The official word from AMD is late this year, so most likely November.
-
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 3 (5000 Series), rumored 17% IPC gain.
- Thread starter AStaley
- Start date
- Status
The official word from AMD is late this year, so most likely November.
+unsubscribes until november+
Soldato
Thank you.
Soldato
Some tests of the new desktop APUs:
http://www.coolpc.com.tw/phpBB2/viewtopic.php?f=52&t=270805
No dGPU testing,but performance looks competitive with the Zen2 CPUs.
http://www.coolpc.com.tw/phpBB2/viewtopic.php?f=52&t=270805
No dGPU testing,but performance looks competitive with the Zen2 CPUs.
4750G vs 3700X, the only outliner is 7z compression. Decompression looks okay.
Guess having a quarter of L3 must impact something.
Renoir is 156mm² and about 10 billion transistors. Zen2 CCD is 74mm² estimated at 3.9 billion plus about 2 billion for the I/O die. So a very similar transistor count. 24MB of extra L3 cache versus an IGP?
https://en.wikichip.org/wiki/amd/microarchitectures/zen_2#Die
https://en.wikipedia.org/wiki/Zen_2#Design
Soldato
I expect the GPU is taking up a lot of transistors? I really want to see how performance with a dGPU is,as we can get some estimations on how much having the I/O die not integrated into the CPU itself,affects gaming performance.
https://www.techradar.com/amp/news/...with-exec-hinting-theyll-be-something-special
Rick Bergman, who is executive VP of computing & graphics at AMD, made the comment at the end of a blog post: “So, what’s next for AMD in the PC space? Well, I cannot share too much, but I can say our high-performance journey continues with our first ‘Zen 3’ Client processor on-track to launch later this year. I will wrap by saying you haven't seen the best of us yet…”
Throw down.
Rick Bergman, who is executive VP of computing & graphics at AMD, made the comment at the end of a blog post: “So, what’s next for AMD in the PC space? Well, I cannot share too much, but I can say our high-performance journey continues with our first ‘Zen 3’ Client processor on-track to launch later this year. I will wrap by saying you haven't seen the best of us yet…”
Throw down.
The top-end IGP looks like it's a bit faster than a RX 550, snore
I thought it was gonna be better than that, though I suppose that's pretty good with 3600 DDR4, when the RX 550 has GDDR5.The top-end IGP looks like it's a bit faster than a RX 550, snore
That is actually very impressive. Seriously impressive considering these are 35-65 watt parts.
I still can't quite understand how's it's taking so long for AMD to integrate dGPUs onto desktop chips to give decent gaming APU performance (not enthusiast level just 60fps 1080 on AAA titles). They have a GPU and CPU division and both are needed in a PC yet this Holy Grail still eludes them.
Soldato
It isn't a holy grail, but a very niche market place. I think you'll see more of what you might be talking about once they hit TSMC 5nm chips, and when DDR5 is around and memory bandwidth is better which a good GPU requires.
The vast majority of laptops/desktops or anything sold requires basic graphical acceleration, but having a range of CPU's with better/more capable GPU does help when comparing it to the closest competitor. Maybe when Xe is out there'll be some marketing pushing these built in GPU's making a push for the dGPU market at the ~$100 mark.
Soldato
Staggered resources. AMD has literally rebuilt itself from the ashes, so there are priorities. Clearly the CPUs needed the most attention, so get those up to speed first. It's not a case of it eludes AMD, it's a case of other priorities.
That being said, you'd think getting a passable GPU onto a Ryzen CPU for the boggo office PC OEM market would be a bigger priority. That's a true Intel stranglehold which needs breaking, unless AMD actually don't care about high volume, low spec disposable office PCs, dumb terminals and thin clients?
Perhaps it's just not a market they're interested in. They're clearly capable though as the APUs in the upcoming PlayStation and Xbox certainly pack a punch.
Don
Both have access to dedicated GDDR6 however.
Even DDR5 won't be enough - even the projected PC5-51200 is "only" 51.2GB/s. Most GDDR6 based GPUs are in the 448-512GB/s area.
The only way to get that sort of bandwidth onto an APU is through the use of HBM or similar on package - it's physically doable (e.g. see below image of size of HBM2 and HBM stacks on Vega and Fiji GPUs), but is still too costly for the target markets
Well it's a niche because performance is so lacklustre. To sell an APU for £250 or thereabouts that can handle 1080p would seem like a huge opportunity. Many kids/adolescents could really use something along those lines at that sort of price as could their parents. Maybe £300 at a push but the rationale for such a chip seems blindingly obvious and the potential sales meteoric in gaming cafes for example across the developing World. Just one simple entry level chip to do the work of two. Seems memory bandwidth constraints make it an issue but surely they can figure a way around this and other issues although the time taken suggests it's a more complex problem than it appears on the surface.
Caporegime
Yes, in APU's there comes a point where shader performance is strangled by memory bandwidth.
The get 60 FPS 1080P level performance from integrated graphics you would need some sort of hybrid on die memory, AMD will do that when that option becomes cheap to do.
The get 60 FPS 1080P level performance from integrated graphics you would need some sort of hybrid on die memory, AMD will do that when that option becomes cheap to do.
Soldato
a modern pc APU on a wide gddr6 bus would be something. you'd only need 4x 16Gb modules on a 256bit bus at a stock 1750mhz clock for about 448Gb bandwidth. That would make for an interesting APU. imagine a CPU that could take advantage of that bandwidth.
Soldato
Managed to order the 4650G (100-000000143) yesterday, no firm delivery date from the vendor as yet, was going to opt for the 4750G but I didn't want to pay the huge premium, as I am only buying it to see what the performance is like vs the normal desktop CPU's.
Managed to order the 4650G (100-000000143) yesterday, no firm delivery date from the vendor as yet, was going to opt for the 4750G but I didn't want to pay the huge premium, as I am only buying it to see what the performance is like vs the normal desktop CPU's.
They are just 3000 series chips with GFX
Soldato
Managed to order the 4650G (100-000000143) yesterday, no firm delivery date from the vendor as yet, was going to opt for the 4750G but I didn't want to pay the huge premium, as I am only buying it to see what the performance is like vs the normal desktop CPU's.
It will be nice to see some results!
Not exactly. Renoir uses a monolithic die,and not chiplets. So the IMC is on the same die,so there will be less memory-CCX latency. There is also less L3 cache,but as many suspect,the huge L3 caches in Zen2 CPUs might be down to the chiplet design.
One of the problems is die area,and the other is memory bandwidth. Since AMD didn't have competitive CPU performance,they couldn't really make a very large APU,because the CPU performance wouldn't warrant a higher price. So they had to balance the die size with potential selling price. Even then the GPUs took up a large percentage of the die,but they started to also hit memory bandwidth problems.
Also for AMD they have traditionally shared their server designs with their desktop CPU lines,as the desktop got the runts of the litter which were not good enough for servers. For example the Phenom II X6,Bulldozer and Piledriver CPU dies,were shared with server equivalents which had two of them integrated as an MCM. Zen,Zen+ and Zen3 dies are shared with their server chips. Their APUs,were targetted towards the value end of the market,as CPU performance wasn't as good as Intel.
Intel "desktop" CPUs are basically the laptop APUs,in a desktop package. So they sell the leakier chips which probably can't make it to laptop,and those which have defects in the IGP.Their HEDT line is based on server chips. So its realistically taken AMD until Renoir to have an APU which is competitive in both CPU performance and GPU performance. But even then they have been conservative on die size - Renoir is a tiny chip,which is the smallest of the laptop APUs they have made. So I expect they were looking at costs again as 7NM is probably still pricey,and again memory bandwidth is a problem for the GPU. Intel tried to get around this with large L4 caches.
Last edited:
- Status