Rocket Lake Review: A waste of sand...

[Calin Banc]

But isnt this because developers need to make their games in mind with what specs most gamers already have? Developers had to stick to 4c/8t max as that was a norm, similarly to how little Vram graphics cards used to have - for a long time Nvidia only put up to 1.5 Gb on their highest end cards.

If a developer wanted to make a game that could utilize 8 cores and 4 Gb Vram, then no one would have been able to play it so it wouldn't have sold. Hardware isn't meant to be made based on what current software demands are, the point is that with better hardware becoming the norm, software developers can then utilize those features.

Almost all of them were limited by the consoles and their own bare minimum mentality. Even hardcore PC developers like Bohemia (ArmA series), didn't care about their engine that much. Why bother when you could get away with less?

You could put it that way I guess . Vram is a little more difficult. As Gpu's have got faster Vram had increased accordingly. There isn't much point in increasing vram unless you have the gpu processing power to actually make use of it. In all my years of pc gaming (my first 'gaming' pc was a 486sx25) I've never had a problem with a graphics cards vram (or lack of it).

Because developers never bother to offer higher quality textures for their assets.

 

[chrcoluk]

You get the feeling Intel are charging Motherboard vendors a small fortune for the chip sets to off set the decreasing price of their CPU's.

I am confused, for both AMD and Intel their premium chipset motherboards are extortionate pricing, You comparing AMD's mid range chipset to intel's top chipset.

Now intel have enabled memory o/c on their mid range chipset and the fact their chips arent worth o/c anymore, their midrange chipset is the way to go now, same as with AMD buy the mid range chipset.

PCIE4 seems to have really screwed the motherboard market, a tech that offers little benefit in return for the price bumps. I dread to think what happens when PCIE5 is pushed out.

 

[KompuKare]

PCIE4 seems to have really screwed the motherboard market, a tech that offers little benefit in return for the price bumps. I dread to think what happens when PCIE5 is pushed out.

What we don't know is how much of the price increases are due to the actual cost of better traces etc., and how much is the motherboard manufacturers just trying it on.
When PCIe 4.0 was brand new, the design cost of getting tracks and traces working might have been high but now that they have experience it shouldn't be a large cost anymore.
The same will apply when PCIe 5.0 comes around.
Power consumption is another worry. The share of power going to interconnect on server chips is now crazy, on desktop it isn't as important.

 

[Joxeon]

What we don't know is how much of the price increases are due to the actual cost of better traces etc., and how much is the motherboard manufacturers just trying it on.
When PCIe 4.0 was brand new, the design cost of getting tracks and traces working might have been high but now that they have experience it shouldn't be a large cost anymore.
The same will apply when PCIe 5.0 comes around.
Power consumption is another worry. The share of power going to interconnect on server chips is now crazy, on desktop it isn't as important.

Don't forget AIBs enabled PCIe 4.0 on x470/B450 before AMD blocked it and forced people to buy X570 if they wanted 4.0

 

[KompuKare]

Don't forget AIBs enabled PCIe 4.0 on x470/B450 before AMD blocked it and forced people to buy X570 if they wanted 4.0

Yes, which implies that short tracks aren't that hard to do for PCIe.
So one NVMe slot near the CPU and the x16 GPU slot should be relatively easy to do with existing designs. Longer tracks are harder, but for ITX and mATX that's enough.
If the faster signalling is responsible for using extra power, having to repeat it to boost/amplify the signal for longer tracks probably consumes extra power.
Haven't seen any mono group review for ages; always used to keep an eye out on idle power on those.
Don't X570 fans only come on with heavy traffic like copying with multiple NVMe drives?
Still haven't read any reviews about LGA 1200 and PCIe 4.0 power consumption - would like to know if x570 Vs z590 or b550 Vs b560 shows anything of interest.

 

[Game]

I am confused, for both AMD and Intel their premium chipset motherboards are extortionate pricing, You comparing AMD's mid range chipset to intel's top chipset.

Now intel have enabled memory o/c on their mid range chipset and the fact their chips arent worth o/c anymore, their midrange chipset is the way to go now, same as with AMD buy the mid range chipset.

PCIE4 seems to have really screwed the motherboard market, a tech that offers little benefit in return for the price bumps. I dread to think what happens when PCIE5 is pushed out.

does anything other than the odd m2 drive even use pci4 yet? not certain

 

[LePhuronn]

does anything other than the odd m2 drive even use pci4 yet? not certain

RDNA, RDNA2 and Ampere are PCIe 4

 

[Game]

but do they gain anything from it

rdna 1+2 dont seem to gain anything from it

 

[neo-omega]

They don't gain anything right now but if there's direct storage, such as in the PS5, then it will be a benefit to have very fast storage.

 

[chrcoluk]

pcie3 works on directstorage, so consider a top end gpu uses maybe 4-8 lanes of pcie3 at worst, then another 4 lanes for the nvme i/o, there is enough there.

If you have a gen 4 nvme drive running on a gen4 slot, then you probably also have the gpu in there as well, and I expect on gen4 a modern gpu only needs 4 lanes.

I have yet to see my 3080 uses more than 16% pcie bandwidth, and im running it on pcie3.

 

[Journey]

pcie3 works on directstorage, so consider a top end gpu uses maybe 4-8 lanes of pcie3 at worst, then another 4 lanes for the nvme i/o, there is enough there.

If you have a gen 4 nvme drive running on a gen4 slot, then you probably also have the gpu in there as well, and I expect on gen4 a modern gpu only needs 4 lanes.

I have yet to see my 3080 uses more than 16% pcie bandwidth, and im running it on pcie3.

As far as MS have stated, DirectStorage is still using the RAM as the middle man (so to speak), so NVMe to RAM, then RAM to GPU GDDR6(x) and decompression done there. There are no dedicated lanes required, or lost from the GPU. Be interesting to see what effect the extra bandwidth DDR5 offers in helping this along once it is developed into some games/programs.

 

[Mesai]

As far as MS have stated, DirectStorage is still using the RAM as the middle man (so to speak), so NVMe to RAM, then RAM to GPU GDDR6(x) and decompression done there. There are no dedicated lanes required, or lost from the GPU. Be interesting to see what effect the extra bandwidth DDR5 offers in helping this along once it is developed into some games/programs.

RTX IO skips the RAM but maybe there is some sort of buffering mechanism with the RAM - main idea is to remove the CPU from the equation (where it makes sense). Guess it all depends on whether storage to VRAM is faster than RAM to VRAM (if they need to do caching).

 

[Journey]

RTX IO skips the RAM but maybe there is some sort of buffering mechanism with the RAM - main idea is to remove the CPU from the equation (where it makes sense). Guess it all depends on whether storage to VRAM is faster than RAM to VRAM (if they need to do caching).

Difficult to know if that is true as Nvidia are being ultra vague with the true technical specs about what RTX I/O add to its use of the Direct Storage API. Either way the idea is to take compressed data, and shift it across the PCI-E bus, straight into the GPU+GPU RAM for decompression. The point being that none of this requires more PCI-E lanes, which people seem to be getting very confused with.

 

[Mesai]

Difficult to know if that is true as Nvidia are being ultra vague with the true technical specs about what RTX I/O add to its use of the Direct Storage API. Either way the idea is to take compressed data, and shift it across the PCI-E bus, straight into the GPU+GPU RAM for decompression. The point being that none of this requires more PCI-E lanes, which people seem to be getting very confused with.

Correct, it's already there. Same deal with REBAR and the CPU having direct access to the GPU, as long as you're using the right layout, you're fine.

 

[humbug]

I am confused, for both AMD and Intel their premium chipset motherboards are extortionate pricing, You comparing AMD's mid range chipset to intel's top chipset.

Now intel have enabled memory o/c on their mid range chipset and the fact their chips arent worth o/c anymore, their midrange chipset is the way to go now, same as with AMD buy the mid range chipset.

PCIE4 seems to have really screwed the motherboard market, a tech that offers little benefit in return for the price bumps. I dread to think what happens when PCIE5 is pushed out.

They actually aren't, this is a much better X570 board from the same vendor, again £160 https://www.overclockers.co.uk/asro...4-x570-chipset-atx-motherboard-mb-16t-ak.html

These days you get nothing but crap from Intel unless you pay

 

[chrcoluk]

They actually aren't, this is a much better X570 board from the same vendor, again £160 https://www.overclockers.co.uk/asro...4-x570-chipset-atx-motherboard-mb-16t-ak.html

These days you get nothing but crap from Intel unless you pay

looks low end to me like the intel board?

I remember when looking for decent top chipset board for amd when I was considering 3700X I couldnt find one for under £300.

 

[humbug]

looks low end to me like the intel board?

I remember when looking for decent top chipset board for amd when I was considering 3700X I couldnt find one for under £300.

Compared to this for the same money?

 

[Grim5]

RTX IO skips the RAM but maybe there is some sort of buffering mechanism with the RAM - main idea is to remove the CPU from the equation (where it makes sense). Guess it all depends on whether storage to VRAM is faster than RAM to VRAM (if they need to do caching).

RTX IO takes data from the storage (SSD) and transfers it to VRAM, on it's way to VRAM, any compressed files are decompressed by RTX IO.

RTX IO does not replace system RAM. RTX IO's sole job is to replace the CPU by doing decompresion on the gpu and not the cpu

Using RTX IO, a RTX3080 GPU has the same decompression performance as a AMD Threadripper 3970x

 

[LordBeggar]

I've been an intel fan for a very long time but now I've switched to my first AMD processor which is 5950x. Now I know why people are talking about intels performance