A rare variant of the ASUS ROG Strix RTX 5090 graphics card has been discovered boasting 192 Render Output Units (ROPs), exceeding the expected specifications and rivaling the professional-grade RTX PRO 6000. This discovery highlights potential inconsistencies in NVIDIA’s chip binning process and raises questions about the true performance potential of the RTX 5090.
The tech world is abuzz with the news of an ASUS ROG Strix RTX 5090 graphics card featuring an unusual configuration: 192 ROPs. This overspec model, unearthed by TechPowerUp, significantly deviates from the expected specifications and brings its performance closer to that of the NVIDIA RTX PRO 6000, a card designed for professional workloads.
What are ROPs and Why Do They Matter?
ROPs are crucial components within a graphics processing unit (GPU) responsible for handling the final stages of rendering, including pixel output, anti-aliasing, and depth testing. A higher number of ROPs generally translates to improved performance, particularly at higher resolutions and with demanding visual settings.
The ASUS ROG Strix RTX 5090 Anomaly
The standard RTX 5090 is expected to feature 176 ROPs. However, this particular ASUS ROG Strix RTX 5090 liquid-cooled card packs a punch with 192 ROPs – an increase of 16 units, or two additional groups. According to TechPowerUp’s testing, this configuration results in an average performance increase of 8% compared to a reference RTX 5090 (simulated by downclocking the card to reference speeds). The advantage is even more pronounced when compared to RTX 5090 models with reduced ROP counts (168 ROPs), where the performance difference can reach as high as 5%.
Furthermore, the ASUS card features a core clock boost of 7.1% (2580MHz vs. 2407MHz) over the reference design. This combination of increased ROPs and higher clock speeds translates to an overall performance gain of approximately 12%, bringing it close to the anticipated performance level of an unreleased RTX 5090 Ti variant.
Potential Chip Binning Errors
The discovery of this overspec card has led to speculation about potential errors in NVIDIA’s chip binning process. The RTX PRO 6000 Blackwell, released last month, utilizes the GB202 chip with all 192 ROPs enabled, along with 188 Streaming Multiprocessors (SMs) – totaling 24,064 CUDA cores (compared to the full potential of 24,576 cores). The RTX 5090, in contrast, is expected to have 21,760 CUDA cores.
TechPowerUp suggests that NVIDIA may have encountered issues during the chip selection process for the RTX 5090. While most chips are configured with the expected 176 ROPs, some may have been mistakenly disabled (resulting in the 168 ROPs variant), while others, like the ASUS card, may have slipped through with the full complement of 192 ROPs.
Implications and Future Outlook
The existence of this overspec RTX 5090 raises several questions:
- Performance Consistency: How widespread are these variations in ROP counts, and what impact do they have on the overall consistency of RTX 5090 performance across different models and manufacturers?
- Chip Yields and Binning: What does this discovery reveal about the yields and binning process for NVIDIA’s GB202 chip?
- Consumer Awareness: Should consumers be concerned about the possibility of receiving an RTX 5090 with a lower-than-expected ROP count?
NVIDIA has previously acknowledged the existence of RTX 5090 and RTX 5070 Ti cards with reduced ROP counts, stating that they affect less than 0.5% of the total production. However, the discovery of an overspec card adds another layer of complexity to the situation.
It remains to be seen whether other manufacturers have also produced RTX 5090 cards with 192 ROPs. IT Home and other tech news outlets are closely monitoring the situation and will provide updates as more information becomes available.
References:
- IT Home News Article: https://www.ithome.com/0/758/241.htm
- (Hypothetical) TechPowerUp RTX 5090 Review (Based on information in IT Home article)
Disclaimer: This article is based on information available from the IT Home news article. Further research and testing are needed to confirm the accuracy and implications of these findings.
Views: 0