SmallDog.com has refurbished towers, laptops, displays, etc., with 12 month warranty.
New DDR 1GHz MP Power Mac versus "Old" SDRAM version Originally posted 8/15/02 by rob-ART morgan, mad scientist Updated 8/17/02 with 'streamlined' graphs, tweaked verbiage, and addition of a very interesting Memory Speed Graph 8/20/02 -- Page TWO added with additional tests using subsystems. Updated 8/27/02 -- Page TWO Quake3 results using three different graphics cards on both DDR and SDR systems Mahalo to my remote mad scientist, Greg, for results from his new DDR Dual 1GHz. (Note to skeptics that doubted Greg's DDR results: On 8/26/02 we re-ran every test in the Bare Feats local lab using a different DDR 1GHz/MP Power Mac with even more memory (1.25GB) and got the same results as he got. This report is not a mistake, not a fake, and not a fluke as some suggested.) Does the new 2.7GB/s DDR memory bus and 167MHz system bus make a difference? Has we taken a leap forward in Power Macs overall system throughput? I figured the best way to tell is to run the NEW DDR Dual 1GHz Power Mac against the "obsolete" SDR (short for SDRAM) Dual 1GHz Power Mac (and the DDR based Xserve Dual 1GHz).               CONCLUSION To my surprise and chagrin, the new Power Mac with DDR RAM has no performance advantage over the old SDR Power Mac running at the same clock speed. The 25% faster system bus seems of no help, either. Depressing. Scandalous! FOR MORE TEST RESULTS THAT STRESS THE DISK AND GRAPHICS SUBSYSTEMS, TURN TO PAGE TWO. 10/20/02 -- MacAddict published their own shootout between a DDR 1GHz G4 and an SDR 1GHz G4. They showed a significant difference in speed but I believe their tests are faulty. All three tests involve the hard drive. Their two test machines had a different model of hard drive. Therefore, the results should be different. In the BARE FEATS tests, I used the same 120GXP hard drive in both machines. The results were virtually identical. Where's the bottleneck? Here are some theories from readers on why the new DDR machine isn't any faster: 1. The two processors share a 1.3GB/s pipe to Apple's custom AGP/Memory controller. So the 2.7GB/s memory gets "starved." 2. The main problem hobbling the PowerMac is the CPU to System Controller Bus, i.e. the System Bus. What is "starved" is not the memory. What is starved are the CPUs of the PowerMac. They are capable of processing much faster than the system can feed it data. This has been a known, nagging, problem with the G4 processor in its current implementation. By design, the PowerPC 7450 used in current PowerMacs CANNOT receive data at faster than 1.3 GBPS speed. And if you use DUAL or QUAD CPUS, they are FORCED to SHARE the bus, slowing down each processor when doing data intensive work. This is why Apple cannot come out with a QUAD processor machine. The four CPUs have to share the same thin 1.3 GBPS bus!!!!. Talk about starving CPUs even more!!! Ideally, the two CPUs should get separate busses to get data. But they cannot. It's forced by Motorola's current design. Note that Pentium 4 also has the same problem - having to share the same system bus among dual CPUs (however, the bus is fatter). Motorola's PowerPC G4 cannot use a faster bus. Only the Athlon can take advantage of using a separate bus per CPU. 3. The DDR Power Mac has a smaller (faster) L3 cache (1MB per cpu versus the SDR's 2MB per cpu). (I am not convinced this is a big factor. Why? Because the DDR based Xserve has 2MB per CPU just like the SDR Power Mac and it's no faster. See memory graph below for more on this. -- rob-ART) 4. The DDR Power Mac is running Jaguar while the SDR Power Mac and Xserve are running Puma (10.1.5). (As you can see above, we re-tested using 10.2 on the SDR Power Mac to shoot holes in that theory -- rob-ART) 5. The advantages of the DDR Power Mac will show only if you load up all the buses (CPU, Memory, PCI, FireWire, AGP, etc.). (Though all the tests apps we used exercise the most of the PCI and AGP buses to some degree, they are primarily CPU and Memory intensive. See PAGE TWO tests for more on graphics and disk stressing. -- rob-ART) Here's a very interesting graph created with data generated by an app named MemPerf written by Basil Achermann. It reveals what's going on with L1 cache, L2 cache, L3 cache, and regular RAM:   The graph above shows that the SDR Mac is as much as 81% faster for several sequential 0.5-2MB data accesses over the same data or whenever its larger L3 cache is used a lot. The DDR Mac can load 21% faster from the main memory but only because of the 25% faster bus speed, not because of the doubled data rate (proof for the bottleneck). Real world applications use both L3 cache and RAM, and that's why they all score about the same on both machines. What will it take to fix the bottleneck? 1. A CPU than can handle the full speed of DDR memory (like the fabled PPC 7470). 2. A redesigned motherboard with separate bus for each CPU. Will the Dual 1.25GHz Power Mac be 25% faster as the specs imply? Will it smoke the Pentium 4 2.53GHz system as Apple claims? We plan to answer those and other questions in about 6 weeks... when the top model ships. Meanwhile, we have already started testing on the Pentium 4 2.53GHz system. Here's a chart comparing the features and specs of the three Dual 1GHz systems: DDR Power Mac SDR Power Mac Xserve MP CPU Clock Speed 1GHz *2 1GHz *2 1GHz *2 System Bus Speed 1.3GB/s 1GB/s 1.1GB/s Memory Bus Speed 2.7GB/s (DDR) 1GB/s 2.1GB/s (DDR) Maximum Memory 2.0GB 1.5GB 2.0GB L3 Cache 1MB per CPU ** 2MB per CPU 2MB per CPU L3 Cache Throughput 4.6GB/s 4GB/s 4GB/s Standard Graphics Card Radeon 9000 AGP GeForce4 MX AGP (64MB DDR) ATI PCI (32MB DDR) * Optional Graphics GeForce4 Titanium AGP GeForce4 Titanium AGP Radeon 8500 AGP (Apple) or any SHORT AGP card you can obtain ATA drive bus speed 66MB/s and 100MB/s (two controllers) 66MB/s 100MB/s ATA drive connections 4 2 4 w/ SMART support full length PCI slots 4 4 2 PCI bus speed 33Mhz 33MHz 66MHz FireWire Ports 2 2 3 Gigabit Ethernet Ports 1 1 2 * Price for Dual 1GHz Base Model $2495 (out of production) $3999 * Combo PCI/AGP short slot can be used for short AGP graphics card or Gigabit Ethernet card but for both. Full length GeForce4 Titanium won't fit. And you'll need a special AGP riser for the short card if you decide to upgrade. The default factory config is a generic ATI PCI graphics card which seems to be on the level of a Radeon 7000. ** The Power Mac G4/1.25GHz MP will ship with 2MB L3 per CPU   RELATED LINKS Apple has published some test results of their own using Photoshop, Final Cut Pro, and DVD encloding (iDVD?). You can download the Technical Overview PDF with these results. They haven't responded yet to my request for details as to how these tests were run so we can't duplicate or verify the results. PowerLogix published an in-depth white paper discussion (PDF ) comparing single data rate static RAM ("SDR") architecture versus double data rate ("DDR") when designing the level 3 cache circuitry for use with the latest Motorola G4/745x processors. Don't forget to TURN TO PAGE TWO and THREE for more DDR vs SDR test results. If you buy a new DDR Power Mac, make sure you specify DDR PC2700 333MHz (Non ECC) 64x64 CL2.5 memory. I found the 512MB modules at Data Memory Systems for $135 (part number DM50 609). Check also with TransIntl.com but be sure to specify "PC2700" since they don't list the speed on their website.   TEST NOTES The "SDR" Power Mac 1GHz MP had 1GB of PC133 CL2 SDRAM. The "DDR" Power Mac 1GHz MP had 1.25GB of PC2700 CL2.5 DDR RAM The "DDR" Xserve 1GHz MP had 1.5GB of PC2100 CL2.5 DDR RAM All three were running from an IBM 120GXP drive. For details on each real world test, read "HOW I TEST."		Search this site powered by FreeFind HOME PAGE SPEED TEST RESULTS by Category LINKS to SPEED tests on other sites HOT DEALS on speed upgrades rob-ART morgan, mad scientist and webmaster         Has Bare Feats helped you? Say "thanks"... with a donation.
© 2002 Rob Art Morgan "BARE facts on Macintosh speed FEATS" Email webmaster at rob-art@barefeats.com (Bare Feats is hosted on a G4 Power Mac server by MacDock.com)