10GB of RAM in 2010 Quad or Hexcore?

[nanofrog]

The temp sensors are available on 1 and 2gb 1333mhz dims, not just the 4 gb modules. These were Kingston parts on ramshopping.com.

Yes, but they're not required (Intel's design, as the smaller DIMM's wouldn't get hot enough to need them). But mixing them with non-thermal sensored DIMMs would cause problems.

Also I'm pretty sure that if you use 1333MHz memory in a system that runs at 1066MHz, it will simply throttle it back to 1066MHz.

Yes. It would just function at 1066 if that's all that the memory controller was designed for (max mem frequency), or as the 2009 systems are currently, limited by the firmware (i.e. a couple of 1333 compliant CPU's came out, 3.2 and 3.33GHz SP Quads, but still run at 1066 as a result).

 

[bzollinger]

I'm bumping this back up to the top to see if anyone has any new insight about tripple, dual, single channel memory operation.

Will the 3.2GHz quad run in dual channel mode with two 1GB DIMMS from apple and two 4GB DIMMS from transintl? Each pair should be "matched" so I'm really hoping I don't get knocked down to single channel mode.....

 

[lucidmedia]

I received an email last night from a OWC tech that confirmed that 2x4GB + 2x1gb (apple) ram does indeed work.

 

[Novablas]

I received an email last night from a OWC tech that confirmed that 2x4GB + 2x1gb (apple) ram does indeed work.

But 2x 4GB + 3x1gb (apple) ram would not work?

If it has to be pairs then why does 3x1gb work?

or 3x any size?

 

[Icaras]

But 2x 4GB + 3x1gb (apple) ram would not work?

Because 2 + 3 = 5.

The quad/hex only has 4 RAM slots.

If it has to be pairs then why does 3x1gb work?

or 3x any size?

Because you're using the DIMMS in triple channel, which is what the Nehalem/Westmere architecture seems to be designed and optimized for. There are a ton of RAM threads going about right now that go more into detail about this.

 

[Novablas]

Because 2 + 3 = 5.

The quad/hex only has 4 RAM slots.



Because you're using the DIMMS in triple channel, which is what the Nehalem/Westmere architecture seems to be designed and optimized for. There are a ton of RAM threads going about right now that go more into detail about this.

Ah I see, thanks. And I just assumed the Quads/Hexs had more ram slots because my 2006 Mac Pro does. Forgot about them being a single processor.

Wait so do 8GB sticks work in the quads/hexs? If I bought one of those could I then use the 3x1GB that comes default with it for 11gb of ram?

Edit: Oh wait nvm, OWC says their 8GB sticks can only be used with other 8GB sticks.

 

[Icaras]

Ah I see, thanks. And I just assumed the Quads/Hexs had more ram slots because my 2006 Mac Pro does. Forgot about them being a single processor.

Wait so do 8GB sticks work in the quads/hexs? If I bought one of those could I then use the 3x1GB that comes default with it for 11gb of ram?

Edit: Oh wait nvm, OWC says their 8GB sticks can only be used with other 8GB sticks.

Yea np, 8GB DIMMS are definitely confirmed. You just have to use all OWC sticks like you mentioned 😛

Also, i chatted with an OWC rep last night and he confirmed to me that you cannot use just a single 8 DIMM. You have to start with two DIMMS at the very least.

 

[bzollinger]

I received an email last night from a OWC tech that confirmed that 2x4GB + 2x1gb (apple) ram does indeed work.

Awesome!! Thanks for the response. So it works which is great, but did they say anything about dual/single channel mode?

Can anyone confirm this? I can't afford 3x8GB so I'm looking into dual channel mode for the Nehalem procs.

If the two pairs will interleave (dual channel I think) then anyone doing this will be in good shape. Granted it won't be as "fast" as triple channel. I can forego a 2% theoretical increase in speed to get my MP up to 10GB from base for $250😱

 

[dolz]

The used Mac Pro 1,1 came with 1GB of memory. I added another 8GB bought from OWC. I had no problems at all. In fact today I just added another 1GB of Apple memory that I got on sale from OWC to fill out my risers.

DIMM Riser A/DIMM 1: 2GB, DDR-2 FB-DIMM, 667 Mhz, Manufacturer 0x05F7
DIMM Riser A/DIMM 2: 2GB, DDR-2 FB-DIMM, 667 Mhz, Manufacturer 0x05F7

DIMM Riser B/DIMM 1: 2GB, DDR-2 FB-DIMM, 667 Mhz, Manufacturer 0x05F7
DIMM Riser B/DIMM 2: 2GB, DDR-2 FB-DIMM, 667 Mhz, Manufacturer 0x05F7

DIMM Riser A/DIMM 3: 512MB, DDR-2 FB-DIMM, 667 Mhz, Manufacturer 0x80AD
DIMM Riser A/DIMM 4: 512MB, DDR-2 FB-DIMM, 667 Mhz, Manufacturer 0x80AD

DIMM Riser B/DIMM 3: 512MB, DDR-2 FB-DIMM, 667 Mhz, Manufacturer 0x802C
DIMM Riser B/DIMM 4: 512MB, DDR-2 FB-DIMM, 667 Mhz, Manufacturer 0x802C

 

[bzollinger]

I just read this and believe it has finally answered this question. In short the 2010 SP MP 3.2GHz quad should have no problem running 10GB of RAM in dual channel mode. Furthermore the benefits of triple channel memory are negligible. 😎

http://macperformanceguide.com/Reviews-MacProWestmere-MemoryBandwidth.html

 

[Chenier]

Visual Evidence

Find out for myself today, though others have already confirmed... 3.2 ghz quad core running with 2 x 4gb ram from OWC + 2 x 1gb ram from Apple = 10 gb ram

 

[bzollinger]

Find out for myself today, though others have already confirmed... 3.2 ghz quad core running with 2 x 4gb ram from OWC + 2 x 1gb ram from Apple = 10 gb ram

It's great to know that it works. Can you confirm that it's running in dual channel mode?
thanks,
Ben

 

[johnnymg]

I'm pretty sure that as long as the memory is of compatible spec in regards to, dual or triple channel kits (matched DIMMS), ECC, proper speed (1066 or 1333MHz), and installed in the proper slots it will work.

As gaspra states if you get the order wrong the MP is smart enough to know, and will guide you to the proper placement.

Also I'm pretty sure that if you use 1333MHz memory in a system that runs at 1066MHz, it will simply throttle it back to 1066MHz. So really if the price is the same why not buy the faster RAM?

FWIW: Diglloyd verified that the hex can be used with 1066 RAM:
http://macperformanceguide.com/Reviews-MacProWestmere-MemoryBandwidth.html

Would be interesting to see if the Hex can handle mixed RAM speed modules. 😛.

 

[Chenier]

It's great to know that it works. Can you confirm that it's running in dual channel mode?
thanks,
Ben

Pardon my naïveté, but how/where do I check for this?

 

[Fiete5401]

Find out for myself today, though others have already confirmed... 3.2 ghz quad core running with 2 x 4gb ram from OWC + 2 x 1gb ram from Apple = 10 gb ram

Hi Chenier!

Could you please run 64-bit Geekbench on your new MP and post your results to the Geekbench database? Your results would help me and others in their buying decision. Thanks!

 

[bzollinger]

Pardon my naïveté, but how/where do I check for this?

I'm sorry I don't know how to check for that. I've searched for a way or a utility but haven't found one yet.

Maybe someone else here know....

 

[Chenier]

Hi Chenier!

Could you please run 64-bit Geekbench on your new MP and post your results to the Geekbench database? Your results would help me and others in their buying decision. Thanks!

Would love to help, but finances are rather tight after all the upgrades. Is there a free alternative to Geekbench that would satisfy your curiosity?

 

[Fiete5401]

I'm sorry I don't know how to check for that. I've searched for a way or a utility but haven't found one yet.

Maybe someone else here know....

Maybe the Memory Configuration Tool (I think it is named like that in english) can help.
You can find it in /System/Library/CoreServices.

Would love to help, but finances are rather tight after all the upgrades. Is there a free alternative to Geekbench that would satisfy your curiosity?

Well, in this case please just run the 32-bit version. As there are no results at all for the 3.2 quad every single bit of information helps.

 

[Chenier]

Maybe the Memory Configuration Tool (I think it is named like that in english) can help.
You can find it in /System/Library/CoreServices.

When I run this (Memory Slot Utility) I only get this simple message:

Well, in this case please just run the 32-bit version. As there are no results at all for the 3.2 quad every single bit of information helps.

Done. For what the 32-bit version is worth, the results are here:
http://browse.geekbench.ca/geekbench2/view/281986

Not familiar with these numbers yet; does this sound ok for my machine specs?

 

[bzollinger]

When I run this (Memory Slot Utility) I only get this simple message:
View attachment 247977



Done. For what the 32-bit version is worth, the results are here:
http://browse.geekbench.ca/geekbench2/view/281986

Not familiar with these numbers yet; does this sound ok for my machine specs?

That sounds pretty good to me!! I have only ran the 32-bit version too. My recently retired Dual 2.0 G5 ran ~1800, and my 2.53GHz late 09 Mini ran ~3500.

Does that put it into perspective a little?

I wonder what the 64-bit results would be? Anyone care to speculate?

 

[Fiete5401]

I wonder what the 64-bit results would be? Anyone care to speculate?

I calculated that the 64-bit results are normally around 13% higher than the 32-bit results. That would result in an extrapolated score in the range of 11500.
So far I can tell you that the 3.2 quad gives you would you pay for if you just compare the Geekbench results and the price difference to the 2.8 quad. You pay 16.7% more and get 16.6% more performance.
It is interesting to see that the 3.2 quad has the highest Geekbench score per core of all 2010 MPs.
If we just look at "bang for the buck" the 3.33 hex and the 2.66 dodeca are the clear winners.
The 2.93 dodeca is the problem child of the family. You pay 24.5% more than the 2.66 dodeca but you only get a 9.1% higher score.

All results are based on comparisons of 64-bit Geekbench results which are nothing more than synthetic and not real world observations.
These results can't reflect any real world workflows but I think they can give you an insight about what can be done with these machines if the software can utilize all the power that's buried in them.

 

[barefeats]

I tested the 3.33GHz 6-core with two different memory configs. Then I ran a 64 bit memory stress test.

4x8G = 32G = 10.5 GB/s
3x8G = 24G = 14.3 GB/s (or 36% faster)

However, when I ran After Effects CS5 project render using 12 cores and 1.5G per core, the 32G config took 72 sec, the 24G took 71 sec. In other words, if there is a difference in real world performance, you would need a stopwatch to detect it.

Why? Because few if any real world apps use the full memory bandwidth. I'm still testing. If I find any app that benefits from the 3 stick config, I'll report it here.

 

[Fiete5401]

Because few if any real world apps use the full memory bandwidth. I'm still testing. If I find any app that benefits from the 3 stick config, I'll report it here.

It's curious to see that most people don't realize that fact. Thanks for your efforts.

 

[bzollinger]

I tested the 3.33GHz 6-core with two different memory configs. Then I ran a 64 bit memory stress test.

4x8G = 32G = 10.5 GB/s
3x8G = 24G = 14.3 GB/s (or 36% faster)

However, when I ran After Effects CS5 project render using 12 cores and 1.5G per core, the 32G config took 72 sec, the 24G took 71 sec. In other words, if there is a difference in real world performance, you would need a stopwatch to detect it.

Why? Because few if any real world apps use the full memory bandwidth. I'm still testing. If I find any app that benefits from the 3 stick config, I'll report it here.

Thanks for the info. That is an interesting test for sure.

 

[lucidmedia]

I calculated that the 64-bit results are normally around 13% higher than the 32-bit results. That would result in an extrapolated score in the range of 11500.
So far I can tell you that the 3.2 quad gives you would you pay for if you just compare the Geekbench results and the price difference to the 2.8 quad. You pay 16.7% more and get 16.6% more performance.
It is interesting to see that the 3.2 quad has the highest Geekbench score per core of all 2010 MPs.
If we just look at "bang for the buck" the 3.33 hex and the 2.66 dodeca are the clear winners.
The 2.93 dodeca is the problem child of the family. You pay 24.5% more than the 2.66 dodeca but you only get a 9.1% higher score.

All results are based on comparisons of 64-bit Geekbench results which are nothing more than synthetic and not real world observations.
These results can't reflect any real world workflows but I think they can give you an insight about what can be done with these machines if the software can utilize all the power that's buried in them.

My 3.33 hexacore with 10GB ram (4x2, 1x2) got a 16135 in geekbench 64 bit... Its posted in the the results browser.