Test software and workstation
The software package includes 4 large software suites from PTC Inc, Dassault Systèmes, Autodesk and Adobe and of course the usual standard software such as Blender. I use professional or standardized workloads and do not use synthetic benchmarks. The only exception here is SPECviewperf 2020, which I deliberately included because it makes two things possible: Firstly, you can see in direct comparison to the full versions where the graphics card manufacturers may have optimized the drivers in order to look particularly good. And secondly, anyone can benchmark this themselves in order to make a direct comparison.
Even if NVIDIA would certainly have liked to see it differently, today it’s only about AI and computers to a limited extent (but also). It was important to me that there are programs in the test that run on all architectures and therefore allow a direct comparison. We will also see that some tasks do not require an ultra-expensive high-end card. These findings are also important when it comes to making an investment decision. Practicality is the new magic word.
The workstation is a PC with an Intel Core i9-13900K on an MSI Z790 Godlike (I know, it’s a gaming board) and 64 GB DDR5 PC6000 CL30, whereby I deliberately only run the memory as DDR5 5200. The performance difference to DDR5 6000 is only marginal in this application area for most applications, especially since I had a case where the system froze shortly before the end after a longer Creo run with 5 iterations with DDR5 6000. Incidentally, this is also the reason why I only allow the Core i9-13900K a PL1 and PL2 of 125 watts. It’s not a gaming PC and, as always, stability comes first. Time is pure money and a potential loss of data is inexcusable. Everything is cooled with a simple 360 Silent-Loop 2 AiO from be quiet! and I’ve put the workstation in a Fractal Meshify XL, just like the VGA test station.
I deliberately opted for a professional solution for the monitor. The almost frameless BenQ PD322020U with its 31.5 inch diagonal and 3840 x 2160 pixels is large enough, has an excellent AHVA panel from AU Optronics and with 10 bit color depth (1.07 billion colors) almost completely covers the required color spaces. The ergonomics are almost perfect and you can turn it (including the pivot function) however you like – one setting always fits. Not quite new, but tried and tested.
My new Keyence VHX 7000 with EA-300 takes care of the material testing and measurement of the cards, enabling both exact measurements and fairly precise mass determinations of the chemical elements. The laser-induced breakdown spectroscopy (LIBS) I now use is a type of atomic emission spectroscopy in which a pulsed laser is directed at a sample in order to vaporize a small part of it and thus generate a plasma. The emitted radiation from this plasma is then analyzed to determine the elemental composition of the sample. LIBS has many advantages over other analytical techniques. Since only a tiny amount of the sample is needed for analysis, damage to the sample is minimal. This technique generally requires no special sample preparation. Even solids, liquids and gases can be analyzed directly.
LIBS can detect multiple elements simultaneously in a sample and can be used for a variety of samples, including biological, metallic, mineral and other materials. And you get true real-time analysis, which is a huge time saver. Since LIBS generally requires no consumables or hazardous reagents, it is a naturally relatively safe technique that also does not require a vacuum. As with any analytical technique, there are limitations and challenges with LIBS, but in many of my applications, especially where speed, versatility and minimally invasive sampling are an advantage, it offers significant benefits.
Test System |
|
CPU | Intel Core i9-13900K @ 125W (PL2 = PL1) |
RAM | 64 (2x 32) GB Corsair Dominator Platinum DDR5 PC6000 CL30 @ DDR5 5200 |
Mainboard | MSI MEG Z790 Godlike |
Cooler | Be Quiet! Silent Loop 2, 360 mm |
PSU | Be Quiet! Dark Power 1000 Watt |
SSD | 2x 2 GB Netac NI7000-t NVMe |
Case | Fractal Meshify XL |
Monitor | BenQ PD322020U |
OS | Windows 11 Professional (all updates) |
Drivers | NVIDIA Branch/Studio: R535 U8 (537.70) WHQL AMD Software Pro Edition: |
Benchmarks
For the sake of simplicity, I am now linking to the updated version of the collection of all my workstation graphics cards for 2023, to which I have of course also added the Radeon Pro W7700 in advance. The reason for not including the charts directly in this article is obvious: The link will lead to the updated benchmarks even after the currently planned addition of four more workstation cards:
Nevertheless, I would of course like to check the advertising claims separately. Let’s start with Solidworks 2022, where I first tested the full version in a standardized workload without FSAA. The Radeon Pro W7700 is around 15.5 percentage points slower here than the older Radeon Pro W6800. However, it is 27.2 percentage points faster than the Radeon Pro W6600 and also 5 percentage points faster than an RTX 5000 with Turing chip and almost 7 percentage points faster than an RTX A4000 with Ampere chip.
But what about SPECviewperf, the free suite with implemented older software versions and pre-built workloads? SPECviewperf is always an easy victim for special, application-related optimizations of the drivers by graphics card manufacturers for marketing purposes. The Radeon Pro W7700 is only around 5 (instead of 15.5) percentage points slower than the older Radeon Pro W6800. What a coincidence. It is suddenly 92 (instead of 27.2) percentage points faster than the Radeon Pro W6600 and also 43 (instead of only 5) percentage points faster than an RTX 5000 with Turing chip and it suddenly even outperforms the RTX 6000! And finally: it is almost 36 (instead of just 7) percentage points faster than an RTX A4000 with an Ampere chip.
This short sample clearly shows why such synthetic, free benchmarks should be enjoyed with the necessary caution, especially since the gap between NVIDIA and AMD shifts further in favor of the green cards when FSAA is activated. Yes, AMD doesn’t perform badly with the Radeon Pro W7000 series, but you really have to use full benchmarks.
17 Antworten
Kommentar
Lade neue Kommentare
Urgestein
Veteran
Urgestein
Urgestein
Urgestein
Urgestein
Urgestein
Veteran
Urgestein
Mitglied
Urgestein
1
Urgestein
Veteran
Urgestein
Alle Kommentare lesen unter igor´sLAB Community →