteron 170 or 175.

Editing videos is by far the heaviest stuff I'm doing on my computer so any power boost is welcome. Can the 175 "go higher" compared to the 170?

I can have the 170 without paying anything if I sell the X2 4200 (about the same price), but I'd have to pay the difference when getting the 175. Is the 175 worth it compared to the 170?

[ripken204]

well it all depends on how much u want to overclock.

with the 170 you are guaranteed at least 2.6

with the 175 you should be hitting at least 2.7-not too mny ppl have these so cant rly tell

if you get a good stepping then your 170 could his as high as 2.8, if you want any higher then you need water cooling

the 175 is capaple or 2.9-3.0 on water cooling

[Tomcat76]

Alrighty...

Will go with the 170 then...

[maximusement]

Well......... A little late but got a dual 170 myself on a DFI Lanparty SLI-DR (2X 1GB Mushkin Blackline: 530) and I clocked at 2.8 at about 34'c on STOCK. Comes with a really good heatsink compaired to a lot of the aftermarket ones I sell.

Save your money, get the opteron.

[teqguy]

Let me get this straight... you want to upgrade from a 2.2Ghz X2 to a 2Ghz or 2.2Ghz Opteron?

I don't know if you realize this, but video editing and encoding benefit more from thread parallelization and load balancing than just raw clock speed. You can overclock an Opteron to your heart's content and it won't be able to equate to the performance you'd get if you had an encoder that supported SMP, SS3, and all of the other nifty features you're probably not making use of.

Before even considering spending any money... ask yourself if you have everything properly configured and tweaked for maximum performance.

For one, Windows does not automatically detect that you have a multicore/multiprocessor/hyperthreaded system... you have to first uninstall the ACPI Uniprocessor manager in Device Manager and replace it with the Multiprocessor manager.

Second, even after Windows is informed that you have such a system, it does not automatically make use of the additional processors/cores. This is because Windows does not feature thread affinity, as demonstrated by Tom's Hardware.

If you really want to invest in something with an inherent benefit, watercool your X2 or fill up your DIMMs with 4GB of ram, then use half of that for a ramdrive. Then, store the file you're editing in the ramdrive and encode it to and from the ramdrive. Hard drives pale in comparison to a ramdrive, when you figure the average hard drive bandwidth is limited to 60MB/s, while a ramdrive can easily hit over a Gbit/s.

[ripken204]

windows does automatically recognize a dual core cpu. anf for affinity, you can set that yourself if u want to. ram drives are very nice but cost a whole lot and take alot of space.

Well......... A little late but got a dual 170 myself on a DFI Lanparty SLI-DR (2X 1GB Mushkin Blackline: 530) and I clocked at 2.8 at about 34'c on STOCK. Comes with a really good heatsink compaired to a lot of the aftermarket ones I sell.

Save your money, get the opteron.

are you prime stable?

[teqguy]

windows does automatically recognize a dual core cpu.

Actually, unless you specifically select that you want to use a multiprocessor hardware abstraction layer, the default HAL(ACPI Uniprocessor) will be installed and remain installed until you manually change it.

Simply displaying multiple processors in Device Manager is not enough to "acknowledge" that the system has multiple processors AND is actually managing them properly.

anf for affinity, you can set that yourself if u want to.

Windows does not provide an option of doing this, nor do many people even understand what thread affinity is, let alone how to set it.

ram drives are very nice but cost a whole lot and take alot of space.

You don't need additional hardware to make a ramdrive, so I don't see where you're factoring in the added expense. Microsoft provides a tool to make a ramdrive, or you can use a graphical tool such as this: http://superspeed.com/desktop/ramdisk.php if you want to manage it easily.

To reaffirm my point, the benefit of a ramdrive in tasks such as encoding is to eliminate hard drive I/O entirely, which slows down the process tremendously.

In any case, switching from an X2 to an Opteron will end up being a downgrade.

[Marsden]

So the extra 1MB of L2 cache in the Opteron 170 makes no difference?

[ripken204]

for the affinity you just use the task manager and right click on the process

for ram drives, it is much better to get an actual pci card that hold the ram

[teqguy]

So the extra 1MB of L2 cache in the Opteron 170 makes no difference?

It's an additional 512KB of L2 cache... and when you consider you're working with files that are hundreds of megabytes in size, that additional 512KB won't make as much difference as going to a faster storage medium to read and write to.

for the affinity you just use the task manager and right click on the process

I've yet to see that option available in Windows without the aid of a third party tool, such as the one offered by Tom's Hardware.

for ram drives, it is much better to get an actual pci card that hold the ram

Actually, with a PCI card, you're limited to the same 60MB/s average a hard drive has. SATA or PCIe would be better solutions, but even then you're stuck at 150MB/s or 500MB/s, respectively.

In comparison, the slowest DDR memory available running in dual channel mode will get you about 3.4GB/s of theoretical bandwidth, while the fastest memory currently tops off at nearly 12GB/s.

Not to mention, the latency is a lot less going from ram to the processor, instead of from ram to an onboard controller(to translate the ram modules to 'x' interface), to another onboard controller, to the southbridge, northbridge, and then finally the processor.

[Porn Loader]

So the extra 1MB of L2 cache in the Opteron 170 makes no difference?

It's an additional 512KB of L2 cache... and when you consider you're working with files that are hundreds of megabytes in size, that additional 512KB won't make as much difference as going to a faster storage medium to read and write to.

for the affinity you just use the task manager and right click on the process

I've yet to see that option available in Windows without the aid of a third party tool, such as the one offered by Tom's Hardware.

for the 1 meg l2, depends on what you are doing. for gamers, it will increase your minimum fps by a bit aka a couple frames max.

as for your second post teq,

granted its not setting it thru taskmanger, but prime remembers which core has affinity. multi core is still relatively new tech. give it time it'll come around. and plz don' give me the "smp has been around since the mid90's" crap. yes it has, but it has actully become affordable to the cheap geek/home consumer now

[teqguy]

for the 1 meg l2, depends on what you are doing. for gamers, it will increase your minimum fps by a bit aka a couple frames max.

To quote the thread starter:

"Editing videos is by far the heaviest stuff I'm doing on my computer so any power boost is welcome."

granted its not setting it thru taskmanger, but prime remembers which core has affinity.

First of all, Prime is one application... and it's not even in the same category of interest being discussed.

Second, core affinity is an entirely different animal than thread affinity. Core affinity applies to load balancing, while thread affinity applies to multithreading and SMP.

Third, while it seems adequate that the software is actually aware of a second processor/core, performance would be better if the operating system was aware of it and could actually put it to good use.

multi core is still relatively new tech. give it time it'll come around. and plz don' give me the "smp has been around since the mid90's" crap. yes it has, but it has actully become affordable to the cheap geek/home consumer now

Actually, SMP has been around long before that, but that's beside the issue.

The fact of the matter is, software that can take advantage of multithreading and SMP has been available on both an open source and consumer level for a long time... a lot of which deals with multimedia tasks, such as video editing and encoding.

So, to reaffirm my original point, "upgrading" from a dual core Athlon to a single core Opteron, despite cache size and clock frequency, will end up being a downgrade in tasks like these.

[Albuquerque]

Quite a bit of misinformation in here...

Multiprocessor kernel detection

Contrary to what seems to be popular belief in this thread, Windows XP can and will detect the presense of and install the proper ACPI Multiprocessor HAL for SMT and SMP capable processors during the install process. This is painfully simple to test: perform an XP install on any AMD or Intel dual-core processor (or an Intel hyperthreaded processor) and simply watch what happens. However, it is worth noting that if you already have built the operating system and upgraded to an SMT/SMP processor package without a reinstall, XP will not change the HAL by itself.

Multiprocessor Affinity

It's odd that someone would argue that affinity doesn't exist without a 3rd party tool. Maybe you should start task manager, click on Help, click on Task Manager Help Topics, and look at the index. The second entry is, oddly enough, affinity. And it says this:

To assign a process to a processor

• On the Processes tab, right-click the process you want to assign, click Set Affinity, and then click one or more processors.
  

Notes

• The Set Affinity command is available only on multiprocessor computers.
  
• Using the Set Affinity command limits the execution of the program or process to the selected processors and might decrease overall performance.
  

As for your comments on "core affinity", there isn't such a thing. Affinity is defined in two ways: process affinity and thread affinity. A "process" would be the entire application, say Prime. You can control the process affinity by way of Task Manager and in some cases by way of the application. Thread affinity would be in the indivudal worker threads that are spawned by a process to make it all work. Applications that are not SMP/SMT aware may not use all the available processor resources; this is a limitation of the application and not the operating system.

Windows NT5 and higher are all quite aware of the processors given to them and they are able to manage the threads and processes across all logical and physical processors in the system. Your contention that the OS doesn't spread it's own threads across all processors meaning it's less-performance is questionable at best. In reality, how much processing power does your OS consume at any given time? Hopefully very little; my CPU useage stays somewhere between 0 and 1% -- even when looking at kernel load. There's not much need for the OS to start throwing its threads all over the available processors; leave those resources for other applications that can use them.

Hard drives are limited to 60mb/sec

Sure they are, if you're talking about 7200RPM IDE drives. Might want to reconsider for 15,000 RPM SCSI drives, or even 10,000RPM IDE drives.

SATA can go faster than IDE drives, to the tune of 150mb/sec versus 60mb/sec

Not. The SATA interface has a higher maximum burst speed than IDE by 17mb/sec (150 vs 133) but the actual data transfer speed from the drive is dictated by the physical construction of the drive itself, not the interface. You cannot lay down any generalization that says a SATA harddrive is "faster" than an IDE disk.

PCI is limited to ~60mb/sec

Not. PCI is a 32-bit 33mhz bus, which gives you 133mb/sec of bandwidth. There are also PCI-X components, as well as PCI-E options too.

RAMDRIVE will make you go so much faster when compositing video

Doubtful. When transferring RAW video from a digital device to your computer, it is quite possible to get bottlenecked by the data streaming capacity of your hard drive. But when doing video editing, you're using a TON of CPU cycles. This isn't a drive bottleneck, this is a CPU bottleneck. It's better to let your video application use that ram for buffering versus trying to overcome a non-existant disk bottleneck. It's also much better for that CPU to have all the bandwidth it needs to ram for doing that buffering, and not having to contend with bandwidth being soaked up by an unneeded ramdisk storing all that information. Talk about a busy northbridge...

As for the actual root of this thread: Opteron 170 is a dual core processor at 2.0ghz, but one-ups the X2 with the additional 512kb cache. Combined with the overclocking potential, I say you're right on track. Buy the 170, crank up the speed and enjoy the extra performance.

Edited February 22, 2006 by Albuquerque

[ripken204]

for ramdrives

http://www.anandtech.com/storage/showdoc.aspx?i=2480&p=1

this shows how much faster a ramdrive is than a raptor, you can google for more info

[Albuquerque]

for ramdrives

http://www.anandtech.com/storage/showdoc.aspx?i=2480&p=1

this shows how much faster a ramdrive is than a raptor, you can google for more info

Yes, ramdrives are "faster" than 10,000 RPM drives, and even 15,000 RPM drives. That does not answer the question of how much "faster" it will be for a given application.

If the application you're wanting to speed up is ATTO, then a ramdisk will do quite well. If the application you want to speed up is video encoding, there are better ways to use your ram.