• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Low cost HBM on the way, will hit mass market soon Plus HBM3 teased up to 64GB

clock speed is still just a number kaap - the two things that matter remain latency and throughput and HBM does excellently in these. There have been many threads where people have talked through the reasons why what you think you're seeing isn't actually caused by HBM :(
 
Hbm3 is good news for amd. While nvidia can stick to ddr5 and keep reducing bandwidth requirements at the same time as increasing performance.
 
While it is always nice to hear about new tech, lets see how well the second generation of HBM graphics cards early next year do first.

Also thinking about it AMD supposedly going to be using next gen memory on Navi rather than HBM, so where does that leave HBM*.
 
The obvious big advantage of hbm is less board space, but on the flip side just like sapphire and others done, you can have the same size board and have a very effective heatsink due to the overlap. This could even be reflected in reference boards instead of just third party boards.

If you look at something like titan x, the actual vapour chamber on the whole is pretty small, its probably 5 inches across or so and is barely keeping the core in check. Imagine what they could do with a vapour chamber that was something akin in size to what was on the sapphire fury tri-x. A lot more surface area to work with which could make the squirrel cage reference fans far more effective if it could be set up that way.

It seems obvious that at least nvidia are intent of keeping to this reference fan design so maybe that's one way they could make it more effective if they went hbm.
 
benjii said:
I was mistaken on how HBM worked, my apologies.
Click to expand...

Well on the plus side you get to join the select group of people who have ever admitted they were wrong on the Internet. Which is a very small percentage of those who actually are. ;) :)

Gerard said:
The obvious big advantage of hbm is less board space, but on the flip side just like sapphire and others done, you can have the same size board and have a very effective heatsink due to the overlap. This could even be reflected in reference boards instead of just third party boards.
Click to expand...

Actually there's another huge advantage to HBM. It uses far less power and generates far less heat. Both of these are as much the limiting factors on GPU performance as the cost of the silicon. GDDR5 eats a lot of power (and necessarily spits it back out as heat). HBM gives a substantial increase in headroom to crank the GPU itself up further.
 
h4rm0ny said:
Actually there's another huge advantage to HBM. It uses far less power and generates far less heat. Both of these are as much the limiting factors on GPU performance as the cost of the silicon. GDDR5 eats a lot of power (and necessarily spits it back out as heat). HBM gives a substantial increase in headroom to crank the GPU itself up further.
Click to expand...

Heat is negated by the fact that the heat is more concentrated on the core with the vram and the core producing a more intense hot spot as opposed to the ram being spread out. Not really sure how much difference in terms of power it makes.
 
Gerard said:
Heat is negated by the fact that the heat is more concentrated on the core with the vram and the core producing a more intense hot spot as opposed to the ram being spread out. Not really sure how much difference in terms of power it makes.
Click to expand...

If I recall GDDR5 uses something along the lines of 5W/chip, I am unsure what GDDR5X uses. While I agree it wouldn't necessarily cause problems with cooling, This power saving would still be immense for any graphics card and mean the core could be pushed a little further potentially.
 
Gerard said:
Heat is negated by the fact that the heat is more concentrated on the core with the vram and the core producing a more intense hot spot as opposed to the ram being spread out. Not really sure how much difference in terms of power it makes.
Click to expand...

This simply isn't how physics work.

Why is a 1080 core harder to cool than a 980ti, because of surface area. While die size went down almost 50%, power reduced only around 30%.

It doesn't matter where the chips are, they have their own surface area and touch the heatsink, the heat is transfer and the heatsink heats up depending on the load applied to it, the heatsink eventually hits a steady state and cooling is determined by the temp difference there.

Also, around 60% of the power saving is achieved within the memory controller rather than the chips themselves. Of the power required to generated 512GB/s of bandwidth about 60% of the power used is the memory controller running at higher speeds, needing to step down the incoming signals and the power required to send signals off package over 50cm+ long traces is a magnitude higher than that to send far slower signals on package over <5cm long traces.

So the majority of the power saving from HBM comes within the GPUs memory controller anyway, chip location wouldn't matter because the power saving is within the gpu die itself.
 
Seanspeed said:
The acid lining to all this is that it implies that HBM2 is too expensive to use in standard consumer GPU's.

That would be a huge shame.
Click to expand...

The real point is no one is using HBM1 0r HBM2 on a gaming card, not even a Pascal Titan X.

Maybe this is to do with price or maybe it is to do with performance. Whichever way you want to look at it both AMD and NVidia have decided to use other solutions for their gaming cards.
 
Kaapstad said:
The real point is no one is using HBM1 0r HBM2 on a gaming card, not even a Pascal Titan X.

Maybe this is to do with price or maybe it is to do with performance. Whichever way you want to look at it both AMD and NVidia have decided to use other solutions for their gaming cards.
Click to expand...

I understood that AMD are using HBM2 for Vega. That's what it says on their slides. Have they announced otherwise?
 
Do we need a dedicated thread for all things HBM related as it could get quite interesting if the OPs is updated regularly.

The OP could include news articles, cards using it and even people writing articles about it.

For example DM could writing a lengthy article about the subject that could go in the OP and if any one else wanted to do the same I would include those too.
 
bru said:
While it is always nice to hear about new tech, lets see how well the second generation of HBM graphics cards early next year do first.

Also thinking about it AMD supposedly going to be using next gen memory on Navi rather than HBM, so where does that leave HBM*.
Click to expand...

Maybe HBM3 is the next gen memory, I can't see how they'd of spent all this time and effort getting HBM up and running as a replacement for G5 just to suddenly have a replacement a couple of years later. Surely we'll have a good 5 or more years of HBM first. Unless there was a problem found late on in it's development meaning they need to move on earlier than planned. Something like it's being on the same interposer means it hold back a chips clocking ability maybe?
 
You must log in or register to reply here.
Back
Top Bottom