Core Ultra 9 285T ES Notes: Q4A7, a B860 Engineering Board, and the 35W Power Wall

Notes on the Core Ultra 9 285T ES sample Q4A7: platform, motherboard, power delivery, memory, performance, gaming results, and buying advice. The specs look tempting, but the 35W power wall, high DDR5 latency, scarce ES boards, and limited BIOS make it better suited to low-power tinkering than a gaming PC.

Some Core Ultra 200 engineering sample processors have recently appeared on second-hand markets at tempting prices. The catch is that ordinary B860 / Z890 motherboards usually do not support these ES CPUs directly. They need an engineering motherboard with an ES PCH to boot.

The main character here is Q4A7, which can be understood as an ES version of the Core Ultra 9 285T. Its specs look attractive: 8P + 16E, 24 cores in total, an NPU, and a new enough architecture. But its TDP is only 35W, and the test platform is a B860 custom board with a very simple BIOS, untunable memory, and trimmed-down power delivery. So the real result is not as simple as “cheap 24-core magic CPU.”

01 What This Platform Is

The CPU is Q4A7. Similar ES models include Q4A9, Q4A6, and others. It is close to the retail Core Ultra 9 285T, with the main differences in frequency and ES status. Functionally, the NPU and 24-core configuration are basically present.

The motherboard is a B860 custom board in an OEM-style design. It is not a retail board, and both expansion and BIOS options are restrained:

  • 2 memory slots.
  • 1 PCIe x16 graphics slot.
  • 2 M.2 slots.
  • 2 SATA ports.
  • 1 wireless card slot.
  • Rear USB 2.0, USB 3.0, USB 3.2 Gen2, Type-C, and 3.5mm audio ports.
  • Front USB and audio headers.

The key reason this board can use Q4A7 is that it has an ES PCH with a model similar to Q3NQ. Retail B860 / Z890 boards do not have this support, so even a cheap CPU is difficult to use directly.

02 Motherboard Power Delivery and Components

The B860 engineering board has rather basic power delivery. The CPU VRM area has no heatsink, and the pads show that the power design has been trimmed further. The PWM controller is a Richtek RT3635BJ, theoretically a three-channel controller that can manage multiple power rails.

In practice, the board does not provide iGPU power delivery and has no video output. The power design is roughly:

  • 4 phases for the core.
  • 1 phase for SA.
  • MOSFETs are from 大中: SM4373 and SM4377.
  • CPU power connector is only 4pin.
  • Motherboard power is 6pin, so a normal ATX PSU needs an adapter cable.
  • The board powers on automatically after receiving power.

That sounds thin, but for the 35W TDP Q4A7, the power pressure is not huge. The real issue is not whether it can run, but how little playroom and tuning space the board provides.

03 Real Shortcomings of This ES Platform

This type of Core Ultra 200 ES platform has two obvious shortcomings:

  1. It can only use DDR5 memory.
  2. Compatible motherboards are rare and not cheap.

These B860 engineering boards cost close to 600 RMB second-hand, which is not exactly bargain-bin pricing. Although Q4A7 itself is much cheaper than the retail 285T, the total platform cost is less dramatic once the motherboard and DDR5 memory are included.

Its advantages:

  • Much cheaper than the retail version.
  • Still has 24 cores.
  • Uses a newer architecture.
  • Temperature and efficiency look good at 35W.

Its disadvantages are just as clear:

  • Scarce motherboards.
  • Very minimal BIOS.
  • Memory cannot be overclocked, and timings cannot be adjusted.
  • ES platform uncertainty.
  • Gaming performance is clearly affected by high latency and low frequency.

So it is more like a low-power tinkering platform than a desktop platform that ordinary users can buy without thinking.

04 BIOS and Identification

The BIOS is typical of OEM machines: very few adjustable options. There is no memory overclocking support. Memory only runs at base frequency, and timings cannot be changed manually.

After installing the system and drivers, CPU-Z cannot display the full model name properly. It only identifies an Arrow Lake ES processor with a 35W TDP and 8P + 16E configuration:

  • 24 cores.
  • 40MB L2.
  • 36MB L3.
  • Maximum boost around 4.4GHz.
  • NPU frequency around 2.6GHz.
  • iGPU/related frequency information around 3.2GHz.

Windows can identify ES2 Q4A7, with information similar to Qray1500. This also shows that it is not a normal retail CPU, so compatibility, stability, and BIOS support should not be expected to match a retail chip.

05 CPU-Z and Cinebench: Split Results

CPU-Z was tested first:

  • Single-thread score around 728.
  • Multi-thread score close to 12000.
  • Compared with a stock i5-14600KF, single-core is about 19% lower.
  • Multi-core is about 17% higher.

Looking only at CPU-Z, this 35W 24-core ES looks fairly strong.

But Cinebench is less flattering:

  • Cinebench 2023 multi-core around 17440.
  • Cinebench 2023 single-core around 1937.
  • Single-core is slightly lower than 14600KF, but considering 4.4GHz versus 5.3GHz, it is still acceptable.
  • Multi-core is about 37% behind 14600KF.
  • Cinebench 2026 multi-thread score is around 4303, about 18% lower than 14600KF.

The key difference is that CPU-Z is a lighter load and not very sensitive to memory performance. Cinebench and 7-Zip are heavier and amplify both the 35W power wall and memory latency problem.

06 Memory Latency Is a Major Problem

The DDR5 memory on the test platform can only run in a state similar to 5600 C46, and AIDA64 shows memory latency as high as around 125ns. Compared with a 14600KF platform tuned to 4400 C18, latency is nearly 1.5x higher.

Although DDR5 still has some bandwidth advantages, high latency directly affects many desktop applications and games. Since this B860 engineering board cannot tune memory frequency or timings, users have little room to optimize through BIOS.

7-Zip also confirms the issue:

  • Q4A7: around 107.253 GIPS.
  • 14600KF: around 129.279 GIPS.
  • Q4A7 is about 21% behind.

This is the awkward part of the platform: many cores, low power, and a new architecture, but memory latency and power limits hold it back in many tasks.

07 Frequency Under the 35W Power Wall

In AIDA64 stress testing, after running FPU for 30 minutes:

  • P-core frequency is only around 1.6GHz - 1.7GHz.
  • E-core frequency is around 1.8GHz.
  • Power is firmly limited to 35W.
  • CPU temperature is only around 32℃.

After switching to the integer CPU test for another 30 minutes:

  • P-core frequency is close to 2.8GHz.
  • E-core frequency is around 2.6GHz.

This shows that cooling is not the problem. The power limit is simply very hard. Temperatures look great, but frequencies cannot climb. For low-power servers, NAS, and long-running light-to-medium workloads, this is an advantage. For burst performance and gaming frame rates, it is a clear weakness.

08 Gaming Performance: Not a Gaming CPU

The gaming part tested five games at 1080P, mainly comparing Q4A7 with i5-14600KF.

CS2:

  • Average FPS is only about 61% of 14600KF.
  • 1% Low is about 60%.
  • 0.1% Low is about 48%.

PUBG:

  • Average FPS is about 65% of 14600KF.
  • 1% Low is only about 32%.
  • 0.1% Low is about 49%.

Black Myth: Wukong:

  • Average FPS is about 79% of 14600KF.
  • 1% Low is about 64%.
  • 0.1% Low is about 43%.

Cyberpunk 2077:

  • Average FPS is about 72% of 14600KF.
  • Both 1% Low and 0.1% Low are about 67%.

Forza Horizon 5:

  • Average FPS is about 87% of 14600KF.
  • 1% Low is about 78%.
  • 0.1% Low is about 74%.

The conclusion is clear: the more a game depends on CPU frequency, latency, and scheduling, the worse Q4A7 performs. In GPU-heavy, well-optimized AAA games, the gap becomes smaller.

09 Why Gaming Performance Is Weak

Q4A7 performs poorly in games for three main reasons.

First, frequency. Once game load rises, CPU frequency drops under power pressure. Some games can stay around 3.8GHz, but others fall to 3.0GHz - 3.3GHz, far below the maximum boost of 4.4GHz.

Second, memory latency. DDR5 5600 C46 plus an untunable BIOS makes memory latency ugly, and many games are sensitive to latency.

Third, the Core Ultra 200 series itself has high inter-core latency issues. Low D2D and NGU frequencies also affect performance. Manual tuning usually requires a high-end Z890 platform, while this test uses a B860 engineering board with almost no tuning space.

So even if you switch to Q4A9 or Q4A6 with slightly higher frequency and power limits, gaming performance may not change dramatically. The root cause is not just one CPU’s frequency, but the limits of the whole platform.

10 How to Choose Against 7500F and 14600KF

If the goal is gaming, Q4A7 is not very worthwhile. In gaming performance alone, it is not only far behind 14600KF, but also worse than AMD’s 7500F.

The real cost also needs to be counted:

  • 7500F is not expensive.
  • Entry-level AM5 motherboards are easy to find.
  • Memory latency is easier to reduce.
  • Platform stability and BIOS tuning are better.

If you only see many cores and a low CPU price and want to build a gaming PC around Q4A7, you will probably be disappointed. This should not be treated as a gaming CPU.

11 Better Use Cases

Q4A7 is better suited to:

  • NAS.
  • Long-term low-power operation.
  • Multi-core workloads that do not require high frequency.
  • Users who can accept ES platform uncertainty.
  • People willing to tinker with rare boards, adapter cables, and BIOS limits.

It is not suitable for:

  • Gaming PCs.
  • Stable daily main machines.
  • Manual overclocking, memory tuning, and BIOS tweaking.
  • Production environments with high compatibility and stability requirements.
  • Buying only because “24 cores are cheap.”

There were also several cases where the system failed to boot for no obvious reason and required clearing CMOS to recover. This is not surprising on an ES platform, but it is a very real annoyance for normal users.

12 Buying Advice

If you know exactly what you want, such as a low-power NAS, long-running light-to-medium workloads, or background multi-core tasks, and you can accept scarce ES boards, a limited BIOS, occasional bugs, DDR5 cost, and platform uncertainty, then Q4A7 can be considered.

But if you want the cheapest possible gaming PC, or want to experience the full playability of Core Ultra 200, this ES platform is not recommended. If you really want to play with Ultra 200, a retail 265K + Z890 setup is clearer in performance, tuning, and stability.

Simple summary:

  • Low-power multi-core tinkering: worth a look.
  • NAS / light server: somewhat attractive.
  • Gaming: not recommended.
  • Ordinary main PC: not recommended.
  • Pure DIY tinkering: not that fun unless you can accept many limits.

Q4A7 does have tempting specs, but the key to this platform is not “cheap 24 cores.” It is the combination of 35W, ES status, B860 engineering board, high DDR5 latency, and a minimal BIOS. Understand those conditions first, then talk about value.

记录并分享
Built with Hugo
Theme Stack designed by Jimmy