RTX 5080 with missing ROPs takes a 12% performance hit in synthetic benchmark
The RTX 5080 is no exception to manufacturing defects.
If you've been out of the loop, a handful of RTX 50-series (Blackwell)GPUs are reported to suffer from a manufacturing flaw that fuses off an ROP partition in the GPU. Nvidia quickly addressed these reports and claimed that only the RTX 5090, RTX 5090D, and RTX 5070 Ti are affected; however, a new Reddit user report indicates that the RTX 5080 is also implicated. Courtesy of Hardwareluxx, the user tested their unit in 3DMark Time Spy, where it suffered a 12% performance loss, enough to put it behind the RTX 4080 Super in some scenarios.
As the news spread, Nvidia followed up with a statement asserting that only 0.5% of units manufactured are affected, with an on-average 4% loss in graphical performance. The defect is consistent across all impacted GPUs, disabling only one ROP partition (eight ROPs). Such manufacturing flaws typically occur when manufacturers fuse off individual components from a full-fat die for binning. For example, GB202 carries 192 SMs; however, the RTX 5090 uses a partially enabled variant with 170 enabled SMs.
The RTX 5080 was considered a fully enabled GB203 chip; however, hardware sleuth MEGAsizeGPU claims that Nvidia disabled one set of media engines on the GPU. Fusing off any part of the chip carries a risk of disabling neighboring components, and this theory may clarify why the RTX 5080 is also seemingly affected. Still, this doesn't explain how or why Nvidia and AIBs let these chips pass QA testing.
In 3DMark Time Spy, a normal RTX 5080 with all 112 ROPs operational scored 32,273 points, while the nerfed unit with 104 ROPs scored 28,118 points, roughly 12% slower. We've gathered that 3DMark Time Spy is extremely sensitive to ROP counts, and the performance hit varies from scenario to scenario. The user's setup features a Core i7-12700K and 32GB of DDR4-3200 memory.
Nvidia has asked affected customers to contact their respective board partners for a replacement, though given the ongoing shortages, there's no guarantee of when they'll receive one given the ongoing shortages.
Blackwell has been quite disappointing with non-existent inventory, inflated prices, manufacturing defects, and the return of the cable meltdown scare. Let's hope RDNA 4 is priced well enough to provide competitive offerings, at least in the budget market.
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Hassam Nasir is a die-hard hardware enthusiast with years of experience as a tech editor and writer, focusing on detailed CPU comparisons and general hardware news. When he’s not working, you’ll find him bending tubes for his ever-evolving custom water-loop gaming rig or benchmarking the latest CPUs and GPUs just for fun.
-
helper800 Reply
I would speculate that this was intentional as a stop gap.Gaidax said:Nvidia clowning. Spend some of your billions on validation, this is embarrassing. -
jp7189 The 5000 series might be the worst launch ever. Is this a result of nvidia shifting attention and resources away from the consumer market?Reply -
Shiznizzle D model love coming your way.Reply
One world. One market. One card.
They already said they have no interest in supplying two sets of cards. A D model for china and a model for the rest of the world. This is the start of conditioning westerns consumer to have the same "power" as china gets.
Then they only have to make one card. Not two.
What is sad, is that for that kind of money you could only get a %5 or %10 max increase in performance if you are an unlucky card holder. And i think the %10 is optimistic considering they themselves are saying you are taking a %12 hit. -
Gaidax Reply
No, the real issue here is a massive bottleneck and slowdown at the foundries. By a mile at that.jp7189 said:The 5000 series might be the worst launch ever. Is this a result of nvidia shifting attention and resources away from the consumer market?
Much of the last few previous gen-to-gen boosts were afforded by default moving to a new node process, but this time around it was not viable to do, because it was just too expensive.
Staying at the same process as Series 40 really crippled Series 50 - architecture-wise Series 50 is actually pretty nice, but without the usual huge transistor boost due to denser process - it is dead in the water.
I hope Series 60 will be viable to make with 3nm process and I bet Series 70 will be like 50s all over again. -
alceryes Reply
It's over 14% difference - not 12%.Admin said:The RTX 5080 is also reported to be plagued by manufacturing flaws. Eight missing ROPs are reported to have reduced performance by up to 12%.
RTX 5080 with missing ROPs takes a 12% performance hit in synthetic benchmark : Read more