The open rendering performance test technique in 3ds max can be considered the introductory word for this review. That is why we will only explain the testbed configuration.
This review is divided into three parts because of a great number of tests and diagrams. The first part will demonstrate the rendering rate in Brazil r/s and with 3ds max 5 renderer (Scanline), the second part will cover the subtests for Final Render Stage-1 SP2 and draw a conclusion about the single-processor systems, and the third one will be devoted to multi-processor configurations and the effect from the Hyper-Threading technology.
Tests in Brazil r/s
Here are three pure groups: the first one is the leading Pentium 4 3.2 eXtreme Edition, the second one consists of two other Pentium 4 CPUs (their performance can be considered equal). AMD's solutions go last. Pentium 4 3.2E comes very close to Pentium 4 3.4 in spite of its lower clock speeds either because the further clock growth in Northwood makes no effect, or the Prescott core is "closer" to SplutterFish renderer in operation with this function, or it's caused by the cache size. The last option is also proved by the breakaway of Pentium 4 eXtreme Edition.
The same three groups, though some CPUs gradually moved from one group to another. Thus, Pentium 4 eXtreme Edition now goes along with Pentium 4 3.2E (Prescott), and the Northwood of the highest clock rate takes the leading position in AMD's group! Athlon XP 3200+ loses to everyone. The assumption about the effect from the L2 cache size is not topical anymore as Northwood has the smallest one among all Athlon 64 CPUs.
Pentium 4 3.2E becomes a leader though it can't boast either of its clock speed (Pentium 4 3.4 Northwood has a 200MHz higher one) or of its cache (Pentium 4 eXtreme Edition has the largest one - 2 MB L2). It must have a certain architectural advantage but the gap is too narrow to make any conclusions.
The picture is similar to Brazil_DOF, but this is just a coincidence as this effect is not used in the final scene. However it can be, the Pentium 4 platform remains a leader. But we can't single out a leader within this platform. The dated 32bit AMD platform is a loser. AMD64 CPUs look better, though there's hardly any difference between Athlon 64 FX-51 and Athlon 64 3400+.
This "fast" scene hasn't changed the layout, only Athlon XP performs even worse. However, there's a leader within the Intel NetBurst platform - it's Pentium 4 eXtreme Edition.
Tests in Scanline
This test demonstrates a well-made optimization. It doesn't make sense to discuss whether the optimization for AMD is poor or the guys at Discreet developed a perfect code for Pentium 4 (there's actually no difference). The results of Pentium 4 3.2E (Prescott) point to the optimization: as they worked very hard on it, the new architecture now doesn't match the code well adjusted for Northwood. Northwood and Gallatin (alias Xeon, alias Pentium 4 eXtreme Edition) differ only in the big L3 cache. That is why their scores must be similar if it's only the core architecture that makes an effect. And we do witness it.
The AMD64 platform improves its scores, Pentium 4 eXtreme Edition goes far ahead... but on the whole, the picture remains the same.
Northwood (Pentium 4 3.4) sweeps the floor again. The eXtreme Edition loses because of its lower clock speeds. Prescott (Pentium 4 3.2E) loses even more because of peculiarities of the architectural optimizations of the software/CPU tandem because the 200MHz gap can't result in such performance gap. The AMD64 platforms looks pretty pale though it runs faster than Athlon XP.
If the Scanline rendering speed is what we are interested in the main competitors are Pentium 4 3.4 Northwood and Pentium 4 3.2 eXtreme Edition. Pentium 4 3.2E Prescott takes the third place in the Scanline group. You can decide yourselves how to consider it - as one of the processors of the leading platform or the worst processor among Pentium 4 models. Athlon 64 (both versions) is not the best choice.
This subtest doesn't change the layout.
The CPUs do change their places depending on the scene type, that is why our
work wasn't useless. At the moment (note that the Final Render results
will be published in the next article) the Pentium 4 platform leads,
with the eXtreme Edition looking really good. Although this solution
seemed to throw dust in eyes, we are glad it scores such excellent
results. The AMD64 doesn't look impressive, but we haven't tested
it in Final Render Stage-1 yet. Also remember that we use only one
application, and the scores obtained must be mostly interested to
those who use it in their real work.
Stanislav Garmatyuk (firstname.lastname@example.org)
Write a comment below. No registration needed!