{"id":201856,"date":"2024-08-07T15:00:50","date_gmt":"2024-08-07T13:00:50","guid":{"rendered":"https:\/\/www.hwcooling.net\/?p=201856\/"},"modified":"2024-09-12T00:42:03","modified_gmt":"2024-09-11T22:42:03","slug":"amd-ryzen-5-9600x-slightly-improved-hexa-core-once-again-review","status":"publish","type":"post","link":"https:\/\/www.hwcooling.net\/en\/amd-ryzen-5-9600x-slightly-improved-hexa-core-once-again-review\/","title":{"rendered":"AMD Ryzen 5 9600X: Slightly improved hexa-core once again"},"content":{"rendered":"

In the first wave of new AMD processors with Zen 5 architecture<\/a>, we took a closer look at a Ryzen 5. The results show many improvements (in speed, in efficiency, in cooling, …) led by unprecedented single-threaded performance. But in terms of the multi-threaded one, it’s still well behind the rival Intel Core i5-14600K<\/a> CPU, albeit with much better power consumption. Gaming performance? Sometimes strange.<\/p>\n

AMD Ryzen 5 9600X in detail<\/h2>\n

The new generation of AMD Ryzen 9000 (Granite Ridge) processors are half very innovative, half a bit “old familiar”. That’s because of how AMD builds powerful desktop CPUs. Following the example of the Ryzen 3000 and 5000, the Ryzen 9000s also adopt the original 6nm I\/O chiplets from the Ryzen 7000 CPUs, but the “CCD” chiplets with the CPU cores have been replaced with new ones. This means that in terms of internal connectivity, for example, the tested Ryzen 5 9600X also behaves similarly to the Ryzen 5 7600X<\/a>. The AMD AM5 platform is also supported again and from the outset, motherboards with 600 series chipsets (i.e. X670E, X670, B650E, B650, B650 and A620) are designed for Ryzen 9000.<\/p>\n

A BIOS with the newer AGESA microcode is required for support and to operate correctly, per expectations. Ryzen 9000s support early BIOSes with AGESA 1.1.7.0 microcode, but it is recommended to use a BIOS with AGESA 1.2.0.0a Patch A. This last microcode is already implemented on the vast majority of Asus, Gigabyte and MSI motherboards. This includes the Gigabyte B650E Aorus Pro X USB4<\/a> (BIOS F4c), on which we tested the Ryzen 5 9600X<\/a>, and we’ll be testing other models from the Granite Ridge family (Ryzen 7 9700X, Ryzen 9 9900X, and Ryzen 9 9950X) on it as well. This Gigabyte motherboard should be conceptually very similar to the X870 chipset motherboards, which we will also cover later in our tests after their release. But now on to the processor, which we’ll cover in this analysis.<\/p>\n

The Ryzen 5 9600X is the successor to the Ryzen 5 7600X<\/a>, to which it also corresponds “topologically”. All the processor cores are in a single chiplet. It has eight cores physically, but in the R5 9600X there are only six active cores. That’s a total of twelve threads, as this processor also uses SMT (Simultaneous Multithreading) technology. PCI Express 5.0 support (\u00d716 and a pair of \u00d74 interfaces for SSDs) and a dual-channel DDR5 memory controller are also retained. There is one change with this though, namely that AMD is now listing DDR5-5600 support, whereas in the Ryzen 7600 generation it was “only” DDR5-5200. Support for the highest RAM speeds (such as DDR5-8000) should also be improved. But that’s already within the scope of overclocking using AMD EXPO profiles.<\/p>\n

\"\"<\/a><\/p>\n

The integrated graphics core \u2013 Radeon with 128 shaders of RDNA architecture at 2200 MHz \u2013 is also adopted from the previous generation. DisplayPort UHBR10 outputs are supported (but the maximum output resolution is only 4K). The multimedia support hasn’t changed either. The role of this iGPU lies only in a normal “display” tool, don’t expect gaming performance from it.<\/p>\n

The baseline all-core boost clock speed in the specs is significantly lower for the Ryzen 5 9600X (3.9 GHz) than the Ryzen 5 7600X (4.7 GHz). This is related to the fact that the R5 9600X has a TDP of only 65 W (PPT 88 W) and the R5 7600X is up to 105 W (PPT 142 W), which throttles the boost to a substantially lesser extent. However, unless you, typically manually, limit the power limits of the processor, you’ll encounter different, higher all-core boost clock speeds. AMD doesn’t specify a maximum, that’s officially only set for single-core boost \u2013 5.4GHz (+100MHz vs. the Ryzen 5 7600X).<\/p>\n

The CPU core chiplet (CCD) is manufactured using the N4P node, a 4-nanometer TSMC technology that should be slightly better than the 5nm N5 node used for Zen 4 in the Ryzen 7000 processors. In this new Ryzen 9000 generation, there has been a significant increase in transistor density (to about 8.3 billion in Zen 5 from about 6.6 billion in Zen 4) on a comparable chip\/chiplet area (~71 mm\u00b2). This is probably mainly due to the architectural and physical design, as the N5 and N4P nodes are still members of the same generation and there are no major differences between them that would correspond to a true transition to a new process generation.<\/p>\n

What the core architecture changes entail was covered in a separate article<\/a>. Virtually the entire core has been redesigned, which mainly has a wider frontend with two clusters of decoders, more ALUs, and a fully 512-bit SIMD unit. This promises very high performance in AVX-512 instructions, which we’ll get to.<\/p>\n

\"\"<\/a><\/p>\n

More efficient cooling of the cores<\/h2>\n

AMD claims that the Ryzen 9000s will be easier to cool compared to the previous generation of processors (Ryzen 7000). We’ll throw a bit of theory at it as well because “there may be something to it”, as our own measurements show. While the chip area hasn’t changed much, not even the heat spreader or TIM\/solder (between the IHS and the CCD), but there should have been some optimizations at the level of the cores themselves. This is with respect to reaching lower temperatures of hotspots, i.e. the places where the highest temperature within the core is reached. This was done by more efficiently distributing blocks at the CPU core level with respect to less concentration in the most critical areas.<\/p>\n

\"\"<\/a><\/p>\n

At the same time, temperature sensors should be more appropriately placed and temperature reporting more accurate. For Zen 4 processors, there was more uncertainty in the measurement due to the greater distance of the sensors from where the highest temperatures are in the chip. Therefore, the processor management system did not know as well whether a critical value had already been reached at a given location. The modeling of the temperatures reported by the processor was therefore reportedly somewhat pessimistic. With Zen 5, the temperature sensors should be better located directly to the sources of the largest hotspots and the measured temperatures should be closer to reality. That is, according to AMD, this is how it should work. Every degree Celsius dip will be good with respect to the height of a single-core boost in a single-threaded workload, for example. This is because achieving a given SC boost clock speed is strongly dependent on the temperature of the loaded core (or cores that alternate on the task<\/a>).<\/p>\n

Please note:<\/strong> The article continues in the following chapters.<\/p>\n

<\/div>
<\/th><\/th><\/th><\/th><\/th><\/tr><\/thead>
Manufacturer <\/td><\/td>AMD <\/td>AMD <\/td>Intel <\/td><\/tr>
Line <\/td><\/td>Ryzen 5 <\/td>Ryzen 5 <\/td>Core i5 <\/td><\/tr>
SKU <\/td><\/td>9600X <\/td>7600X <\/td>14600K <\/td><\/tr>
Codename <\/td><\/td>Granite Ridge <\/td>Raphael <\/td>Raptor Lake Refresh <\/td><\/tr>
CPU microarchitecture <\/td><\/td>Zen 5 <\/td>Zen 4 <\/td>Golden Cove (P) + Gracemont (E) <\/td><\/tr>
Manufacturing node <\/td><\/td>4 nm + 6 nm <\/td>5 nm + 6 nm <\/td>7 nm (\u201eIntel 7 Ultra\u201c) <\/td><\/tr>
Socket <\/td><\/td>AM5 <\/td>AM5 <\/td>LGA 1700 <\/td><\/tr>
Launch date <\/td><\/td>08\/08\/2024 <\/td>09\/26\/2022 <\/td>10\/17\/ 2023 <\/td><\/tr>
Launch price <\/td><\/td>279 USD <\/td>299 USD <\/td>319 USD <\/td><\/tr>
Core count <\/td><\/td>6 <\/td>6 <\/td>6+8 <\/td><\/tr>
Thread count <\/td><\/td>12 <\/td>12 <\/td>20 <\/td><\/tr>
Base frequency <\/td><\/td>3.9 GHz <\/td>4.7 GHz <\/td>3.5 GHz (P)\/2.6 GHz (E) <\/td><\/tr>
Max. Boost (1 core) <\/td><\/td>5,4 GHz (5,45 GHz unofficially) <\/td>5.3 GHz (5.45 GHz unofficially) <\/td>5.3 GHz (P)\/4.0 GHz (E) <\/td><\/tr>
Max. boost (all-core) <\/td><\/td>N\/A <\/td>N\/A <\/td>5.3 GHz (P)\/4.0 GHz (E) <\/td><\/tr>
Typ boostu <\/td><\/td>PB 2.0 <\/td>PB 2.0 <\/td>TB 2.0 <\/td><\/tr>
L1i cache <\/td><\/td>32 kB\/core <\/td>32 kB\/core <\/td>32 kB\/core (P), 64 kB\/core (E) <\/td><\/tr>
L1d cache <\/td><\/td>48 kB\/core <\/td>32 kB\/core <\/td>48 kB\/core (P), 32 kB\/core (E) <\/td><\/tr>
L2 cache <\/td><\/td>1 MB\/core <\/td>1 MB\/core <\/td>2 MB\/core (P), 4\u00d7 4 MB\/4 cores (E) <\/td><\/tr>
L3 cache <\/td><\/td>1\u00d7 32 MB <\/td>1\u00d7 32 MB <\/td>1\u00d7 24 MB <\/td><\/tr>
TDP <\/td><\/td>65 W <\/td>105 W <\/td>125 W <\/td><\/tr>
Max. power draw during boost <\/td><\/td>88 W (PPT) <\/td>142 W (PPT) <\/td>181 W (PL2) <\/td><\/tr>
Overclocking support <\/td><\/td>Yes <\/td>Yes <\/td>Yes <\/td><\/tr>
Memory (RAM) support <\/td><\/td>DDR5-5600 <\/td>DDR5-5200 <\/td>DDR5-5600\/DDR4-3200 <\/td><\/tr>
Memory channel count <\/td><\/td>2\u00d7 64 bit <\/td>2\u00d7 64 bit <\/td>2\u00d7 64 bit <\/td><\/tr>
RAM bandwidth <\/td><\/td>89.6 GB\/s <\/td>83,2 GB\/s <\/td>89.6 GB\/s\/51.2 GB\/s <\/td><\/tr>
ECC RAM support <\/td><\/td>Yes (depends on motherboard support) <\/td>Yes (depends on motherboard support) <\/td>Yes (with vPro\/W680) <\/td><\/tr>
PCI Express support <\/td><\/td>5.0 <\/td>5.0 <\/td>5.0\/4.0 <\/td><\/tr>
PCI Express lanes <\/td><\/td>\u00d716 + \u00d74 + \u00d74 <\/td>\u00d716 + \u00d74 + \u00d74 <\/td>\u00d716 (5.0) + \u00d74 (4.0) <\/td><\/tr>
Chipset downlink <\/td><\/td>PCIe 4.0 \u00d74 <\/td>PCIe 4.0 \u00d74 <\/td>DMI 4.0 \u00d78 <\/td><\/tr>
Chipset downlink bandwidth <\/td><\/td>8.0 GB\/s duplex <\/td>8.0 GB\/s duplex <\/td>16,0 GB\/s duplex <\/td><\/tr>
BCLK <\/td><\/td>100 MHz <\/td>100 MHz <\/td>100 MHz <\/td><\/tr>
Die size <\/td><\/td>70.6 mm\u00b2 + 118 mm\u00b2 <\/td>71 mm\u00b2 + 118 mm\u00b2 <\/td>~257 mm\u00b2 <\/td><\/tr>
Transistor count <\/td><\/td>8.16 + 3.37 bn. <\/td>6.57 + 3.37 bn. <\/td>? bn. <\/td><\/tr>
TIM used under IHS <\/td><\/td>Solder <\/td>Solder <\/td>Solder <\/td><\/tr>
Boxed cooler in package <\/td><\/td>No <\/td>No <\/td>No <\/td><\/tr>
Instruction set extensions <\/td><\/td>SSE4.2, AVX2, FMA, SHA, VAES (256-bit), AVX-512, VNNI <\/td>SSE4.2, AVX2, FMA, SHA, VAES (256-bit), AVX-512, VNNI <\/td>SSE4.2, AVX2, FMA, SHA, VNNI (256-bit), GNA 3.0, VAES (256-bit), vPro <\/td><\/tr>
Virtualization <\/td><\/td>AMD-V, IOMMU, NPT <\/td>AMD-V, IOMMU, NPT <\/td>VT-x, VT-d, EPT <\/td><\/tr>
Integrated GPU <\/td><\/td>AMD Radeon <\/td>AMD Radeon <\/td>UHD 770 <\/td><\/tr>
GPU architecture <\/td><\/td>RDNA 2 <\/td>RDNA 2 <\/td>Xe LP (Gen. 12) <\/td><\/tr>
GPU: shader count <\/td><\/td>128 <\/td>128 <\/td>256 <\/td><\/tr>
GPU: TMU count <\/td><\/td>8 <\/td>8 <\/td>16 <\/td><\/tr>
GPU: ROP count <\/td><\/td>4 <\/td>4 <\/td>8 <\/td><\/tr>
GPU frequency <\/td><\/td>400\u20132200 MHz <\/td>400\u20132200 MHz <\/td>300\u20131550 MHz <\/td><\/tr>
Display outputs <\/td><\/td>DP 2.0, HDMI 2.1 <\/td>DP 2.0, HDMI 2.1 <\/td>DP 1.4a, HDMI 2.1 <\/td><\/tr>
Max. resolution <\/td><\/td>3840 \u00d7 2160 px (60 Hz)? * <\/td>3840 \u00d7 2160 px (60 Hz) <\/td>7680 \u00d7 4320 (60 Hz) <\/td><\/tr>
HW video encode <\/td><\/td>HEVC, VP9 <\/td>HEVC, VP9 <\/td>HEVC, VP9 <\/td><\/tr>
HW video decode <\/td><\/td>AV1, HEVC, VP9 <\/td>AV1, HEVC, VP9 <\/td>AV1, HEVC, VP9 <\/td><\/tr><\/tbody><\/table><\/div>Wordpress Table Plugin<\/a>\n
* We have no certainty on this parameter. AMD does not specify the maximum resolution and maximum refresh rate in publicly available materials. However, it is possible that it will be the same as for Ryzen 7000, i.e. 3840 \u00d7 2160 px (60 Hz).<\/strong><\/h6>\n