News AMD's AM5 Platform Launches With Only DDR5 Support for Ryzen 7000, Dual-Chipset Design

[KananX]

Yeah probably just quoting what so any say about Intel all these years when they were on top from Core 2 Duo in July 2006 all the way until July 2019 with Zen 2 release.

DO you think Intel was greedy forcing changes to new sockets and motherboards, or were their technical reasons for it despite them being way ahead of AMD. Like there are technical reasons for AMD requiring DDR5 for Zen 4??

Mixed, sometimes you have to change the socket sometimes not. If it was Skylake with a different name, like Kaby Lake etc, the change was less likely needed, but of course I would need the technical details to make a 100% statement. Fact is, Intel had a cadence of using the same socket for only 2 gens, AMD did the opposite, they changed sockets when it was needed so, so nobody trusts intel anymore when they say it was a technical reason.

Fact is, AM5 will be a long going platform so it’s not needed to have a outdated memory standard built in, it’s so not needed, it’s planned for the long term, unlike intel which was just planned for 2 gens again, “12th” and “13th”.
And I wanna add one more thing: AMD isn’t in the position of corruption to do the “Intel” it’s not a company with 80% or more market share. Alone for this reason, you alleging AMD to be greedy, is wrong. They simply can’t afford it. AMD atm is going the way of appeasing via social media, “the friend from Reddit” way of advertising. It’s quite the opposite of being greedy and corrupt, it simply wouldn’t fly.

 

[InvalidError]

Fact is, AM5 will be a long going platform

Fact is AMD struggled to make Zen 2 work on AM4 and were it not for AMD previously trumpeting that AM4 would last four years, Zen 2 would quite possibly have gotten a new socket to solve whatever issues AMD ran into. So far, there has been no official announcement on how long AM5 is going to last, only Su projecting in her CES speech that it "it should be similar" followed immediately by "I have no specific number."

 

[Wolverine2349]

Mixed, sometimes you have to change the socket sometimes not. If it was Skylake with a different name, like Kaby Lake etc, the change was less likely needed, but of course I would need the technical details to make a 100% statement. Fact is, Intel had a cadence of using the same socket for only 2 gens, AMD did the opposite, they changed sockets when it was needed so, so nobody trusts intel anymore when they say it was a technical reason.

Fact is, AM5 will be a long going platform so it’s not needed to have a outdated memory standard built in, it’s so not needed, it’s planned for the long term, unlike intel which was just planned for 2 gens again, “12th” and “13th”.
And I wanna add one more thing: AMD isn’t in the position of corruption to do the “Intel” it’s not a company with 80% or more market share. Alone for this reason, you alleging AMD to be greedy, is wrong. They simply can’t afford it. AMD atm is going the way of appeasing via social media, “the friend from Reddit” way of advertising. It’s quite the opposite of being greedy and corrupt, it simply wouldn’t fly.

Do you think Intel will change their behavior now that AMD is providing competition and keep same socket for longer and be less corrupt and greedy??

 

[spongiemaster]

Do you think Intel will change their behavior now that AMD is providing competition and keep same socket for longer and be less corrupt and greedy??

No, because changing sockets more frequently is unlikely to be making Intel any significant amount of money. Many enthusiasts seem to think the world revolves around you. It doesn't. The overwhelming majority of CPU's are sold in business computers that will never get upgraded. In Q4 2021, AMD had 25.6% of the global CPU market. Same quarter, Lenovo had 23.5% of global PC shipments. HP was at 21.2% and Dell at 18.6%.

Intel doesn't change sockets so frequently to make money off the miniscule niche of a niche market of home builders who want to upgrade their CPU, but end up replacing their whole system because they can't. It makes more sense from a business standpoint and simplifies development to do what Intel does so they don't end up with the dumpster fire situation AMD put themselves in by promising support for a socket longer than is practical. It was so bad, AMD tried to back out of the promise until they got blasted by the online mob.

 

[abufrejoval]

Faster memory mostly comes into play when all caches miss. Since AMD will have the cache advantage at least until Raptor Lake launches, I'd expect Zen 4 to be even less memory-sensitive than Alder Lake provided Zen 4 has enough IF : DRAM clock flexibility to allow people to run fast IF (~1.9GHz on Zen 3) without locking themselves into premium DDR5 bins.

DDR5 is an APU choice. What you say may be true for CPU workloads, but while all GPUs need bandwidth, iGPUs may be even more sensitive yet. AMD (and Intel) have done an incredible job with Tiger Lakes Xe and on Cezanne to squeeze an incredible amount of graphics performance from ~45GByte/s DRAM bandwidth.

Kaveri A10-7850K had almost the same bandwidth at DDR3-2400 and also 512 SMs, but ran against a DRAM bottleneck at only 720MHz GPU clock (the performance difference from 384 to 512 SMs was tiny). Cezanne has the same number of SMs but clocks to 2GHz: without extensive wizardry on cache memory all those clocks would be for naught, because even at 720MHz on Kaveri the iGPU was mostly RAM bandwidth limited.

Similar with Intel: Where the 48EU iGPUs in earlier Iris Plus iGPUs required 64/128MB EDRAM to achieve 50% better than their 24EU bethren, the 96EU XE iGPU on TigerLake makes do without EDRAM yet delivers pretty much 4x linear speed increases from the 24EU 14nm UHD iGPUs. My only explanation is extremely smart use of L3 caches for textures, but those are extremely limited so iGPUs need to resort to DRAM, where bandwidth rules by far over latencies.

If DDR5 can get 70-80GB/s bandwidth the Zen4 iGPUs seem ready to tackle the FHD gaming market without discrete GPUs and thus pay for the DDR5 premium. And if you add or expose the ECC features native to DDR5, I don't even see a price increase, just a sensible paradigm shift back to parity on RAM, which incidentally the original IBM-PC from 1981 had, too.

I know, I was there...

 

[InvalidError]

DDR5 is an APU choice.

IGPs were always a given, though nearly everybody here who does any meaningful amount of gaming already owns something at least just as fast as Phonenix' IGP is going to be, so it isn't all that much of a selling point for most people who put a new system together in the last 4-5 years.

 

[gruffi]

For more than double the cost, which is horrible value.

If you buy a 500 EUR/USD processor then spending 100 more for memory won't kill you. I bought a 32G DDR4-3200 kit in 2019 for ~150 EUR. At the moment I could buy a 32G DDR5-4800 kit for ~200 EUR. That's far from horrible. And it might be even less expensive when Zen 4 / Raptor Lake arrives. The more demand of DDR5 the more it will be produced. Which can only be good for adapting a new standard.

For 4-5X the cost, which is even worse value.

32G DDR5-6000+ kits start at ~300 EUR. Sure, that's still much more than 32G DDR4 kits which start at ~100 EUR. But not 4-5x the cost. But as I said, you don't need that. DDR5-4800 is fine for most people. I know a lot of people who not even use DDR4-3200.

Can you name any component for which any sane person would recommend spending 100+% more for ~0% net performance gain

Such arguments are pointless. The whole platform counts. Memory alone doesn't give you any performance. If the new platform gives me a sufficient advantage over my current platform then pricing is justified. If you buy a DDR4 board now and upgrade to DDR5 later makes no real difference. You still have to buy at least 3 components. Two boards and one memory kit or one board and two memory kits. You actually win nothing. Or do you want to stay on DDR4 forever? ^^

And by the way, I rather upgrade my memory than my board. Memory is a simple drop in. Replacing a board actually needs some work. Memory also is easier to resell than a board, in my opinion.

 

[Wolverine2349]

If you buy a 500 EUR/USD processor then spending 100 more for memory won't kill you. I bought a 32G DDR4-3200 kit in 2019 for ~150 EUR. At the moment I could buy a 32G DDR5-4800 kit for ~200 EUR. That's far from horrible. And it might be even less expensive when Zen 4 / Raptor Lake arrives. The more demand of DDR5 the more it will be produced. Which can only be good for adapting a new standard.


32G DDR5-6000+ kits start at ~300 EUR. Sure, that's still much more than 32G DDR4 kits which start at ~100 EUR. But not 4-5x the cost. But as I said, you don't need that. DDR5-4800 is fine for most people. I know a lot of people who not even use DDR4-3200.


Such arguments are pointless. The whole platform counts. Memory alone doesn't give you any performance. If the new platform gives me a sufficient advantage over my current platform then pricing is justified. If you buy a DDR4 board now and upgrade to DDR5 later makes no real difference. You still have to buy at least 3 components. Two boards and one memory kit or one board and two memory kits. You actually win nothing. Or do you want to stay on DDR4 forever? ^^

And by the way, I rather upgrade my memory than my board. Memory is a simple drop in. Replacing a board actually needs some work. Memory also is easier to resell than a board, in my opinion.

I got a brand new 12900KS and would have probably gone with fast DDR5 and still thought about it, but my own testing and seeing OCCT Large Data set variable test with same exact CPU and vcore and clock speeds and LLC set run 10-15C hotter scared me off.

So I stuck with fast low latency DDR4. Plus I think the jump from DDR4 to DDR5 is much less i any gains and more close to same performance and sometimes maybe worse than from DDR3 to DDR4. DDR4 still outperforms or is on par to even fast DDR5 in high FPS games mostly. Plus you cannot even run DDR5 in Gear 1 where as you could DDR4 initially when it was released. That may be why the performance benefits at least now and in near future are not there??

I also think DDR4 will be much more relevant years after DDR5 than DDR3 years after DDR4. We are and really have been hitting laws of diminishing returns in CPU IPC and memory speed increases having much performance impact. Its more core count and video card computational power increasing that seems to matter more.

 

[-Fran-]

If both Intel and AMD are greedy, why is it that AMD kept same socket for whole Zen to Zen 3 generation? Where as Intel changes them.

Are there other ways AMD is greedy where Intel is by requiring socket changes so often.

Though to be fair, AMD even though socket stayed the same, its just as much f a problem if the chipset needs to be changed for new CPUs which happened at times. Though to be fair some micro code updates AMD has allowed older chipsets to support Zen 3 going back to X470 or even X370 I think but not sure how reliable it is?? I mean even Intel stuck on LGA 775 for a long while going from the first Pentium 4 Prescott in Spring 2004 all the way to the best Core 2 Quads released in Spring 2008. Though a new chipset and board were required when Core 2 Duo hit in most cases.

AMD decided to keep the same socket because they thought it would be a good idea. I think the most simple explanation works best: build consumer trust. As you say, then Intel had to make a comeback with Core2, they did the exact same thing AMD did just now. Would it be a coincidence? I don't think so, but instead believe (yes, no hard evidence here) it may be a "text book" strategy in the industry.

No, because changing sockets more frequently is unlikely to be making Intel any significant amount of money. Many enthusiasts seem to think the world revolves around you. It doesn't. The overwhelming majority of CPU's are sold in business computers that will never get upgraded. In Q4 2021, AMD had 25.6% of the global CPU market. Same quarter, Lenovo had 23.5% of global PC shipments. HP was at 21.2% and Dell at 18.6%.

Intel doesn't change sockets so frequently to make money off the miniscule niche of a niche market of home builders who want to upgrade their CPU, but end up replacing their whole system because they can't. It makes more sense from a business standpoint and simplifies development to do what Intel does so they don't end up with the dumpster fire situation AMD put themselves in by promising support for a socket longer than is practical. It was so bad, AMD tried to back out of the promise until they got blasted by the online mob.

LOL, no. Forcing clients to upgrade each generation was making Intel ridiculous amounts of money. As you say, the consumer market is minuscule in comparison and I agree. They (Intel) were doing this in all areas, but in server it was just monkey business. AMD's biggest fighting chance was to offer clients a platform they could use and upgrade down the road (server) without spending stupid amounts of money on re-tooling, re-training and re-wiring of rooms full of machines. No done on every single SPx socket, but close enough to give them good grace, I'd say. Consumer was just to regain a bit of trust in the platform and actually make it a good deal for people to invest in AMD again. This will go away once AMD can secure a spot in people's mind and their sales fortify to the point where they don't need this "good deal" mentality anymore. Luckily for us, Intel and nVidia will never allow that to happen.

So yeah, AMD is doing this as part of their "grabbing marketshare" strategy on all fronts and, I have to say, it has worked somewhat fine. The thing to look forward to is how Intel reacts to AMD eating their breakfast and menacing eating their lunch. Will they try and do the same (they've done it before)? Will they believe their name has enough weight to keep the masses (including server) attracted to their lineups even with short upgrade cycles? I think Intel needs to adjust slightly, but not change things radically.

Regards.

 

[gruffi]

So I stuck with fast low latency DDR4. Plus I think the jump from DDR4 to DDR5 is much less i any gains and more close to same performance and sometimes maybe worse than from DDR3 to DDR4. DDR4 still outperforms or is on par to even fast DDR5 in high FPS games mostly. Plus you cannot even run DDR5 in Gear 1 where as you could DDR4 initially when it was released. That may be why the performance benefits at least now and in near future are not there??

I also think DDR4 will be much more relevant years after DDR5 than DDR3 years after DDR4. We are and really have been hitting laws of diminishing returns in CPU IPC and memory speed increases having much performance impact. Its more core count and video card computational power increasing that seems to matter more.

The higher bandwidth alone makes DDR5 very attractive for new platforms. Especially if you think about iGPUs. I think DDR5-6400 will be standard quite soon. Which will give you also in desktop systems a significant boost in memory sensitive applications. And there are other nice features like on-die ECC. If I needed a new platform now I would definitely go with DDR5. Take the cheapest 16/32G kit you can find. You won't really recognize a difference to DDR4 with better latency. And upgrade in 2 or 3 years with better and cheaper DDR5 modules.

 

[InvalidError]

And there are other nice features like on-die ECC.

Thing to keep in mind is that the reason why ECC has become standard with DDR5 DRAM chips is because it has become necessary to avoid reliability falling off of a cliff as DRAM fabs push beyond 10nm. Doesn't take much energy to flip femto-farad cells by 0.5V and this only gets worse with each cell shrink.

 

[Wolverine2349]

The higher bandwidth alone makes DDR5 very attractive for new platforms. Especially if you think about iGPUs. I think DDR5-6400 will be standard quite soon. Which will give you also in desktop systems a significant boost in memory sensitive applications. And there are other nice features like on-die ECC. If I needed a new platform now I would definitely go with DDR5. Take the cheapest 16/32G kit you can find. You won't really recognize a difference to DDR4 with better latency. And upgrade in 2 or 3 years with better and cheaper DDR5 modules.

Good point for iGPU it helping a lot Though for those with dedicated video cards, the iGPU does not matter at all. I like the latency of DDR4 much better and I gated that DDR5 caused Alder Lake to run 10-15C hotter in my experience.

DDR5-6400 with Gear 1 support and lower power consumption and not causing hotter running CPU I am on board. Until then Fast DDR4.

 

[Wolverine2349]

The higher bandwidth alone makes DDR5 very attractive for new platforms. Especially if you think about iGPUs. I think DDR5-6400 will be standard quite soon. Which will give you also in desktop systems a significant boost in memory sensitive applications. And there are other nice features like on-die ECC. If I needed a new platform now I would definitely go with DDR5. Take the cheapest 16/32G kit you can find. You won't really recognize a difference to DDR4 with better latency. And upgrade in 2 or 3 years with better and cheaper DDR5 modules.

DO you foresee DR5 6400 being standard within Alder Lake generation, or next chipset generations?? Part of me thinks about going that route, but the higher CPU temps for just DDR5 6000 that I saw on 2 different Asus motherboards with exact same CPU and clock speed and VCORE and LLC level scared me off. I am trying to build a future proof system for next gen gaming that I do not have to replace motherboard or even CPU in for at least 3-4 or more years and just do video card upgrades.

It was during an OCCT Variable Standard Large Data set test that CPU temps were 10-15C higher at exact same settings than DDR4 Asus counterpart. And that is a test that puts a 100% load on CPU, but does not generate much heat nor uses much RAM. Just checks general stability unlike small FFTs or data sets or Blend Prime95.

 

[InvalidError]

Part of me thinks about going that route, but the higher CPU temps for just DDR5 6000 that I saw on 2 different Asus motherboards with exact same CPU and clock speed and VCORE and LLC level scared me off.

If you over-volt the memory controller by 200-300mV to run at higher speeds than the CPU officially supports, OF COURSE you will end up with higher CPU temperatures. DDR5 is spec'd at 1.1V while those 6000+MT/s DIMMs call for ~1.4V. If you go from 1.2V DDR4 to 1.4V DDR4 combined with 3200MT/s to 6400MT/s, you end up with ~3X as much power going to memory IO.

2-3 years from now, there will likely be 6400MT/s memory that works fine at ~1.15V and DDR5 will still require about double the IO power simply from operating twice as fast.

 

[Wolverine2349]

If you over-volt the memory controller by 200-300mV to run at higher speeds than the CPU officially supports, OF COURSE you will end up with higher CPU temperatures. DDR5 is spec'd at 1.1V while those 6000+MT/s DIMMs call for ~1.4V. If you go from 1.2V DDR4 to 1.4V DDR4 combined with 3200MT/s to 6400MT/s, you end up with ~3X as much power going to memory IO.

2-3 years from now, there will likely be 6400MT/s memory that works fine at ~1.15V and DDR5 will still require about double the IO power simply from operating twice as fast.

Does that higher IO power cause higher RAM temps which works IMC of CPU harder and thus higher CPU temps even if the DIMM voltage is not higher?? Just like faster clock speed causes higher CPU amperage current and thus temps even if CPU vcore is the same??

 

[InvalidError]

Does that higher IO power cause higher RAM temps which works IMC of CPU harder and thus higher CPU temps even if the DIMM voltage is not higher?? Just like faster clock speed causes higher CPU amperage current and thus temps even if CPU vcore is the same??

IO power is related to the operating voltage of whatever IO is in question, Vcore has nothing to do with it.

And yes, memory needs to spend IO power following the same scaling as the CPU's memory IO does since the memory needs to drive IOs to send data to the CPU on reads and memory reads significantly outnumber writes in most workloads.

The general formula for the amount of power needed to switch capacitive loads is: P = C * V^2 * F where C is the bulk capacitance of what is being switched, V is the voltage and F is the frequency. So the part of bus power caused by parasitic IO capacitance is linearly proportional with frequency and scales squared with bus voltage. If you keep the bus voltage the same, you still have a linear increase from frequency.

 

[KananX]

Fact is AMD struggled to make Zen 2 work on AM4 and were it not for AMD previously trumpeting that AM4 would last four years, Zen 2 would quite possibly have gotten a new socket to solve whatever issues AMD ran into. So far, there has been no official announcement on how long AM5 is going to last, only Su projecting in her CES speech that it "it should be similar" followed immediately by "I have no specific number."

Didn’t struggle at all, it worked perfectly since release, better than Zen and Zen+, don’t know what you’re talking about.
And yes Su said AM5 will be a long standing platform just like AM4. AMD isn’t known for changing sockets too often, never really was.

Do you think Intel will change their behavior now that AMD is providing competition and keep same socket for longer and be less corrupt and greedy??

Only if they lose a lot of market share and are forced into a underdog “consumer friendly” role like AMD was / is. Otherwise they will keep their cadence with 1 socket for 2 gens, with sometimes minimal improvement between gens making it moot as well.