Deceptive packaging, bottom of the barrel or stroke of genius? – Patriot Viper Venom DDR5-6400 CL32 2x 16 GB kit with SK Hynix 8 Gbit “H-Die” ICs in test with teardown and overclocking

1 - Unboxing and specifications 2 - SPD information and special properties of the ICs 3 - Heatsink test with external K-type sensor 4 - Teardown and PCB analysis 5 - Compatibility and overclocking 6 - Benchmarks – Synthetics 7 - Benchmarks – Gaming 8 - Summary and conclusion

Benchmarks – Synthetics

The basis of our test system is an MSI MAG X870E Tomahawk WiFi with BIOS 2.A91 (AGESA 1.2.0.3g). Since the performance in RAM-intensive applications was improved by 1-3% with this AGESA, it is unfortunately not possible to directly compare previously measured results. Apart from that, the methodology and system setup have remained the same as in the Ryzen 9000 X3D reference guide, which I have linked here.

AMD Ryzen 9000 X3D RAM Reference Guide – 8200 2:1 vs. 6400 1:1, 2R vs. 1R, Optimized vs. EXPO vs. JEDEC in Synthetics and Gaming

EXPO/JEDEC (AXMP) configurations

For a better overview and to save you the clicking orgy, here are the RAM configurations tested today in compact form as Zentimings screenshots.

6000 1:1 36-38 EXPO M16H 1R (Crucial CP16G64C32U5B)

6000 1:1 30-36 "EXPO" H16A 1R (G.Skill F5-6800J3445G16GX2-TZ5RK)

5600 1:1 46-45 JEDEC N16A 1R (Nanya NT16GF64F88AX3F)

6000 1:1 36-44 EXPO S32M 1R (Kingston KF560C36BBE2K2-64)

6000 1:1 36-38 EXPO M16D 1R (Crucial CP16G60C36U5B)

6400 1:1 32-40 EXPO* M16H 1R (Crucial CP16G64C32U5B)

6400 1:1 32-40 AXMP* H8H 2R (Patriot PVV532G640C32K)

6400 2:1 32-40 AXMP H8H 2R (Patriot PVV532G640C32K)

6000 1:1 30-40 AXMP H8H 2R (Patriot PVV532G640C32K)

Even if the kit doesn’t actually have an EXPO profile, the profiles can still be loaded on the vast majority of motherboards using technologies such as A-XMP at MSI or DOCP at Asus. As often mentioned, the values stored here are effectively the same as for an EXPO profile, only in a different memory address on the SPD EEPROM. Accordingly, the Patriot kit is marked here with “AXMP”, but is nevertheless sorted with the EXPO/JEDEC kits in the diagram.

In the synthetic benchmarks, the Patriot kit with its XMP profiles is positioned where one would expect it to be based on the timings alone. However, the OCCT read test is interesting, where our IC specials with all 3 profiles are significantly faster than the competition. Presumably, the special internal layout of the memory chips is particularly suitable for this particular benchmark.

Otherwise, the fastest profile with DDR5-6400 and forced 1:1 mode is at the top and just ahead of the Crucial Pro Overclocking Kits with comparable timings. The faster tRAS timing of 84 instead of 102 should give the Patriot kit the decisive advantage here.

In latency-sensitive benchmarks such as PYPrime 2.0 or OCCT Latency, however, there is a slight deficit compared to other kits with slower timings. This is probably where we stumble across the unusual halving of the banks per bank group, because in order to process data in 4 banks, for example, 2 bank groups must now be activated, not just 1 as with “normal” DDR5 ICs.

Manually overclocked configurations

6400 1:1 28-37 * H16A 1R (G.Skill F5-6800J3445G16GX2-TZ5RK)

6400 1:1 32-29 * N16A 1R (Nanya NT16GF64F88AX3F)

6400 1:1 32-44 * S32M 1R (Kingston KF560C36BBE2K2-64)

8000 2:1 34-45 * H16A 1R (G.Skill F5-6800J3445G16GX2-TZ5RK)

6400 1:1 36-35 * M16D 1R (Crucial CP16G60C36U5B)

6400 1:1 30-39 * M16H 1R (Crucial CP16G64C32U5B)

6400 1:1 26-37 * H8H 2R (Patriot PVV532G640C32K)

In the manually optimized state, the Patriot kit can place itself around the full-fledged Hynix-based kits. It’s never quite enough for a top spot, but at least the kits with Micron or Samsung chips can be consistently outperformed.

What the figures do not show here, but I would like to mention, is the repeatability of the results. The typical deviation from one benchmark run to the next, for example in the Geekbench3 multi-core memory score, is around 50 points in my test setup. So if a config achieves, say, 13000 points, the worst of the 3 runs is usually not below 12950 points. However, the 8 Gbit H-Die ICs show significantly greater fluctuations here, of up to 300 points in this example. These inconsistencies in the results can also be observed in other benchmarks.

For better conciseness and comparability, the RAM configurations have been abbreviated. Examples of this are:

• 6400 1:1 32-40 AXMP* H8H 2R: DDR5-6400 in forced 1:1 MCLK to UCLK mode, with tCL 32, tRCDRD 40 and other timings of the modules’ XMP profile loaded with A-XMP. The kit uses SK Hynix 8 Gbit H-Die memory ICs on two ranks per channel.
• 6000 1:1 36-44 EXPO S32M 1R: DDR5-6000 in 1:1 MCLK to UCLK mode, with tCL 36, tRCDRD 44 and other timings of the EXPO profile of the modules. The kit uses Samsung 32 Gbit M-Die memory ICs on one rank per channel.
• 8000 2:1 34-45 * H16A 1R: DDR5-8000 in 2:1 MCLK to UCLK mode, with tCL 34, tRCDRD 45 and fully optimized subtimings (*). The kit uses SK Hynix 16 Gbit A-Die memory ICs on one rank per channel.
• 6400 1:1 30-39 * M16H 1R: DDR5-6400 in 1:1 MCLK to UCLK mode, with tCL 30, tRCDRD 39 and fully optimized subtimings (*). The kit uses Micron 16 Gbit Rev H memory ICs on one rank per channel.