Contents

• ABSTRACT
• 1. INTRODUCTION
• 2. The SPEC BENCHMARKS
• 3. THE DL_POLY BENCHMARK
• 4. SUMMARY
• 5. References

ABSTRACT

This report compares the performance of a number of different computer systems using the DL_POLY software. The benchmark suite used includes a set of six MD calculations using the DL_POLY molecular simulation code. The comparison involves forty nine computers, including scientific workstations from IBM, Sun, Hewlett Packard, Digital and Silicon Graphics, and Pentium-based PCs.

1. INTRODUCTION

Workstations that have been benchmarked, include those from

• IBM, with the Power, Power2 and Power2 Super chip (P2SC), and the recently released Power3 IBM Risc system RS6000-based models. The power3 CPU is available in the dual processor IBM RS/6000 43P Model 260 (200 MHz), and in the quad-processor WinterhawkII CPU clocked at 375 MHz.
• Hewlett Packard with the PA-RISC based HP model 9000 series, including the PA8000 and PA8200 CPUs, the former in the C160 (160 MHz), the latter in the 200 MHz HP PA-9000/V2200 (the Exemplar-V class) and PA-9000/C200 and in the 236 MHz PA-9000/C240 and PA-9000/V2250. Also included are the PA8500 CPUs, in the PA-9000/785 Model 360 (367 MHz) and more recent HP PA-9000/C3000 (400 MHz), HP PA-9000/J5000 and PA-9000/N4000 (440 MHz) models.
• Compaq/DEC, with the AXP Alpha EV5-based 500, PW AU and 8400 series. Of more interest are the variety of systems housing the A21264 EV6 CPU. These include the DEC Alpha 8400/6-575, the entry-level 466 MHz Compaq AlphaServer DS10, the 500 MHz A21264-based machines (the single CPU XP1000 personal workstation, and the dual processor Compaq DS20 and quad processor ES40 Alphaservers), plus the Compaq GS140 with 525 MHz A21264. Note that the DS10 features a 2 MByte L2 cache, in contrast to the 4 MByte cache of the XP1000, DS20 and ES40. Also included are (i) the Samsung AlphaPC 264DP-500, which with 2 MByte L2 cache runs Linux, in contrast to the Tru64 OS on the other EV6-based machines, and (ii) the new EV6.7 processor clocked at 667 MHz. The latter CPU features in the linux-based UP2000 dual-processor CPU from API, and in both the Compaq PW XP1000 and an engineering prototype ES40 Alphaserver (with 8 MByte L2 cache). Note that data generated on the latter machine should be viewed as estimates only, since the final product may differ.
• Silicon Graphics, with the R10k (Power Challenge, Origin200, Origin2000, Octane and the ill-fated O2) and R5k CPUs (Indy and O2 R5k). The R10k CPUs include those rated at 175, 195, 225 and 250 MHz.
Finally, we focus on the most recent R12k-based machines, with the Origin 2000 (300 MHz R12k), and the 270 MHz Octane R12k and O2 R12k.
• SUN, with the Ultra-SPARC-2 processors (Ultra-2/200, Ultra-2/300, Ultra30/300 and Enterprise HPC4500, 336MHz). More recent additions include the UltraSPARC-IIi based HPC4500 with 400 MHz CPU and Ultra80 (with 450 MHz CPU). Both the latter processors included 4 MByte L2 caches.

We should stress from the outset that our access to much of the hardware evaluated herein has been at best short lived, and has often involved the temporary loan or donation of machines as part of one of the hardware evaluation exercises run at the Daresbury Laboratory. In many cases these machines were not optimally configured in terms of either memory, or high speed disk, and consideration of the results presented here should be viewed in that light.

Following an introductory evaluation of hardware based on the SPEC Benchmarks (section 2), we present in sections 3 and 4 results using the DL_POLY simulation program [1].

Note that the present results are taken from a more detailed report on computational chemistry benchmarks; the associated MS powerpoint presentation is also available.

2. The SPEC BENCHMARKS

One of the most useful indicator of CPU performance is provided by the SPEC (``Systems Performance Evaluation Corporation'') benchmarks. This benchmark suite contains non-tuned application-based code to measure processor speed for both integer (SPECint) and floating point (SPECfp) arithmetic. While earlier versions of the suite (e.g. SPECmark89) had certain well-advertised flaws, the more recent offerings, SPECfp95 and SPECint95 have become industry standards in measuring primarily the performance of a system's processor, memory architecture, operating system and compiler.

SPECfp95 is derived from the results of ten floating-point benchmarks compiled with aggressive optimization. It is the geometric mean of ten normalized ratios (one for each floating-point benchmark). SPECint95 is derived from the results of eight integer benchmarks compiled with aggressive optimization. It is the geometric mean of eight normalized ratios (one for each integer benchmark) Note that the level of optimization is not mandated. While highly aggressive optimization is permitted, results derived from benchmarks compiled with conservative optimization (as in SPECbase) can be submitted.

SPECfp95 and SPECint95 results for many of the CPUs discussed in this paper are given in Table 1.

An analysis of trends in SPECfp95 ratings up to mid 1998 suggested that the leading CPUs from each of the key workstation vendors exhibited comparable performance figures. As of July 1998, the P2SC/160 CPU in the IBM RS/6000-397 exhibited the most impressive SPECfp95 rating. With a value of 26.6, the P2SC was marginally faster than the HP PA-9000/C240, 1.2 times faster than the HP PA-9000/C200 and SUN Enterprise HPC4500/336, 1.3 times faster than the DEC Alpha 8400/EV5-625 and SUN Ultra-2/300, and 1.4 times faster than the R10k-based SGI Origin2000/195.

This picture changed quite drastically with the arrival of both the EV6/A21264 from Compaq/Digital and the PA8500 CPU from Hewlett Packard, and more recently the EV67/A21264A from Compaq and PA8600 from HP. The 667 MHz EV67 exhibits SPECfp95 ratings of 72.20 and 65.50 in the Compaq ES40 and XP1000 respectively, while the 500 MHz EV6/A21264 show ratings of 58.7 and 57.7 in the Compaq DS20 and Compaq ES40 respectively. The figure of 54.00 for the 440 MHz PA8500 CPU in the HP PA-9000/J500 is almost identical to the 667 MHz EV67 in the API UP2000/6-667 (noticeably slower than that in the Compaq XP1000). The figure of 52.40 in the EV6-based Compaq PW XP1000/500 is almost identical to that of the 400 MHz PA8500 CPU in the HP PA-9000/C3000 (52.40). While the Winterhawk-II CPU from IBM is of comparable performance (SPECfp95 of 50.90), the current range of leading CPUs from other vendors, notably SUN and Silicon Graphics, are markedly inferior e.g., 34.4 for the SGI 300 MHz R12k in the Origin 2000), and 27.90 for the SUN UltraSPARCIIi in the Ultra80/450. The 200 MHz RS6000 power3 CPU in the IBM/RS6000-43P Model 260 shows comparable performance, with a SPECfp rating of 30.10. The following points should be noted regarding the values specified in this Table;

1. SPEC ratings for a number of the systems from SGI have not been published, and are estimates based on related CPUs. A number of other vendors have only published SPECbase_fp95 values, and not SPECfp95 ratings.
2. Note that values quoted for the Cray T3E systems in Table 1 are estimates based on extrapolations from the corresponding Digital EV5 specifications. No SPEC figures have ever been submitted by Cray/SGI for the T3D or T3E series.

Using the Compaq PW XP1000/667 value of 65.5 to normalises the SPECfp ratings, we would expect the XP1000 and ES40/6-667 to be somewhat ahead of the APU UP2000/6-667 and other EV6-based machines (the Compaq Alpha DS20, XP1000/500, ES40/500, DS20/500, Alpha GS140, and Alpha 8400/6-575) and the PA8500-based systems (the HP PA-9000/J5000, PA-9000/C3000 and PA-9000/N4000). These eleven machines, together with the IBM RS/6000-SP/375 appear far superior to the remainder. Based on this performance metric, the PW XP1000/667 is seen to be 2.2 times faster than the power3-based 200 MHz IBM RS/6000-43P/260 and 1.9 times the 300 MHz R12k-based SGI Origin2000. All other CPUs are projected to be significantly less than half the speed.

The EV67-, PA8500- and EV6-based machines from Compaq/DEC and HP also seen to dominate the SPECint95 ratings, with typical values of 37, 32-34 and 24-27 respectively. The 667 MHz EV67 is rated at twice the SGI Origin2000/R12k and SUN Ultra80/450. The value of 24.40 for the IBM RS/6000-SP/375 is seen to be competitive with the EV6-based CPUs. We also note that the SPECint95 ratings suggest that Pentium III/550 is 1.63 times slower than the Compaq XP1000/667, while the SPECfp95 ratings point to the Pentium being a factor of 4.8 times slower. Corresponding values for the AMD Athlon K7/600 are reduced to 1.34 and 3.34 respectively.

When considering the present benchmarking results, there are several factors we wish to consider in assessing the usefulness of the SPEC ratings;

i. Do the SPECfp95 values provide a reliable metric for evaluating the capabilities of hardware in computational chemistry? If so, we would expect to find a close mapping of the ratios for the various chemistry benchmarks onto the SPECfp ratios;

ii. Does any particular CPU consistently ``underperform'' based on the SPECfp criteria? - this would manifest itself as the ratios from the chemistry benchmarks falling below the SPECfp ratios. In particular we shall look for indicators of the memory problems of the SGI O2-R10k [2] impacting on the benchmarks.

We will attempt to address these issues below. Finally, we note that A SPEC FAQ describing the SPEC benchmark suite and the SPEC consortium is periodically posted to comp.benchmarks, and can be found on the WWW at

http://www.specbench.org/spec/faq

An excellent summary of the SPEC benchmarks that is periodically updated is available via anonymous ftp from ftp.cs.toronto.edu in the file /pub/spectable More SPEC-related information is available at the SPEC WWW site,

http://www.specbench.org

and at the Performance Database Web site,

http://performance.netlib.org/performance/html/spec .html#specsite.

Finally, we note that the next generation of SPEC benchmarks, SPEC CPU2000,

http://www.specbench.org/osg/cpu200 0/results/cpu2000.html

has been announced and will replace SPEC95 during the year. At present there is only a limited set of CPU2000 results available and we have not included these in the present report.

3. THE DL_POLY BENCHMARK

The benchmark summarised below is designed to reflect the typical range of simulations undertaken by the molecular dynamicist. It includes 6 calculations carried out using the DL_POLY molecular dynamics code, and includes the following functionality;

• Benchmark 1: Simulation of a sodium-potassium disilicate glass (1080 atoms, 300 time steps);
• Benchmark 2: Simulation of metallic aluminium with Sutton-Chen potential at 300K (256 atoms, 8000 time steps);
• Benchmark 3: Simulation of valinomycin in 1223 water molecules (3837 atoms, 100 time steps);
• Benchmark 4: Dynamic Shell model water structure (768 atoms, 1024 sites, 1000 time steps);
• Benchmark 5: Dynamic Shell model MgCl₂ structure (768 atoms, 1280 sites, 1000 time steps);
• Benchmark 8: Simulation of a model membrane with 2 membrane chains, 202 solute molecules and 2746 solvent molecules (3148 atoms, 1000 time steps).

The data presented in Table 2 is collected under control of the UNIX command time where available, and includes CPU time (both user and system), total elapsed time and Efficiency, measured as CPU versus elapsed. The total user CPU timings of Table 2 refer to the summed user CPU timings over all 6 calculations of the benchmark. Note that in contrast to the QC benchmark, little I/O is performed by the DL_POLY calculations, so that efficiency should always be high assuming the benchmarks were conducted on a dedicated resource.

The total CPU timings of Table 2 suggest that the Digital/Compaq Alpha EV67 CPU is dominant, with the ES40/667 and XP1000/667 showing comparable run times (10.8 and 11.2 mins. respectively), both ca. 1.2 times faster than the same CPU in the API UP2000/667. The EV67 is seen to be ca. 1.2-1.3 faster than the EV6-based machines from Compaq, the AlphaServer GS140 (13.9 minutes), AlphaServer DS20 and DS40 (14.3 and 14.5 minutes respectively) and XP1000/500 (14.8 mins). The EV67-based Compaq XP1000/667 outperforms the EV5-based DEC Alpha 8400/5-625 (19.8 mins.) and Alpha PW/600AU (20.2 mins.) by a factor of 1.78. Of the 11 leading machines of Table 17, only two are not Alpha-based, with the Hewlet Packard (HP PA-9000/N4000) and Silicon Graphics (SGI Origin2000/R12k) outperformed by the XP1000/667 by factors of 1.42 and 1.67 respectively.

When considering the performance of the CPUs from SUN, IBM and Hewlett Packard, we would note the following:

• The treatment of dynamic memory within Version 2.11 of DL_POLY relies on usage of the f90 compiler. This in part explains the exceedingly poor performance of DL_POLY on the SUN Ultra80/450 and SUN HPC4500/400, for SUN's current f90 compiler continues to generates rather inefficient code, in some cases running only half the speed of that generated using f77. The timings of Table 17 suggest the Ultra80/450 is a factor of 3.9 times slower than the Compaq XP1000/667 compared to the SPECfp95 ratio of 2.3
• While Hewlett Packards f90 compiler has also exhibited performance problems historically, there is some evidence of improvement here. Thus the run time of 19.6 mins on the J5000 (using HP F90 1.0.100) is reduced to 15.9 mins. on the N4000 when using HP F90 v2.3, both machines housing the 440 MHz PA8500. The run-time ratio on the N4000 compared to the XP1000/667 (1.42 times slower) is still greater than would be expected from the SPECfp95 ratio of 1.27.
• All CPUs from IBM perform badly on this benchmark. The power3 IBM RS/6000-SP/375 and RS/6000-43P Model 260 are seen to be factors of 1.94 and 3.2 times slower than the Compaq XP1000/667, the power2-based P2SC SP2/120Thin a factor of 6.0 times slower, and the IBM RS/6000-59H a factor of 9.6 times slower. These factors are much higher than those suggested by the SPECfp95 ratings (1.3, 2.2, 3.9 and 6.2 respectively).
• The performance of the Cray T3E is also disappointing. Just why the T3E/1200 is outperformed by the DEC Alpha PW/600AU by a factor of 2.1, and the T3E/900 by the DEC Alpha PW/433AU by a factor of 1.8, remains unclear.
• In contrast to the workstations from IBM, SUN and HP, the PC-based CPUs exceed the SPECfp95-based predictions. Thus the AMD Athlon K7/600 and PentiumII/550 are found to be 2.53 and 3.06 times slower than the Compaq XP1000/667, compared to the corresponding SPECfp95 ratios of 3.03 and 4.34 respectively.

4. SUMMARY

As a summary of this work, we present in Table 3 the relative performance of 93 of the leading CPUs against the Compaq XP1000/667 in terms of the SPECfp95 and SPECint95 benchmarks, and those from the present DL_POLY evaluation and from the Matrix-97, Chemistry Kernels and GAMESS-UK benchmarks (detailed in computational chemistry benchmarks).

Based on the published SPECfp95 ratings, and normalising with respect to the Compaq XP1000/667 value of 65.5, we would expect (see section 1) the XP1000 and ES40/6-667 to be somewhat ahead of the API UP2000/6-667 and other EV6-based machines (the Compaq Alpha DS20, XP1000/500, ES40/500, DS20/500, Alpha GS140, and Alpha 8400/6-575) and the PA8500-based systems (the HP PA-9000/J5000, PA-9000/C3000 and PA-9000/N4000). These eleven machines, together with the IBM RS/6000-SP/375 appear far superior to the remainder. Based on this performance metric, the PW XP1000/667 is seen to be 2.2 times faster than the power3-based 200 MHz IBM RS/6000-43P/260 and 1.9 times the 300 MHz R12k-based SGI Origin2000. All other CPUs are projected to be significantly less than half the speed. Based on these relative SPECfp values given in the table, we expect a factor of 52.3 between the fastest and slowest processor, the SUN SPARC/10-41.

The EV67-, PA8500- and EV6-based machines from Compaq/DEC and HP are also seen to dominate the SPECint95 ratings, with typical values of 37, 32-34 and 24-27 respectively. The 667 MHz EV67 is rated at twice the SGI Origin2000/R12k and SUN Ultra80/450. The value of 24.40 for the IBM RS/6000-SP/375 is seen to be competitive with the EV6-based CPUs. We also note that the SPECint95 ratings suggest that Pentium III/550 is 1.63 times slower than the Compaq XP1000/667, while the SPECfp95 ratings point to the Pentium being a factor of 4.8 times slower. Corresponding values for the AMD Athlon K7/600 are reduced to 1.34 and 3.34 respectively.

When analysing the results, we wish to consider based on the present evaluation exercise, (i) do the SPECfp95 values provide a reliable metric for evaluating the capabilities of hardware in computational chemistry? If so, we would expect to find a close mapping of the ratios for the various chemistry benchmarks onto the SPECfp95 ratios, (ii) does any particular CPU consistently ``underperform'' based on the SPECfp criteria? - this would manifest itself as the ratios from the chemistry benchmarks falling below the SPECfp ratios, and (iii) do the ``simple'' Matrix and Chemistry Kernel benchmarks lead to the same conclusions as the GAMESS-UK and DL_POLY benchmarks?

In the interests of providing a single Performance Index (PI) covering all machines of Table 3, we have provided an average value of the Matrix-97, Chemistry Kernels and GAMESS-UK benchmarks. Note that that at this stage we have not included the DL_POLY benchmark results in computing the PI, since that these results are only available on a sub-set of the machines. The value of such an index is somewhat debatable, for not only does it omit the DL_POLY benchmark, it weights the chemistry kernels on an equal footing with the end-application codes, which is not ideal. Note that we have only provided PI estimates for those machines where data on all three benchmarks is available.

Summarising the major conclusions from the figures of Table 3, we would note the following;

• While the EV67-based Compaq AlphaServer ES40/6-667 provides the optimum CPU based on the benchmarks conducted in this report, its superiority is not as pronounced as might be expected from just a consideration of the SPECfp95 rankings.
• Following the Compaq AlphaServer ES40/6-667 (112%), the optimum CPUs based on the benchmarks and PI values of Table 3 would appear in order to be the IBM RS/6000 SP/WII-375 (103%) and Compaq PW XP1000/667 (100%), the HP PA-9000/J5000 (95%) and N4000 (93%), the API UP2000/6-667 (86%), the HP PA-9000/C3000 (81%) and the EV6-based Compaq Alpha DS20/6-500 (81%), Compaq Alpha ES40/6-500 (79%), and Compaq PW XP1000/500 (75%). Two other EV6-based machines follow, the Compaq Alpha GS140 (73%) and DEC Alpha 8400/6-575 (71%).
• There is a sizable performance gap between the 12 machines above and those next in the list the Compaq Alpha DS10 (63%), the IBM Power3 RS/6000-43P (62%) and the SGI Origin2000/R12k and HP PA-9000/785 C360, both with an FI of 61%. All other machines considered would appear to be less than half the speed of the Compaq AlphaServer ES40/6-667.
• The apparent strength of the IBM SP/WII-375 and RS/6000-43P must be considered in light of the DL_POLY molecular dynamics benchmarks); inclusion of these figures would have lowered the PIs significantly (from 103 to 90% for the IBM SP/WII and from 62% to 55% for the 43P).
• With few exceptions, all non-Compaq machines above machines exhibit PI's significantly greater than the corresponding SPECfp95 ratios. Particularly striking is the IBM SP/WII-375 (SPECfp95 Ratio of 78% vs. the PI value of 103%) and the PA8500 based HP PA-9000/J5000 and N4000 (SPECfp95 ratios of 82 and 78% vs. PI values of 95 and 93%)
• The poor performance of the SUN Ultra80/450, SUN HPC4500/400 and SUN HPC4500/336 on the DL_POLY benchmark, caused by the status of SUN's f90 compiler at the time of benchmarking (see also [4] for similar Linpack statistics). The UltraSPARC systems from SUN, whose PI values exceed the SPECfp95 ratings, would perform more in line with the SPECfp95 ratings if this benchmark had been included in the final assessment.
• All R12k- and R10k-based machines from Silicon Graphics appear to exceed the SPCfp95-based performance predictions in all benchmarks. Particularly noticeable is the SGI Octane/R12k-270 (55% vs 38%) and the Octane/R10k-250 (45% vs 31%).
• The cost-effective performance of the AMD Athlon K7/600 and PentiumIII/550; the K7/600 exhibits a PI value of 47%, with its performance in all benchmarks exceeding expectations based solely on SPECfp95 figures. The PI value for the PentiumIII/550 (36%) is also significantly higher than the SPECfp95 ratio (23%).
• In the chemistry kernel benchmark all machines, with the exception of the EV6-based machines from Compaq and the Ultra-based SUN hardware, show higher performance ratios relative to the Compaq PW XP1000/667 than expected based on their SPECfp95 rating. This effect is often quite marked; the IBM RS/6000-43P, SGI Origin2000/250, SGI Octane/R10k-250, DEC Alpha PW/600AU, IBM RS/6000-397 and P2SC/160 exhibit performance ratios ranging from 62% to 52%, to be compared with the corresponding the SPECfp95 ratios that range from 46% to 39%. Clearly the PW XP1000/667 is underperforming in these four benchmarks. The impact of the improved memory bandwidth of the Compaq DS20 and ES40 compared to the DEC Alpha 8400/6-575 is particularly noticeable in the JACOBI benchmark.

In terms of perfromance ordering of CPUs, we find that the PI values and all the chemistry benchmarks are broadly in line with the SPECfp predictions; notable exceptions in addition to those already mentioned are summarised below:

• The good performance of the EV5-based PWAU series from Digital, with both the DEC Alpha PW/433AU and 600AU often outperforming other CPUs with higher SPECfp95 ratings.
• The memory bandwidth problems associated with the SGI O2 R12k/270 and R10k/175 [6], accounting for the poor relative SPECfp95 values (18% and 12% respectively), are not shown by either the Matrix-89 or chemistry kernels (with the exception of Jacobi). They are more evident in the GAMESS-UK benchmarks, where these CPUs performs in far inferior fashion to the corresponding SGI Octanes.
• The SPECfp95 rating of the IBM RS/6000-397 (26.6) is significantly higher than that expected from just clock speed considerations. Those CPUs featuring the same P2SC processor, the IBM RS/6000-595 (135 MHz) and IBM SP2/120Thin (120 MHz) exhibit SPECfp95 ratings some 68% and 64% of the 160 MHz RS/6000-397; based solely on clock speed, one might expect higher ratios of 84% and 75%. This effect has been attributed to the enhanced memory bandwidth of the IBM RS/6000-397. There is ample evidence that the benefits of this bandwidth are not apparent in most of the present chemistry benchmarks. Thus the relative figures for the matrix, kernels and GAMESS-UK benchmarks on the IBM RS/6000-595 are 83%, 79% and 81%, those on the IBM SP2/120Thin node, 74%, 75% and 76% respectively. These values are far more in line with the clock-speed predictions above than the enhanced SPECfp95 for the IBM RS/6000-397.

5. References

1

DL_POLY is a parallel molecular dynamics simulation package developed at Daresbury Laboratory by W. Smith and T.R. Forester under the auspices of the Engineering and Physical Sciences Research Council (EPSRC) for the EPSRC's Collaborative Computational Project for the Computer Simulation of Condensed Phases (CCP5) and the Molecular Simulation Group (MSG) at Daresbury Laboratory. The package is the property of the Central Laboratory of the Research Councils.

2

In theory the O2-R10k should have outperformed the corresponding Indigo2; with better memory bandwidth, superior I/O and more tightly coupled integration it should have done well. However SGI made a design decision which has seriously impaired the performance of the O2 in some application areas. It took about 3 months for this "flaw" to be fully identified. Until December 1996, SGI were claiming that the O2 R10k would perform in the region of 10, 12 or even 15 SPECfp95. Indeed on some benchmarks it does indeed achieve performance that matches these figures. However, the O2 has a Unified Memory Architecture which uses main system memory as memory for the graphics display and operations. Despite the impressive bandwidth figures for the O2, it does seem that the O2 memory architecture severely impedes the performance of the R10k processor, particularly when compared with the Octane, Indigo2 and Origin systems. This is shown by the SPEC comparisons of Table 1; we suspect that the two main factors limiting performance in the memory subsystem are the main memory speed and the CRIME chip.

The CRIME chip, which acts as the memory interface between the memory and the three drains on it - the CPU (800 MByte/second), I/O engine (500 MByte/sec) and the monitor display (700 MByte/second) - is probably the main bottleneck. This chip was designed to work as a built in memory controller, but the design was biased toward the R5k; it can't work directly with the R10k because the R5k expects 32 byte cache refills while the R10k wants to have 64 or 128 byte refills. Therefore SGI supply a custom ASIC with the R10k daughter board. This interfaces the R10k's level 2 cache with the CRIME chip. Performance problems are caused by the ASIC having to break each 128 byte cache refill operation into 4, 32 byte refills. The net impact of this effect is that the O2 R10k will only work well with problems that fit into the L2 cache (1 MByte). Not surprisingly, the memory intensive SPECfp95 figures are badly affected, although the impact on less memory intensive applications is not so severe. It should be noted that this type of incident is very rare; chips often fail to deliver but not system architectures designed for existing chips.

3

see the following recently opened web page to obtain Pentium Pro Optimized BLAS and FFTs for Intel Linux: http://www.cs.utk.edu/ ghenry/distrib

4

More evidence of the poor performance of Sun Fortran f90 Compiler is shown in the following figures Linpack 100X100 all Fortran benchmark kindly provided by Hans-Hermann Frese of ZIB, Berlin. The results using different Sun Fortran compiler releases on a Sun Ultra 60 Model 2360 used one UltraSPARC CPU at 360 MHz. The performance of the latest NAGWare f95 is also compared:

Linpack 100X100 Performance as a function of FORTRAN compiler.

Compiler	Options	Linpack 100X100
Sun f77 4.2	-fast -O5	164 Mflop/ s
Sun f77 5.0	-fast -O5	166 Mflop/ s
Sun f90 1.2	-fast	61 Mflop/ s
Sun f90 2.0	-fast -ftrap=no%division	68 Mflop/ s
NAGWare f95 4.0	-Wc,-fast	166 Mflop/ s

These figures underline that the performance of the Sun Fortran 90 compiler is still poor compared with Sun's Fortran 77 compiler. Surprisingly, the performance of the NAGWare Fortran 95 compiler which produces intermediate C code is as good as Sun's native Fortran 77 compiler.