• 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.

*** AMD "Zen" thread (inc AM4/APU discussion) ***

subbytna said:
Raven will really shake up the market. Intel IGP will be trounced once again. Budget builds may not even need a GPU anymore.
Click to expand...

Agreed, AMD now have the CPU horsepower to match a decent IGPU. The Zen APU's could be very good for budget builders and HTPC's. I imagine not even needing a DGPU for the lower tier at all.
 
They will have competitive CPU performance but I don't see how the IGP still isn't hamstrung by limited bandwidth without HBM (even with Vega's updated compression), and I cant see them using HBM, as it stands to excessively increase the price.
3000MHz DDR4 is ~50GBs and less than half that of the rx460.
 
eddyr said:
They will have competitive CPU performance but I don't see how the IGP still isn't hamstrung by limited bandwidth without HBM (even with Vega's updated compression), and I cant see them using HBM, as it stands to excessively increase the price.
3000MHz DDR4 is ~50GBs and less than half that of the rx460.
Click to expand...
Exactly. Using DDR4, I don't think you could even match the bandwidth of an old AMD 7750 (gddr5) GPU.

The sums are easy enough, if someone wants to check
 
I'm eagerly awaiting the Zen based APUs, loved my previous 7850k setup.
Have considered purchasing a 9800 and AM4 ITX board just now then changing the CPU out when they are out but... Waiting is the smarter choice.
 
cheesyboy said:
Exactly. Using DDR4, I don't think you could even match the bandwidth of an old AMD 7750 (gddr5) GPU.

The sums are easy enough, if someone wants to check
Click to expand...

There was talk of using a very small amount of HBM1/2 as cache (128/256MB or similar) which could help, and if its small, it may keep the cost impact down. Something small like that may keep the pricing much more reasonable. I also seem to remember around October time, Hynix IIRC was going to start offering HBM2 as opposed to just Samsung, so as Raven Ridge isn't here, it's still possible. In fact, it could even be Raven Ridge is delayed to allow for increased HBM shipments.

This is all talk however, it'll be interesting to see how crippled Vega is by low memory bandwidth, and whether it can surmount those to at least be reasonable; and safer to assume it won't have HBM.
 
Alexrose1uk said:
There was talk of using a very small amount of HBM1/2 as cache (128/256MB or similar) which could help, and if its small, it may keep the cost impact down. Something small like that may keep the pricing much more reasonable. I also seem to remember around October time, Hynix IIRC was going to start offering HBM2 as opposed to just Samsung, so as Raven Ridge isn't here, it's still possible. In fact, it could even be Raven Ridge is delayed to allow for increased HBM shipments.

This is all talk however, it'll be interesting to see how crippled Vega is by low memory bandwidth, and whether it can surmount those to at least be reasonable; and safer to assume it won't have HBM.
Click to expand...
If they do an equivalent to the Intel Iris Pro/Plus line, then I would expect some sort of on-package memory.

Going back to Kaveri, they had GDDR5 support pencilled in to the memory controller spec in the Kernal Developers Guide. It never happened, but it was certainly thought about.

Limited by dual channel DDR4 at, let's say, 3000mhz offers memory bandwidth of 48GB/s, compared with an AMD 7750 GPU offering 72GB/s. It would be a nice little APU, but won't blow anyone's socks off GPU-side.
 
jigger said:
That's not a good comparison. The Xbox one uses an APU with DDR3 and offers 1.2Tflops.
Click to expand...
If memory serves, XBOne GPU has on-package ESRAM to give a serious boost to memory bandwidth: in excess of 200GB/s, compared to that 48GB/s I mentioned above using traditional (fast) DDR4 dimms.
 
cheesyboy said:
If memory serves, XBOne GPU has on-package ESRAM to give a serious boost to memory bandwidth: in excess of 200GB/s, compared to that 48GB/s I mentioned above using traditional (fast) DDR4 dimms.
Click to expand...

It's not getting used and never took off. MS dropped the ERAM and would have got more performance from using that space for graphics pipelines. Look at the PS4 it has a huge amount of bandwidth and is only slightly faster than the Xbox one.
 
jigger said:
It's not getting used and never took off. MS dropped the ERAM and would have got more performance from using that space for graphics pipelines. Look at the PS4 it has a huge amount of bandwidth and is only slightly faster than the Xbox one.
Click to expand...
They've dropped it for Scorpio, instead moving to GDDR5 giving 300+GB/s memory bandwidth. It's still used for the standard Xbone: it has to be as the rest of the system memory is dead slow DDR3
 
Devs never really adopted It. Look at the MX150 it has 48Gb/s of bandwidth and offers 1.1Tflops. It's memory is also dead slow and that's passing data back and forth over the system.
 
jigger said:
Devs never really adopted It. Look at the MX150 it has 48Gb/s of bandwidth and offers 1.1Tflops. It's memory is also dead slow and that's passing data back and forth over the system.
Click to expand...
Devs absolutely have been using ESRAM: the problem was just that its size (32mb) means they often cannot hit 1080p.
 
I think its clear that memory bandwidth isnt a direct measure of graphics performance and will have less of an impact to a SoC based system.
 
jigger said:
I think its clear that memory bandwidth isnt a direct measure of graphics performance and will have less of an impact to a SoC based system.
Click to expand...
Oh yes, but when we're talking about potential of Ryzen APUs, it's still worth comparing, say, the performance of the DDR3 and GDDR5 versions of a low power card like an AMD 7730: the DDR3 version has something like 30GB/s bandwidth, but is horribly throttled by it. If 3000mhz DDR4 offers only 48GB/s bandwidth, then the ceiling of performance isn't very high.

AMD 7730 review, comparing DDR3 and GDDR5 versions:
http://www.tomshardware.co.uk/radeon-hd-7730-cape-verde-review,review-32751.html
 
1-2Tflops maybe a little more depending on clock speeds I would say. If DDR4 is the bottleneck for an APU at those kinds of performance remains to be seen.

We Know 40Gb/s is enough for 1.1Tflop and that's without the advantage of having the GPU and CPU as one chip. We also know DDR3 isnt holding the XBox back.
 
What gives 1.1 Tflop at 40GB/s?

A 7730's 0.6 Tflop is bottlenecked by a 28.8GB/s bandwidth.

And Ddr3 is only not holding the xbox back thanks to the 200GB/s memory bandwidth of the esram!
 
ESRAM has seen very little use from developers. If 32mb of ERAM could magically boost performance by a factor of 3 then why would the idea get dropped?
The Xbox perfomance is 1.3Tflops with DDR3.
 
You must log in or register to reply here.
Back
Top Bottom