Reply 53040 of 55555, by Trashbytes
- Rank
- Oldbie
Ozzuneoj wrote on 2024-05-23, 19:47:Okay, I think there must have been a misunderstanding. I wasn't talking about the AGP bus bottlenecks in my reply because I was […]
Trashbytes wrote on 2024-05-23, 16:04:Ozzuneoj wrote on 2024-05-23, 15:46:I don't think it's accurate to make sweeping generalizations about bottlenecks. Performance is always dependent on the software and settings first, then the hardware. If you're trying to play a graphically intensive game at a high resolution or with anti-aliasing on an HD 3850 + QX6800, the CPU will not be the bottleneck. If you're trying to play Quake II at 1024x768, then sure, the CPU will probably be the limit. Or, if you're trying to run something that is notorious for making poor use of CPU resources (any Elder Scrolls game), then yes, the CPU will be the bottleneck. In general though, the HD3850 is not an overly powerful card in relation to a good Core2 Duo or Quad. It may be newer, but depending on the game it tends to swing about 15-20% faster or slower than an 8800GT, and usually the 8800GTX outpaces it by a pretty significant margin. Those were the high end cards contemporary to the QX6800.
I would love to see some benchmarks that cover this specific setup though! I have an HD3850 and a couple of HD4650 AGP cards, but I don't have a Socket 775 AGP board myself. It would also be interesting to see different CPUs dropped in to see scaling in different situations, as well as some PCI-E GPUs to compare them.
I'm not wrong here the AGP 8X bus will never be a bottleneck, its very unlikely that any CPU GPU combination using that bus will ever be able to fill the 2.1GB a second bandwidth available to it for it to become the bottle neck. Since I was asked a question about the AGP 8x bus being the bottleneck my answer was focused on that, the reality is that there isn't a CPU on that bus that could saturate it. Its one of the biggest reasons these cards really seem rather silly for AGP cards, any game that could take full advantage of them would require a better CPU to do so. Moving to PCIe and a E8600 for instance, but that's outside of the question here as they are AGP cards.
Benchmarks for both these cards that I have seen pretty much confirm this, feel free to go look them up yourself they are an interesting read.
Okay, I think there must have been a misunderstanding. I wasn't talking about the AGP bus bottlenecks in my reply because I was referring to this statement:
For these cards likely the CPU, even with the CPU overclocked these two GPUs would still be waiting on the CPU to feed them data quickly enough
... in saying this, it sounds like you're saying that even an overclocked QX6800 would be the bottleneck on these GPU (period). But you were actually just saying that the GPU and CPU would be free to "talk" as quickly as they wanted because the AGP 8x bus isn't a bottleneck. That is a perfectly reasonable assessment, I just didn't get that out of that particular statement when I read it, which is why I didn't comment on AGP bottlenecks at all.
I think we're on the same page now. 😁
Yes, people seem to forget that AGP 8X and PCIe 1.0 X16 are the same speed and bandwidth. AGP 8X was a really huge improvement over AGP 4x and AGPs demise was mainly due to other issues not related to bandwidth or speed.
