1.5 HD Texture pack stitched images greater than 2048 resolution result in massive performance drop.

I can confirm this issue. Running fresh Windows 7 install, fully updated drivers, and fully updated Java. I loaded a fresh copy of MC and Sphax 128 together and when I run this texture pack I get less than 10fps. Typically I run MC at around 120fps. I can also confirm this is not an issue with my OS or any software or driver files as it ran the same before I reinstalled my OS. In fact, one reason I reinstalled everything was because I thought something in my system was corrupt. Running 1st generation i7 with Radeon HD5770.

These screenshots show that the issue is current in the 1.5.1 pre-release.

isn't this related to your computer not being able to render this quality? most other games give you a graphic option like: low-normal-fantastic. maybe your pc can't handle "fantastic" graphics(texturepacks)

It shouldn't in this game. Theoretically most computers should be able to handle up to 1024x texture packs for a game like this without any issue at all. For example, my PC can do texture settings for Battlefield 3 on High without any FPS drops, and its 6-7 years old now. That uses almost 1GB of video memory, while MC shouldn't go above 400mb of video memory, hell, I'd even be inclined to say 200mb or 100mb, but I don't know how everything in MC is stored.

Usually in games, texture settings have minimal impact on performance so long as you have a decent amount of video memory. The only reason I cannot go to ultra on BF3 for textures is that you need 1.5gb of video memory and I only have 1.

There is something Mojang could do for textures, but it wouldn't make the game run much higher as it only increases loading times. Though, what I am about to say is more related related to engines like UDK, Cryengine, and Quake 3 Engine (ID Tech 3) and may not fully apply here as there is hardly the same amount of textures to be loaded. That is converting the PNG images to TGA. I have found, as well as other developers, that PNG reduces load times because it has to decompress the image twice, vs once. This is because the PNG is packaged usually inside a compressed file like a JAR or ZIP. It has to decompress one before decompressing the PNG for rendering. With TGA it only has to do the zip/jar decompression and with a TGA being in a compressed package, it compresses itself very well, almost to the point of a PNG. In terms of loading efficiency: TGA>JPG>PNG. In terms of quality: TGA>PNG>JPG at least in my tests. The only problem with TGA is that it does not automatically create an alpha channel. Though that can be considered as a good thing, as people are less likely to steal content from other texture packs as it requires a little bit of effort to create that alpha channel.

Idk, enough rambling. 😛

uhm do you have evidence of mc only taking up 400 mb of video memory with 2k textures? seems most unlikely. note RAM is not video memory...

Its more or less logical, most of the texture data for MC is miniscule in comparison to other games. I've designed multiple scenes in the engines that I mentioned, with way more high res textures with normals, specular, etc maps as well on top of that and I didn't get more than 500mb on the vram on an entire 40,000 unit x 40,000 unit level. I haven't done any specific tests on MC, but this is based on previous game experience.

There also shouldn't be a lot of overlay in the buffers because well they're solid box objects and there isnt a lot of transparency, especially with the fact that MC uses voxels and not polygons. So that shouldn't be taking up a ton of video memory either. I don't know about the chunk format, and that could be partially saved in video memory as well as ram, but the data within them for a normal view distance still shouldn't cause the whole thing to surpass 400mb of video memory.

And yes, I do know the difference between video memory and RAM. I wouldn't even be able to run BF3 on 1gb of ram as that is what I said I only have, but I was talking about video memory instead.

In the end though, if it does take more than 400mb of the vram (we already know MC can be a ram hog, but vram is a different story) that Mojang has some optimizations to do because it really shouldn't even get that high.

Most games have optimizations that reduce render times/memory which is not possible in minecraft.. do you have any tool to check vram while running minecraft?

Why wouldn't it be possible? MC only runs on a single core, and mods like optifine have shown it can at the very least do chunk loading multi-core and improving overall performance and quality at the same time. Things like Spigot and Spout have shown that current chunk formats are inefficient and are often prone to memory leaks. In-game MC usually says its only using 300 or so MB out of 1gb using F3 yet checking task manager it is clearly using a lot more. So its largely things Mojang has to do because Java is capable of it.

I don't currently have a tool to check vram as I usually used in-game profiling.

This has returned for me in 1.5.1 on my GTX 280.

I also tested this on 3 other graphics cards including my new one. These cards are an 8800GTS 640mb, GTX 550 TI, GTX 660 2gb all of which have the same result.