The small, cube-shaped PCs pioneered by Shuttle are becoming almost blasé. We’re starting to see an endless stream of small-form-factor systems based around the Intel 865G chipset or, more recently, socket 754 systems in miniature. Shuttle seems to be the only manufacturer who tries to push the performance envelope in these tiny PCs. The company still produces the only small-form-factor system we’ve seen that uses the Intel 875P chipset. While there are other LGA775 systems based on the 915G chipset now,
So along comes Shuttle with a socket 939 system for users who crave maximum Athlon 64 performance in a small package. Using Nvidia’s Nforce3 Ultra chipset, the SN95G4 fully supports the socket 939 format, including the 128-bit memory controller. Like Napoleon Bonaparte, Shuttle seems to hunger for power in a small package.
So now Shuttle has a pair of fairly modern systems: one based on Intel’s 915G, the other running nVidia’s nForce3 Ultra. We figured it would be fun to compare the two, using CPUs that are roughly in the same ballpark.
Before diving into the nuts and bolts of benchmarking, let’s open up the case and take a look at the innards of the SN95G4.
The SN95G4 uses the by-now-familiar Shuttle heat pipe to cool the processor. However, the heat pipe is modified a bit. Here, we show the SN95G4 heat pipe adjacent to the one used in the older, socket 754-based SN85G4.
The cooling fan is larger—90mm versus 80mm for the SN85G4—to match the larger radiating fin area.
The SN95G4 uses an active cooler over the nVidia north bridge, which draws air over the chip laterally, exhausting it towards the vents in the side of the case. The system also flips the AGP and PCI slot, in much the same manner as other similar Shuttle designs. This allows the graphics card active cooler to exhaust air to the side vents as well.
This particular XPC ships with a 240W power supply. If you want to build a freakishly powerful system, the power supply may be a limiting factor. An Athlon 64 FX-55 burns 105W, and a GeForce 6800GT burns another 65W or so. That’s pushing pretty close to the limit of the power supply. However, our Athlon 64 3500+ and X800XT ran flawlessly, even during heavy LAN party gaming action.
We configured a Shuttle SB81P and the SN95G4 with a set of components as identical as possible. However, the SB81P uses PCI Express, while the SN95G4 is AGP based. Still, both graphics cards were based on ATI’s X800 XT GPU, so only the interfaces differed. Lets’ check out the system configurations:
|
|
SN95G4 System |
SB81P System
|
|
Processor:
|
Athlon 64 3500+ at 2.2 GHz (512KB L2 cache)
|
Pentium 4 model 550 at 3.40GHz
|
|
Motherboard and chipset:
|
Shuttle FB95; nVidia nForce4 Ultra Core Logic
|
Shuttle FB81; Intel 915G Core Logic
|
|
Memory:
|
2 x 512MB Corsair XMS low latency DDR400 at CAS 2-2-2-5
|
2 x 512MB Corsair Twin2X 5300C4Pro DDR2/533 at CAS3-4-4-12
|
|
Graphics:
|
ATI Radeon X800 XT AGP; 256MB Memory; Catalyst 4.10 Drivers
|
ATI Radeon X800 XT PCIe; 256MB Memory; Catalyst 4.10 Drivers
|
|
Hard drive:
|
Seagate Barracuda 160GB SATA
|
Seagate 160GB SATA Drive with support for Native Command Queuing
|
|
Optical drive:
|
ATAPI DVD-ROM Drive
|
ATAPI DVD-ROM Drive
|
|
Audio:
|
Sound Blaster Audigy 2
|
Sound Blaster Audigy 2
|
|
Operating system:
|
Windows XP Professional with SP2
|
Windows XP Professional with SP2
|
We performed a clean install of Windows XP Professional with Service Pack 2. Then we ran the following benchmarks:
- BAPCO SYSmark 2004, which replaces our now-obsolete Winstone suite.
- 3D Studio Max R6 performance, using the SPEC APC test. This benchmark tests both interactive and rendering performance.
- A pair of 3D Studio Max R6 software rendering tests and a LightWave 7.5 rendering test.
- PCMark04, to gauge memory and CPU performance.
- 3DMark05 CPU Test as a memory and CPU check.
- Windows Media Encoder 9 video encode, to check throughput on encoding.
- A multitasking scenario using Norton Antivirus 2003 and Photoshop elements. Photoshop Elements runs a scripted sequence of filters while NAV2003 performs a virus scan of a directory of known size.
- Four games: Unreal Tournament 2003 (ET multiplayer timedemo), Flight Simulator 2004, Doom 3, and Halo.
- SPECviewperf 8.01. This is a graphics intensive test using recordings from a suite of professional 3D applications. It hammers on the graphics interface and actually stresses both AGP and PCI Express a bit more than the game benchmarks.
Note that SYSmark test was run on a clean install, as well as the LightWave and 3D Studio tests. All other benchmarks were installed together on each system.
The Pentium 4 550 outpaces the Athlon 64 3500+ by a substantial margin in the Internet Content Creation test and by a narrower lead in the Office Productivity test. Note that the Athlon 64 3500+ only has 512KB of L2 cache, which may affect its performance in these benchmarks.
PCMark04 Results
PCMark04 is a purely synthetic test suite, although it isolates individual subsystems. The 3500+ suffers in the overall score, but paces the Intel processor closely in the individual CPU and memory scores.
Interestingly, the 3500+ fares very poorly in the 192KB memory tests, even though 192KB blocks should fit within the L2 cache. The results are a bit better in the 4MB block tests, though the AMD CPU still trails its Intel competition.
We used three tests. The first is the SPEC APC 3D Studio R6 benchmark, which tests both interactive performance and rendering performance. The second is a pair of pure 3D Studio rendering tests. Finally, we use LightWave 7.5 to render a scene using radiosity rendering techniques.
The Prescott core has never been strong at software rendering tasks, but seems to have legs against the 3500+. The smaller cache size probably plays an effect here. However, as we’ve seen with faster CPUs, the Athlon 64 performs remarkably well in the SPEC APC 3D Studio Max 6, which models actual application usage.
SPECviewperf 8.01
It’s difficult to gauge how real the gap is between the two CPUs in the SPECviewperf 8.01 benchmark. One reason is that Shuttle uses an Award BIOS, and we’ve seen similar BIOSes on other Pentium 4 systems perform poorly using PCI Express graphics running on top of an Award BIOS. Given a BIOS update for the SB81P, the performance difference is likely to decrease, though by what amount?
The Pentium 4 550 really shows its mettle in multitasking and WMV encoding here, easily outclassing the 3500+ in these tests. The smaller cache and lower clock rate really have an impact on the AMD processor’s performance in these benchmarks.
Here, Intel and AMD swap leads, with the 550 doing slightly better in the synthetic 3DMark05 benchmark. However, in benchmarks based on actual games, the Athlon 64 3500+ simply outclasses the P4. Although the Athlon 64 has a smaller cache, the efficiency of its memory controller and superior x87 floating point probably play major factors in the gaming tests.
The SN95G4 ran all of our benchmarks flawlessly. When equipped with low-latency memory, the system ran Memtest 86+ for a solid hour without a single error, even when set to CAS 2-2-2-5.
We also ran the system for a good six hours during a LAN party, playing Joint Operations and Unreal Tournament 2004. The system never burped or hiccupped even once. So we can pronounce the system stable under severe stress.
How much will this particular configuration set you back? Let’s price both the SN95G4 and SB81P systems we used for our performance testing.
|
Component:
|
SN95G4 System
|
Price
|
SB81P System
|
Price
|
|
Shuttle XPC:
|
SN95G4 barebones
|
$319
|
SB81P barebones
|
$304
|
|
Processor:
|
Athlon 64 3500+
|
$270
|
Pentium 4 550
|
$278
|
|
Memory:
|
1GB Corsair XMS
|
$270
|
1GB Kingston DDR2
|
$295
|
|
Graphics card:
|
AGP X800 XT
|
$450
|
PCIe X800 XT
|
$515
|
|
Sound card:
|
Audigy 2 OEM
|
$65
|
Audigy 2 OEM
|
$65
|
|
Hard drive:
|
Seagate 160GB SATA
|
$115
|
Seagate 160GB SATA
|
$115
|
|
DVD-ROM:
|
Sony DDU-1612
|
$28
|
Sony DDU-1612
|
$28
|
|
Total:
|
|
$1,517
|
|
$1,600
|
The Intel system costs about $80 more than the AMD-based system. For $80 less, you get one hotrod gaming system, but the Intel system will be a better performer on more mainstream applications. Most of the price premium for the SB81P system lies in the memory and the PCI Express graphics card. Both, however, are likely to drop in price over time to achieve price parity with their older cousins. In effect, you’re paying for an uplift for newer technology.
Both systems are good systems, but our focus here is on the SN95G4. While we have some concerns about the 240W power supply, the system was rock solid running a 3500+ and ATI Radeon X800 XT. We do find ourselves wishing that the 3500+ cost a bit less, or had a bigger cache. Still, for $270, you get a true 64-bit processor that can certainly outpace its nearest Intel rival for PC gaming, and doesn’t fall short by too wide a margin in other applications.
|
Product:
|
Shuttle SN95G4 Socket 939 Small-Form-Factor PC
|
|
Company:
|
|
|
Pros:
|
Good performance; stable; quiet running
|
|
Cons:
|
Somewhat pricey; 240W power supply
|
|
Summary:
|
Offering excellent system performance and stability. This is a great socket 939 PC if you don’t need more expansion.
|
|
Price:
|
$319
|
|
Score:
|
|
|
Product:
|
AMD Athlon 64 3500+ at 2.2GHz
|
|
Company:
|
|
|
Pros:
|
Exceptional PC gaming performance for the price
|
|
Cons:
|
Trails the equivalent Pentium 4 in most other applications; 512KB L2 cache
|
|
| |
|
Summary:
|
If you want a great CPU for PC gaming at a reasonable price, the 3500+ is worth a look.
|
|
Price:
|
$270
|
|
Score:
|
|