Floating Point Performance - SPEC CFP2000
base number (peak number)
Test
Athlon/AMD 760 (PC2100 DDR)
P3/i840 (PC800 Dual RDRAM)
Athlon/VIA KT133 (PC133)
P3/i815 (PC133)
168.wupwise
334 (476)
348 (360)
316 (437)
346 (350)

Let's start with the 168.wupwise test, off the bat the i840 with its Dual Channel RDRAM comes out ahead. This test is more of a computational test than one that deals with a large data set which would stress memory bandwidth performance. In this case, both the Intel platforms come out ahead of the two AMD platforms. This helps to eliminate the platform as the reason for the lower scores, and pinpoints the CPU as being at fault. The only explanation that exists here is an architectural advantage of the Pentium III over the Athlon.

Turning on peak optimizations changes things a bit as AMD can now take advantage of more aggressive optimizations. Only with the upcoming 5.0 Compiler from Intel will the Pentium III on the 840 be able to catch up.

Floating Point Performance - SPEC CFP2000
base number (peak number)
Test
Athlon/AMD 760 (PC2100 DDR)
P3/i840 (PC800 Dual RDRAM)
Athlon/VIA KT133 (PC133)
P3/i815 (PC133)
171.swim
763 (763)
439 (438)
496 (488)
285 (285)

While the wupwise test was very floating point intensive and not very stressful on the memory bus, this next test, 171.swim is the exact opposite. This test is a weather prediction program that works with a huge 1335 x 1335 array of data run over 512 distinct timesteps.

The Athlon does hold a bit of an advantage here as it has a larger total cache than the Pentium III (384KB vs 256KB) so more of the data being manipulated can be stored in its cache, but once that cache is full the performance drop is enormous since the CPU must rely on the slow path to system memory.

The Athlon on a VIA KT133 is thus a little faster than the Pentium III on the i840 because of its larger L2 cache, however with the newer Intel 5.0 Beta compiler the 840 would be virtually on par with the KT133. The i815 gets penalized the most here since it has the same amount of memory bandwidth as the KT133 but it lacks the Athlon's large on-die cache.

Now take a look at the Athlon on the AMD 760 with a full 2.1GB/s of CAS2.5 PC2100 DDR SDRAM. The performance is no less than 54% greater than the second fastest KT133. Even with the new Intel 5.0 Compiler Beta the Athlon on the AMD 760 would enjoy at least a 50% performance advantage. The high latency of RDRAM, even in spite of the dual channel nature of the i840's implementation of it, is too high to compete with DDR SDRAM in this case.

The peak performance does not change, further supporting the idea that this is mostly a cache/memory test and there's very little you can do to make it perform better on any given CPU other than increasing the clock speed or dramatically changing the architecture.

Floating Point Performance - SPEC CFP2000
base number (peak number)
Test
Athlon/AMD 760 (PC2100 DDR)
P3/i840 (PC800 Dual RDRAM)
Athlon/VIA KT133 (PC133)
P3/i815 (PC133)
172.mgrid
254 (346)
251 (251)
238 (277)
208 (208)

172.mgrid is another heavy computational test that doesn't really stress memory bandwidth to the degree that the 171.swim test did. There is much less of a focus on memory latency as there is on memory bandwidth in this test. If you'll notice, the Pentium III on the i815 comes in last, however with the 3.2GB/s of memory bandwidth on the i840 the Pentium III very quickly becomes faster than the Athlon on the KT133 platform. It takes the added bandwidth of DDR SDRAM to give AMD the slight lead here over the more expensive 840. It seems like AMD's large L1 and L2 caches come in handy yet again as latency isn't a major factor in this benchmark, rather pure memory bandwidth.

The peak SPECfp2000 numbers indicate that there is still much room for the Athlon to grow in terms of performance if optimized properly for the CPU.

Floating Point Performance - SPEC CFP2000
base number (peak number)
Test
Athlon/AMD 760 (PC2100 DDR)
P3/i840 (PC800 Dual RDRAM)
Athlon/VIA KT133 (PC133)
P3/i815 (PC133)
173.applu
377 (394)
248 (249)
282 (303)
219 (221)

The APPLU benchmark plots the solution of five Partial Differential Equations on a 3D-grid. In this case you're dealing with quite a bit of data yet again, however the Athlon architecture in general seems to be providing quite a bit of the performance advantage over the Pentium III here. For example, the Athlon on the KT133 is 14% faster than the Pentium III on the i840. Drop the Athlon in an AMD 760 board with PC2100 DDR SDRAM and a 266MHz FSB and all of the sudden you've got a 52% advantage over Intel's fastest solution.

The performance benefit here comes from a combination of the DDR SDRAM and the increased FSB. For the two Intel platforms, peak optimizations did not do much however on both of the Athlon platforms the more aggressive optimizations improved performance by a noticeable degree.

Floating Point Performance - SPEC CFP2000
base number (peak number)
Test
Athlon/AMD 760 (PC2100 DDR)
P3/i840 (PC800 Dual RDRAM)
Athlon/VIA KT133 (PC133)
P3/i815 (PC133)
177.mesa
256 (390)
391 (391)
256 (382)
387 (386)

The situation deviates from the "norm" (at least for the past few benchmarks) with the Mesa test. The Mesa test performs the following operation:

"The input data is a 2D scalar field. The scalar data is mapped to height, creating a 3D object with explicit vertex normals. Contour lines are mapped onto the object as a 1D texture."

The description of the test has SSE written all over it, and considering that Intel's SSE optimizations are superior to the current state of AMD's 3DNow! it doesn't seem surprising that the Pentium III completely tramples over the Athlon here. Only when using aggressive optimizations (peak) can the Athlon come close to the performance of the Pentium III, however the new 5.0 Intel Compilers should help extend that lead even further.

There is almost no stress here on memory performance as the i840 and i815E perform within 1% of one another and the AMD 760 and KT133 scores are identical. It just goes to show you that DDR won't be your savior in all situations.

Floating Point Performance - SPEC CFP2000
base number (peak number)
Test
Athlon/AMD 760 (PC2100 DDR)
P3/i840 (PC800 Dual RDRAM)
Athlon/VIA KT133 (PC133)
P3/i815 (PC133)
178.galgel
533 (532)
292 (295)
451 (429)
277 (281)

The 178.galgel test is a test in computational fluid dynamics and inherently benefits from the Athlon's superior FPU, thus giving the Athlon on the KT133 the immediate lead over both the i840 and the i815 platforms. The combination of the 266MHz FSB and the PC2100 DDR SDRAM gives the Athlon an 18% advantage over the Athlon on the KT133 platform, not to mention an even greater advantage on the two Intel platforms.

Floating Point Performance - SPEC CFP2000
base number (peak number)
Test
Athlon/AMD 760 (PC2100 DDR)
P3/i840 (PC800 Dual RDRAM)
Athlon/VIA KT133 (PC133)
P3/i815 (PC133)
179.art
292 (304)
328 (331)
208 (208)
271 (275)

As we noticed at the start of this review with our Linpack benchmarks, the Athlon on a KT133 offers worse memory performance than an equivalently clocked Pentium III on an i815 as the data size increases beyond the initial 384KB cache of the Athlon. Thus it isn't surprising that the Athlon on the KT133 comes out slower than the Pentium III on the i815 chipset here, but the fact that the Pentium III is 30% faster is pretty surprising.

We can't attribute this to a shortcoming of the Athlon as the same CPU with a faster FSB and DDR SDRAM on the AMD 760 produced a score greater than that of the Pentium III on the i815. And we also know that the FSB isn't the cause for the bulk of the performance improvement as the Athlon already has effectively a 200MHz FSB. So this benchmark obviously stresses memory bandwidth. With the Pentium III on the 840 offering at least a 12% advantage over the Athlon with DDR SDRAM, you can expect the data accesses to come in a very serialized fashion and thus taking full advantage of the i840's 3.2GB/s of memory bandwidth.

The only thing we're not taking into account here is price, in which case the i840 loses quite a bit of its appeal.

SPEC CFP2000 Performance SPEC CFP2000 Breakdown (continued)

Log in

Don't have an account? Sign up now