Geekbench Score Worse than slower computer?!

[Zaithe]

Here's my current benchmarks in 64 bit:

http://browser.primatelabs.com/geekbench3/7274906



And in 32bit:

http://browser.primatelabs.com/geekbench3/7274925



But these scores are better.. with a slower processor and less ram.

http://browser.primatelabs.com/geekbench3/1103986



I have a 12 core 3.46Ghz processor with 64GB 1333mhz ddr3 ram with a PCIe SATA 3 controller as the main disk going near 800MB/s

 

[Stageshoot]

Your Multicore Memory Performance is about identical to your single core (Memory not cpu) scroll down the listing to see it

Multicore memory performance is normally about double single as you can see in the other test you pointed to.

So thats something to start looking at

 

[Maxx Power]

Here's my current benchmarks in 64 bit:

http://browser.primatelabs.com/geekbench3/7274906



And in 32bit:

http://browser.primatelabs.com/geekbench3/7274925



But these scores are better.. with a slower processor and less ram.

http://browser.primatelabs.com/geekbench3/1103986



I have a 12 core 3.46Ghz processor with 64GB 1333mhz ddr3 ram with a PCIe SATA 3 controller as the main disk going near 800MB/s

The biggest difference is the memory portion, for multi-core tests. My guess by looking at your memory amount (64GB) and the other config (24GB) is that the (24GB) Mac Pro is properly configured as balanced triple-memory-channel (24/3 = 8GB per channel, so that probably means 4GB Dimms per slot across 6 slots), while yours is not optimal (64/3 does not result in integers, so your Dimms are such that one channel has bigger sticks than others).

May be Geek Bench is reflecting this ?
[doublepost=1468047865][/doublepost]Here is what Apple says about these machines:
"
Each processor’s memory
controller has three
memory channels. DIMM
slots 1, 2, 5, and 6 have
their own channels; slots
3 and 4 share a channel
and slots 7 and 8 share a
channel.

For optimal memory
bandwidth, all six memory
channels should be used,
and memory should be
balanced across the six
channels."

See "https://manuals.info.apple.com/MANUALS/0/MA918/en_US/Mac_Pro_2009_2010_Memory_DIMMs_DIY.pdf".
[doublepost=1468048177][/doublepost]Also have a look at this comparison test of balanced vs. proper triple-channel setup in terms of memory bandwidths:


http://macperformanceguide.com/Reviews-MacProNehalem-Tests-Memory.html

Quote: "
Memory copy speed is 52% faster with six modules instead of eight (triple vs dual channel);
Double-sided modules are about 10% faster than single-sided modules;
Six modules is 68% faster than 3 modules (this might not matter if only a few cores are in use, but consider the 4-core Mac Pro with only 4 memory slots);"

 

[Zaithe]

Thanks guys! This is very helpful. Guess i will need to get 2 more 16GB sticks. Here is the ram i have been using for reference:
http://www.kingston.com/dataSheets/KVR13LR9D4_16.pdf

Do you have any other advice that could boost performance? Could i safely overclock the processor?

Cpu temps right now are 45C and another is 36C. Plan on actually using thermal Surface Purifier and thermal paste remover with Artic Silver 5 High density Polysynthetic silver compound to get it a little cooler.

 

[Maxx Power]

Thanks guys! This is very helpful. Guess i will need to get 2 more 16GB sticks. Here is the ram i have been using for reference:
http://www.kingston.com/dataSheets/KVR13LR9D4_16.pdf

Do you have any other advice that could boost performance? Could i safely overclock the processor?

Cpu temps right now are 45C and another is 36C. Plan on actually using thermal Surface Purifier and thermal paste remover with Artic Silver 5 High density Polysynthetic silver compound to get it a little cooler.

What is your current memory configuration ? What are you targeting your machine for (usage scenario) ?

It really depends on if you need memory speed vs. memory quantity. Also do keep in mind that Geekbench memory scores don't necessarily translate well into real world usage performance.

If you are doing any professional work, I would not recommend that you overclock, even if you could on the Mac Pro.

I guess you currently have 4 x 16GB sticks spread across 4 out of the 8 DIMM slots ?

 

[AidenShaw]

Thanks guys! This is very helpful. Guess i will need to get 2 more 16GB sticks. Here is the ram i have been using for reference:
http://www.kingston.com/dataSheets/KVR13LR9D4_16.pdf

Do you have any other advice that could boost performance? Could i safely overclock the processor?

Cpu temps right now are 45C and another is 36C. Plan on actually using thermal Surface Purifier and thermal paste remover with Artic Silver 5 High density Polysynthetic silver compound to get it a little cooler.

How do your real applications perform?

Memory test benchmarks are deliberately written to more or less "disable" the CPU caches and measure the raw DIMM speeds.

Well written applications try to make the best use of the caches.

This means that memory configs that are significantly different on a memory benchmark might perform nearly the same on an application benchmark.
_____________

Read this thread https://forums.macrumors.com/thread...th-various-mem-configs.1704700/#post-18745317 where an MP6,1 was tested for application performance with one to four DIMMs. All configs were virtually the same performance except for the memory test and two others.

 

[Zaithe]

Yeah i have 4 spread out in DIMM slots 1-4. My applications are performing great but i want to get the most performance out of this machine as possible by setting up everything correctly. It's mostly a workstation and gaming rig using bootcamp and will probably retire into a server.

I have a PCIe 2.0 x16 m.2 controller with 4 ACHI blades which i plan on turning into a 4x 512GB raid 0 using sm950's. Want to also make sure i'm not going to bottleneck the m.2 carrier card anywhere. Theoretical max speed is 8GB/s in a PCI 2.0 x16 slot.

Going to throw in my GTX 970 after it comes back. It's being EFI flashed. I'll install it into my x8 slot. It's x16 but benchmarks show only a 2% increase in performance by putting it in the x16 slot.

I also have a 1TB sata 3 controller which maxes out at 750MB/s i will put in a x4 slot. and a USB 3.0 card in the other x4 slot.

 

[Stageshoot]

Yeah i have 4 spread out in DIMM slots 1-4. My applications are performing great but i want to get the most performance out of this machine as possible by setting up everything correctly. It's mostly a workstation and gaming rig using bootcamp and will probably retire into a server.

I have a PCIe 2.0 x16 m.2 controller with 4 ACHI blades which i plan on turning into a 4x 512GB raid 0 using sm950's. Want to also make sure i'm not going to bottleneck the m.2 carrier card anywhere. Theoretical max speed is 8GB/s in a PCI 2.0 x16 slot.

Going to throw in my GTX 970 after it comes back. It's being EFI flashed. I'll install it into my x8 slot. It's x16 but benchmarks show only a 2% increase in performance by putting it in the x16 slot.

I also have a 1TB sata 3 controller which maxes out at 750MB/s i will put in a x4 slot. and a USB 3.0 card in the other x4 slot.

The 4.1/5.1 has 2x16 slots and 2x4 there are no x8 slots so not sure where you are coming from on that one re GPU performance.

 

[Zaithe]

Thanks for the correction. I've been up for 20 hours and am starting to get forgetful. Out of curiosity, the motherboard chipset for my motherboard is the Intel 5500, http://ark.intel.com/products/36784/Intel-5500-IO-Hub

Why does it say that it only supports PCIe 1.1, two x8 and two x4 PCIe slots? How did they pull off having it support two x16's and two x4's at PCIe 2.0? http://ark.intel.com/products/36784/Intel-5500-IO-Hub

 

[Maxx Power]

Yeah i have 4 spread out in DIMM slots 1-4. My applications are performing great but i want to get the most performance out of this machine as possible by setting up everything correctly. It's mostly a workstation and gaming rig using bootcamp and will probably retire into a server.

I have a PCIe 2.0 x16 m.2 controller with 4 ACHI blades which i plan on turning into a 4x 512GB raid 0 using sm950's. Want to also make sure i'm not going to bottleneck the m.2 carrier card anywhere. Theoretical max speed is 8GB/s in a PCI 2.0 x16 slot.

Going to throw in my GTX 970 after it comes back. It's being EFI flashed. I'll install it into my x8 slot. It's x16 but benchmarks show only a 2% increase in performance by putting it in the x16 slot.

I also have a 1TB sata 3 controller which maxes out at 750MB/s i will put in a x4 slot. and a USB 3.0 card in the other x4 slot.

Well, if you are going to get memory at any time at all, you might as well do it now. DDR3 is super cheap right now and I don't foresee DDR3 1333 to be even cheaper down the line as production will soon cease to fully transition into DDR4s. Although you'll still only see a difference if you end up utilizing those two additional memory channels (which are tied to the 2nd CPU, AFAIK, as the first one is fully maxed out). I suppose you'll do something like slots (1,2,3) and (5,6,7) fully occupied as per Apple.

 

[h9826790]

It seems that you are quite new to Mac Pro, or Mac.

1) There is no tools to OC the CPU, regardless in Windows or OSX. Unless you can develop one, there is no way to do that, no matter safe of not.

2) How do you know the chipset is 5500? You remove the NB heat sink and check the chipset model? And no matter what the chipset is, there are 2 x16 slots and 2 more slots which share x4 bandwidth, but physically support x16 PCIe card.

3) My test shows that a HD7950 has about 2.4% performance penalty by installing in the x4 slot. That's inline with your result. I put my 2nd 7950 in slot 3 for much better cooling. However, if no specific reason, it's always recommended to put the GPU in slot one. That can avoid the graphic card blocking another slot.

4) GeekBench is just a benchmark, that may be much to do with the real world performance. Not optimising the triple channel can take a performance hit, some background process can cause lower score. As long as the score is within 10% variation, I will say that's a good enough proof that your Mac can perform normally. No need to chase the highest score.

5) 45C and 36C looks quite normal to me (depends on the ambient temperature), I doubt if you can make any significant improvement by re-applying the thermal paste.

 

[Zaithe]

Here is where i got the chipset from:
https://en.wikipedia.org/wiki/Mac_Pro#Specifications

Would splitting up my memory sticks right now from only using 1-4 to using 1-2 and 5-6 make an impact right now? I will have to wait until next month to get more. I spent all my available money already getting what i have right now.

 

[Maxx Power]

Here is where i got the chipset from:
https://en.wikipedia.org/wiki/Mac_Pro#Specifications

Would splitting up my memory sticks right now from only using 1-4 to using 1-2 and 5-6 make an impact right now? I will have to wait until next month to get more. I spent all my available money already getting what i have right now.

I am fairly certain it would result in an increase in memory bandwidth, I think. Apple actually recommends that if you have only 4 DIMMS, spread it across 1-2 and 5-6 in the user manual.

E.g., like this according to Apple's memory upgrade manual:

What I am not certain of, is how this will change the Geek Bench scores. Presumably it will go up...

 

[h9826790]

Here is where i got the chipset from:
https://en.wikipedia.org/wiki/Mac_Pro#Specifications

Would splitting up my memory sticks right now from only using 1-4 to using 1-2 and 5-6 make an impact right now? I will have to wait until next month to get more. I spent all my available money already getting what i have right now.

There was a guy here who actually remove the NB heatsink and took pictures of both the single processor and dual processor cMP, from memory, end up the chip are different. And the picture clearly show the chip model number. May be the search function can help to locate that thread, and we can have a more accurate info than WiKi to work on.

 

[Maxx Power]

There was a guy here who actually remove the NB heatsink and took pictures of both the single processor and dual processor cMP, from memory, end up the chip are different. And the picture clearly show the chip model number. May be the search function can help to locate that thread, and we can have a more accurate info than WiKi to work on.

The single processor and multi-processor chipsets are indeed different. I think the single proc uses the X58 chipset while the multi-proc uses the 5500 series (could be 5520, which has more PCI-E lanes that are x16).

Found it for you (multi-proc 2009 MP):

Single proc 2009 MP:

 

[h9826790]

The single processor and multi-processor chipsets are indeed different. I think the single proc uses the X58 chipset while the multi-proc uses the 5500 series (could be 5520, which has more PCI-E lanes that are x16).

Found it for you (multi-proc 2009 MP):

Single proc 2009 MP:

Thanks! This is exactly what I want to find. And 5520 has 36 PCIe 2.0 lanes, so, that perfect match the cMP's config. 2 independent x16 slot + 2 slots share the rest x4 lanes. Which obviously the 5500 cannot do with just 24 lanes.

 

[Zaithe]

Thanks! This is exactly what I want to find. And 5520 has 36 PCIe 2.0 links, so, that perfect match the cMP's config. 2 independent x16 slot + 2 slots share the rest x4 links. Which obviously the 5500 cannot do with just 24 links.

Has anyone checked the multi-CPU chipset for the 5,1 model? 2 x16 slots plus 2 x4 slots would be 40 lanes. Take one of those slots away and you have 36. So theoretically it seems the chipset can't support all it's slots using full capacity let alone a PCIe expansion kit to add more slots.

 

[h9826790]

Has anyone checked the multi-CPU chipset for the 5,1 model? 2 x16 slots plus 2 x4 slots would be 40 lanes. Take one of those slots away and you have 36. So theoretically it seems the chipset can't support all it's slots using full capacity let alone a PCIe expansion kit to add more slots.

The chipset is 5520 as per the picture. And if you read my post carefully, I said twice, slot 3 and slot 4 SHARED the same x4 PCIe lanes. So, only 36 lanes required in total.

Slot 3 and 4 are connected via a PCIe switch. Slot 1 and 2 are independent, but 3 and 4 are not.

 

[Zaithe]

Thank's for the clarification. I updated the wiki page to have the correct chipset.