Wednesday, August 10th 2022

Ryzen 7000 Said to Have a DDR5-6000 Memory "Sweet Spot"

If you remember, there were quite a lot of discussions about memory speed "sweet spots" for both the Ryzen 3000- and Ryzen 5000-series, with the user experience not always meeting AMD's sweet spot for memory clocks. Now details of the Ryzen 7000-series memory sweet spot has arrived courtesy of Wccftech and the speed is said to be DDR5-6000. This is 400 MHz higher than the apparent official maximum memory clock speed of DDR5-5600, but as we know, the manufacturer's max memory clock is rarely the actual max. In AMD's case, things obviously work a bit differently, as the Infinity Fabric clock should ideally run at a 1:1 ratio with the memory in the case of the AM4 platform, to deliver best possible system performance and memory latencies.

That said, as we're using DDR memory, the actual clocks are only half of the memory speeds, so the IF clock is operating at no more than 2000 MHz if the memory is DDR4-4000. However, if the same applies to the Ryzen 7000-series, it appears that AMD has managed to bump the IF clocks by a not insignificant 1000 MHz, as the IF fabric would now be operating at up to 3000 MHz. This could see the Ryzen 7000-series offering better memory latencies than Intel's Alder Lake and upcoming Raptor Lake CPUs, as Intel is running DDR5 memory at a 2:1 ratio or a 4:1 ratio. AMD is said to still have a 2:1 ratio as well, but as with the AM4 CPUs, this offers worse overall performance.

Update 11:49 UTC: Yuri Bubliy aka @1usmus has confirmed on Twitter that the max IF frequency of 3000 MHz and it seems like AMD has added a range of new memory and bus related features to the AM5 platform, going by the additional features he posted.
Sources: Wccftech, @1usmus (1), @1usmus (2)
Add your own comment

70 Comments on Ryzen 7000 Said to Have a DDR5-6000 Memory "Sweet Spot"

#1
oxrufiioxo
Should be interesting for sure. Although how easy it will be to actually hit 3000 IF will be more interesting. Out of the 10 zen 3 chips I worked with only 1 hit 2000 IF stable.
Posted on Reply
#2
Sabotaged_Enigma
This again tells us how garbage is the I/O die for Zen to Zen 3 made by GF...
Glad to see this improvement. Ryzen 7000 is born at a more proper time, with 12th gen Core being the pioneer to open up DDR5 market, even though DDR5 is still relatively expensive.
Posted on Reply
#3
watzupken
Sweet spot for the processor, but not sweet spot for buyers. DDR5 6000 can be quite pricey at this point.
Posted on Reply
#4
Sabotaged_Enigma
oxrufiioxoOut of the 10 zen 3 chips I worked with only 1 hit 2000 IF stable.
I have never been able to push it beyond 1800 MHz on my 3700X. Quite annoying.
Posted on Reply
#5
oxrufiioxo
RidiculousOwOI have never been able to push it beyond 1800 MHz on my 3700X. Quite annoying.
I worked with 15 different Zen 3 chips only 4 of them did 1900 stable with low latency timings one 3600 couldn't even do 1800. Thankfully 1800 low latency in dual rank is the sweet spot in my testing...
Posted on Reply
#6
TheLostSwede
News Editor
watzupkenSweet spot for the processor, but not sweet spot for buyers. DDR5 6000 can be quite pricey at this point.
Well, the rumour mill claims DDR5 prices will drop quite a bit very soon.
RidiculousOwOI have never been able to push it beyond 1800 MHz on my 3700X. Quite annoying.
Same on my 3800X and 5800X, but could be my cheap ass RAM as well.
Posted on Reply
#7
oxrufiioxo
TheLostSwedeSame on my 3800X and 5800X, but could be my cheap ass RAM as well.
Kinda surprised your 5800X doesn't hit 1900 If. All 4 I worked with did.... All decently binned Samsung Bdie though.

Edit. My buddies 5800X can't even do 1800 but I haven't worked with it personally so not sure what's up with it.
Posted on Reply
#8
TheLostSwede
News Editor
oxrufiioxoKinda surprised your 5800X doesn't hit 1900 If. All 4 I worked with did.... All decently binned Samsung Bdie though.

Edit. My buddies 5800X can't even do 1800 but I haven't worked with it personally so not sure what's up with it.
Sorry, doh, yes, I was thinking wrong here, I did hit 1900 on both, not 1800 :oops: :banghead:
Posted on Reply
#9
oxrufiioxo
TheLostSwedeSorry, doh, yes, I was thinking wrong here, I did hit 1900 on both, not 1800 :oops: :banghead:
Yeah, I was confused for a second I vaguely remembered your 3800X doing 1900 with ok timings with that patriot viper kit you had.
Posted on Reply
#10
TheLostSwede
News Editor
oxrufiioxoYeah, I was confused for a second I vaguely remembered your 3800X doing 1900 with ok timings with that patriot viper kit you had.
Yeah, I keep mixing up 3800 (2x 1900) with 1800 for some daft reason...
Posted on Reply
#11
Bomby569
it's wccftech so it can be 6000 or any other value, but if you want the cutting edge tech you have to be prepared to pay
Posted on Reply
#12
Taraquin
We did a survey last year on Zen3 ram oc, over 60% of responders hit 1900fclk or more.

My bet is 5800-6000 will be the usual speed people get, while a few lucky on 1DPC/2dimm boards may do 6200+.
Posted on Reply
#13
Chrispy_
Okay, is that 3GHz FCLK or 1500MHz FCLK?

If it's 3GHz, that's one hell of a leap forwards from Zen3...
Posted on Reply
#14
tabascosauz
RidiculousOwOThis again tells us how garbage is the I/O die for Zen to Zen 3 made by GF...
Glad to see this improvement. Ryzen 7000 is born at a more proper time, with 12th gen Core being the pioneer to open up DDR5 market, even though DDR5 is still relatively expensive.
IFOP is more the problem than the IO die, but yeah in 2021 we paid extra for new CCDs attached to IF and IO die from 2019......in hindsight, it really was a rather bold price hike (ignoring context of complete lack of competition from 11th gen)

Come to think of it, AMD has been content to hand the DDR4 crown back to Intel for the past few years. Nothing haopened on the chiplet side, Renoir and Cezanne were a step up but low effort on AMD's part and ignored (aside from a brief stint where 4750G took the WR I think?). I wouldn't be surprised if we see 3 years worth of UMC and IF development all stacked in 7000.

If the rumor is really true and AMD is doing 3000 1:1 leaving intel behind in Gear 2, I'd bet money that they've come up with a new way to link CCD-IOD (or even an extra CCD-CCD). old ass IFOP needs to go. For intel's upcoming chiplet venture (14th) they're bringing out EMIB.

I still have little confidence in their ability to roll out day 1 processors that lives up to those promises. If there's any lesson to be learned in these past three years, wait two or three months if you want a good chance of clocking well.
Posted on Reply
#15
JD100
So basically these early DDR5 kits we have out today already exceed the Zen4's ability to scale (in 1:1) for future faster kits.
Bamboozled by AMD IOD once again :( But on the plus side it's still in 1:1 ratio, rather than 1:2 on Intel, so there could be major latency benefits for gaming scenarios.

I wouldn't hold my hopes up for most people achieving the 3000 IF lottery. Most likely the majority will settle at 5600-5800 if prior Ryzen generation are anything to go off of.
We'll see.
Posted on Reply
#16
Chrispy_
JD100I wouldn't hold my hopes up for most people achieving the 3000 IF lottery. Most likely the majority will settle at 5600-5800 if prior Ryzen generation are anything to go off of.
We'll see.
It depends what WCCF's source is. If they spoke to Robert Hallock then he's been right on the money for both Zen2 and Zen3's launch, saying that most samples will achieve 3600 and 3800 respectively.
Whilst there have been a handful of chips that couldn't handle that, it's always been lowest end of the product stack that didn't quite make those sweet spots.
Posted on Reply
#17
Jimmy_
watzupkenSweet spot for the processor, but not sweet spot for buyers. DDR5 6000 can be quite pricey at this point.
u r spot on :)
Posted on Reply
#18
Wirko
TheLostSwedeWell, the rumour mill claims DDR5 prices will drop quite a bit very soon.
I told AMD to buy a couple thousand wafers of those PMICs when they were the bottleneck in DDR5 manufacturing ... so they can release them to memory makers at the right time, possibly subsidised too. Did they listen?
Posted on Reply
#19
jesdals
I am looking forward to the real life test and optimization of these cpu - Inifinity settings at 3000 would be a remarkable boost
Posted on Reply
#20
TheLostSwede
News Editor
Chrispy_Okay, is that 3GHz FCLK or 1500MHz FCLK?

If it's 3GHz, that's one hell of a leap forwards from Zen3...
1:1 should be 3000 MHz no?
My FLCK is running at 1900 MHz 1:1 with my 3800 MHz DDR4 memory.
Chrispy_It depends what WCCF's source is. If they spoke to Robert Hallock then he's been right on the money for both Zen2 and Zen3's launch, saying that most samples will achieve 3600 and 3800 respectively.
Whilst there have been a handful of chips that couldn't handle that, it's always been lowest end of the product stack that didn't quite make those sweet spots.
This is the "sweet spot" though, which was always a little bit higher than what most chips could do.
Posted on Reply
#21
Axaion
regressing to 4 cores per ccx scares me, if that part is correct.
Posted on Reply
#22
RedelZaVedno
The "Sweet Spot" costs 120€ (for 16 gigs) and 255€ (32GB). Double the price of DDR4 3600 CL18. All I can say to AMD atm is: "thanks but no thanks. Keep your sweet spot. I have a very decent alternative in AM4 and LGA 1700."
Posted on Reply
#23
Punkenjoy
Axaionregressing to 4 cores per ccx scares me, if that part is correct.
The only reason i would see AMD switching back to 4 cores per CCX is if they greatly improved CCX to CCX latency and were able to share L3 accross CCX. Not impossible, but i can't see why AMD would have to go back on something that greatly improved their performance.

My theory, for what it's worth: The leaker assume 4 core per CCX because the I/O die could connect to 4 CCX. (like Zen 2/3 I/O die). Since Zen 4 seem to only have room for 2 CCD, he is probably assume it's back to 2 CCX per die. I only see that happening if AMD put Zen4c on desktop (with or without 3d-vcache). I suppose but not sure that those have 2 8 cores CCX per CCD.

But a theory that i put a bit less weight into would be that Zen 4 have a new topology where they would have a ring bus and a shared l3 for all the 8 cores of the CCD but they would have 2 independent bi directional IF link to the I/O die per CCD instead of 1.

Possible, but in my opinion, it's probably just the I/O die have 4 IF link and he is making assumption.
Posted on Reply
#24
ir_cow
My suspicion is 6000-6400 range. But we won't know until its in more hands and tested. Remember not all Ryzen CPUs from the same generation could do the same FCLK.
oxrufiioxoI worked with 15 different Zen 3 chips only 4 of them did 1900 stable with low latency timings one 3600 couldn't even do 1800. Thankfully 1800 low latency in dual rank is the sweet spot in my testing...
Dual Rank says it all here. Zen3 single rank is easy to get 1900-2000. For Dual Rank you had to up the voltage and prey to the silicon gods 1900 Mhz would work.
Posted on Reply
#25
Wirko
AMD will once again have a single variant of CCD silicon (not counting the 4c) and I see it as a disadvantage. A new die probably wouldn't cost more than a few million $ to develop. Then they would have a 10-core and a 6-core variant for example, allowing them to offer a range of 4- to 20-core Ryzens with more flexibility and less waste. It wouldn't hurt the Epyc part either, as now there are models having anywhere between 1 and 8 working cores per die.
Posted on Reply
Add your own comment
Nov 21st, 2024 11:07 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts