8400mhz DDR5? Damn, that is fast.

That will likely be later, initially we will get the slower one i suppose. Any word on latency? usually that goes up in cycle count.

ooh that on die ECC. That may be a game changer and really when we are talking 64+ GB of memory if you want stability ECC is needed.

Maybe now iGPUs won't be bottlenecked by memory bandwidth. Maybe.

Who's DDR4 is running @ 1.2v? Most DDR4 is running overclocked enthusiast profile running @ 1.350v. So maybe DDR5 will have the same but from 1.1v - 1.2v or something.

schmidtbag:

Maybe now iGPUs won't be bottlenecked by memory bandwidth. Maybe.

It will sure promote much more powerful iGPU's and 128GB Ram in a dual channel system is sure nice but we all know when DDR5 comes out it's gonna be very very expensive, which is why AMD is going to be behind Intel on that. Still getting (8400MT/s) 134GB/sec vs present (3200MT/s) 51GB/sec or best case Ice Lake (4266MT/sec 68GB/sec) is impressive from a dual channel system. Would sure be impressive if they were triple channel as it would give integrated graphics as much bandwidth as a RX5500XT. I want to know what crappy latencies there are all new RAM has latencies stupidly high.

Reddoguk:

Who's DDR4 is running @ 1.2v? Most DDR4 is running overclocked enthusiast profile running @ 1.350v. So maybe DDR5 will have the same but from 1.1v - 1.2v or something.

True but JDEC spec is 1.2v

schmidtbag:

Maybe now iGPUs won't be bottlenecked by memory bandwidth. Maybe.

It'll be better but it's still nowhere near GDDR ram speeds to my knowledge. DDR5 i believe officially "supports" up to 51.2GB/s speed GDDR6 i believe officially "supports" up to 768GB/s speed No amount of "going beyond specs" for DDR5 will make up that gap lol

DDR5 is going to be a game changer, especially for high core count CPUs.

This all looks very ambitious and i like it if it holds true, also this will be nice for APU's they're very capable once u throw in higher bandwidth memory.

schmidtbag:

Maybe now iGPUs won't be bottlenecked by memory bandwidth. Maybe.

Sadly, it would be a very odd combo to pair an CPU and insanely fast ram, specially if we are talking about the 7000mhz + ram, which i assume is what you're talking about. Least price when it first comes out, think ram has settled a bit, but i assume the top end ram will be rather expensive. Having a look at my site 16gb of 4600 ram is looking about 480 whilst 2400 ram is about 70-80. Would still make more sense to get a gpu with that 400 pounds. We have 4000mhz ram now, has anyone done tests with Zen or intels Igpu on that? Though it would be interesting to know if we would ever get a console kind of APU since we are reaching these high mhz on standard ram now. Edit seen anandtech did a test, from 2133 to 3466, most titles seem to hit a limit around 2800-3333mhz sadly they didnt go higher but assume it wouldnt keep going up much since it seems to scale less and less. One big issue, long with ram prices is often a good motherboard is needed to hit the high mhz, most cheap boards dont allow for it or cant to it. adding more expense. If this was at least a fight for console level APU's like in the xbox one x could see something nice, but at their current level its a total waste of money, and unless they change they can get 16000mhz ram in them and still be too limited

Ricepudding:

Sadly, it would be a very odd combo to pair an CPU and insanely fast ram, specially if we are talking about the 7000mhz + ram, which i assume is what you're talking about. Least price when it first comes out, think ram has settled a bit, but i assume the top end ram will be rather expensive. Having a look at my site 16gb of 4600 ram is looking about 480 whilst 2400 ram is about 70-80. Would still make more sense to get a gpu with that 400 pounds. We have 4000mhz ram now, has anyone done tests with Zen or intels Igpu on that? Though it would be interesting to know if we would ever get a console kind of APU since we are reaching these high mhz on standard ram now. Edit seen anandtech did a test, from 2133 to 3466, most titles seem to hit a limit around 2800-3333mhz sadly they didnt go higher but assume it wouldnt keep going up much since it seems to scale less and less. One big issue, long with ram prices is often a good motherboard is needed to hit the high mhz, most cheap boards dont allow for it or cant to it. adding more expense. If this was at least a fight for console level APU's like in the xbox one x could see something nice, but at their current level its a total waste of money, and unless they change they can get 16000mhz ram in them and still be too limited

XBox One and PS4 both use AMD's Jaguar architecture. It's the same architecture that was used for the AM1 Athlon and Sempron processors.... Also, XBox One and PS4 technically don't use an APU. They use an SoC....

sykozis:

XBox One and PS4 both use AMD's Jaguar architecture. It's the same architecture that was used for the AM1 Athlon and Sempron processors.... Also, XBox One and PS4 technically don't use an APU. They use an SoC....

Should i rephrase and say it would be nice if we could get an APU similar to a console SoC but in APU format, would that be better? and yes technically its the same architecture, but we aren't getting that level of performance at all which is partly a shame. And partly most likely too many limitations, unless some OEM wants to strap gddr5 to the motherboard but never seen them do that.

Yay ECC even on consumer? That is great!

I do wonder how much longer off chip memory has in the laptop and desktop segment. Surely the ultimate aim is a desktop APU with 32GB HBM3 stacked system memory. I know this sounds crazy but that would be your CPU (16 Zen 4 Cores on 5nm = $800), your GPU (120CU RDNA 3.0 on 5nm) $1000) and your replace your system RAM ($300) = $2100, would that APU cost that much more than that? With 5nm densities and efficiencies plus Zen 4 power efficiencies plus RDNA 3.0 being 50% more power efficient than RDNA 2.0 which is 50% more efficient than RDNA 1.0, I think this is all doable and saleable too, so you could have half those specs for half the price. Maybe 5nm for all this is optimistic and we need to wait for 3nm but I would say this is where we are heading.

GIMME..... GIMME....... GIMME........!!.!. Maybe now after DDR5 memory will it be a bottleneck any more. Then again storage is still holding us truly back. Unless you go with some raided m.2's then you're still not as fast as today's memory's capabilities. Can't imagine the day when the whole "system" is able to move data throughout at speeds of 90+ GB/s

I bet I'll be waiting a while to get something that would be a significant bandwidth increase over my DDR4-4000 Cas 17-17-17-37 kit I currently have with my i9-9900K. DDR4-4000 @ Cas 17 has a latency of 8.5ns with 64GB/sec bandwidth. I think I'll be waiting for DDR5-8000 @ Cas 33 which would have a latency of 8.25ns and 128GB/sec bandwidth. Or at least something close to that! Every ram upgrade I've done I've always tried to lower latency slightly. Or I'd have to get Cas 34 which would be the same 8.5ns latency. It is weird to see Cas latency that high but as clock frequencies go up the latency will continue to get higher and higher! And I'm sure the first DDR5-8000 kits will have be nowhere close to that low latency. I'm guessing they will probably start around Cas 40?

Shadowdane:

I bet I'll be waiting a while to get something that would be a significant bandwidth increase over my DDR4-4000 Cas 17-17-17-37 kit I currently have with my i9-9900K. DDR4-4000 @ Cas 17 has a latency of 8.5ns with 64GB/sec bandwidth. I think I'll be waiting for DDR5-8000 @ Cas 33 which would have a latency of 8.25ns and 128GB/sec bandwidth. Or at least something close to that! Every ram upgrade I've done I've always tried to lower latency slightly. Or I'd have to get Cas 34 which would be the same 8.5ns latency. It is weird to see Cas latency that high but as clock frequencies go up the latency will continue to get higher and higher! And I'm sure the first DDR5-8000 kits will have be nowhere close to that low latency. I'm guessing they will probably start around Cas 40?

Test memorylatency with Aida64 🙂 Under 40ns is good with 9900k. Under 35ns is Very good. 60ns is with stock slow xmp memory @ ~3000

Undying:

8400mhz DDR5? Damn, that is fast.

I don't think it is running at 8400MHz. You see the big thing about DDR5 is that it is doubling data units per transfer over DDR4. So, it can run at half the frequency for the same amount of bandwidth. So 8400Mbps DDR5, is possibly running at same frequency as 4200Mbps DDR4 would run at, but transfer twice as much. EDIT: updated to explain better

nizzen:

Test memorylatency with Aida64 🙂 Under 40ns is good with 9900k. Under 35ns is Very good. 60ns is with stock slow xmp memory @ ~3000

Yah those latency values i listed are first-word latency or basically latency to transfer 1-byte to memory. You'll typically see memory latency listed in first-word, fourth-word and eighth-word if your looking at technical documentation. Which is 1 byte, 4 bytes and 8 bytes of transferred to memory. https://en.wikipedia.org/wiki/CAS_latency AIDA64 latency test from what i understand transfers 64 bytes to memory for the latency test. So it's latency value for the benchmark is basically only relevant to AIDA64. Other memory benchmarks will give a completely different value depending on how much data is actually transferred to to test the latency.

Aura89:

It'll be better but it's still nowhere near GDDR ram speeds to my knowledge. DDR5 i believe officially "supports" up to 51.2GB/s speed GDDR6 i believe officially "supports" up to 768GB/s speed No amount of "going beyond specs" for DDR5 will make up that gap lol

51.2GBps is per channel my friend. So, a dual channel will hit over 100GBps. To your point certainly no where close to GDDR6, but it will most definitely allow APUs to double the core counts without being memory starved. I am certainly ok with APUs with double the performance for a laptop or for my kid's PC.