AMD Readies Even More Ryzen 5000 Series Desktop SKUs for April

[RedBear]

When AMD transitioned from DDR3 to DDR4 there were afaik, no CPUs with support for both memory standards, as AM3+ was DDR3 and AM4 was DDR4.
As such, it's not so strange the company did the same now when they transition to a new socket, CPU architecture and chipset, even if the latter isn't really relevant to memory support.

The problem in the comparison is that AMD transitioned relatively late to DDR4 back then, IIRC Intel transitioned in 2015 with LGA 1151, AMD did it only two years later.

why would a 8c/16t zen 3 chip lose to a 6c/12t zen 3 chip at gaming?

5700 is a Cezanne part, a 5700G with disabled graphics apparently, it comes with half the cache of the 5600, which is a Vermeer part, this could impact performance in the specific comparison (5600X vs 5700G should give you a rough idea, but we don't know just how much/little slower the 5600 will be in comparison to the 5600X).

 

[bug]

To many variations maybe. DDR4 DDR5 of course chiplet swap but still you have to make two of them.

Neah, AMD just got cocky. And then reality called.

AMD is memory sensitive and I'm quite sure AM5 CPUs will exhibit same thing. DDR4 is reaching limits and I think DDR5 is making more sense for AMD. New platform new stuff it does make sense. Also it is not like it will show up tomorrow. There's still time for the DDR5 to make it more affordable and faster.

It takes DDR5-6000 to get the same overall performance you can get from DDR4-3200. And DDR4 can go much faster than 3200. Whatever limits DDR4 may be reaching, Zen4 will be obsolete by the time DDR5-6000 becomes affordable. But you go ahead, pay the tax on the new shiny, while thanking AMD they didn't make DDR4 an option.

 

[mechtech]

Too bad that R7 4700 didn’t have igpu.

 

[defaultluser]

The best part of this: AMD still has yet to release any of the parts they promised this month ( and yet here we are, already getting distracted hyping-up next month!)

 

[ncrs]

muh security patches, muh gaming on linux

The latter is becoming less and less of a meme nowadays. We'll also have to see how Microsoft will react to BHI.

That 26% is purely synthetic and it's only about I/O code. When doing I/O, the actual duration of the task is dictated by the data transfer, not by the time taken to initiate it.

Unless you're measuring stuff like IOPS or transactions/s (or to lesser extent FPS) when it does matter. Having to perform a dance like that in the kernel slows down everything, up to a point obviously.

Plus, these mitigations are a fact of life by now. They're heavier at first, people figure out how to lessen the perf hit after a while. And they affect both Intel and AMD. And ARM. And Apple, if anyone cares to look hard enough.

I'm not negating that, but the facts are that Intel has been historically way more vulnerable to mitigations ( ) than ARM and AMD. The impact has also been heavier on Intel's side - for example CPUs without PCID/INVPCID took a heavy hit from PTI mitigations.

 

[bug]

Unless you're measuring stuff like IOPS or transactions/s (or to lesser extent FPS) when it does matter. Having to perform a dance like that in the kernel slows down everything, up to a point obviously.

Yes, these things appear much more stringent when you measure them. Irl, you can barely feel a thing. It may impact servers trying to do stupendous amounts of IOPS, but even that is not very relevant for a desktop CPU/

I'm not negating that, but the facts are that Intel has been historically way more vulnerable to mitigations ( ) than ARM and AMD. The impact has also been heavier on Intel's side - for example CPUs without PCID/INVPCID took a heavy hit from PTI mitigations.

I doubt you have anything to back up that claim. It's hard to measure who was impacted the most and by how much. Just counting the number of CVEs doesn't tell you much, you have to correlate that with the severity/complexity of each CVE. Even if you did that, how much the 7th Core generation or the 1st Zen generation was impacted has little to do with the impact on AL or Zen3+/4.

 

[ncrs]

Yes, these things appear much more stringent when you measure them.

That's what benchmarks are for

Irl, you can barely feel a thing. It may impact servers trying to do stupendous amounts of IOPS, but even that is not very relevant for a desktop CPU/

Having to disable PTI on servers because its impact was noticeable to my users was enough proof for me. The internal documentation for CPU vulnerabilities, their impact and mitigation implementation leans heavily into Intel's side as well. We do not track ARM, so I only have information from general internet reporting on their issues, but we do track AMD. We're a smaller organization so we do have servers spanning multiple generations.
Their impact is also noticeable on older desktop hardware we sometimes dabble with after we're done with it at work. We usually update them with cheap SSDs if needed, use the still working Win7/8 -> 10 free upgrade from MS (since they still have OEM licenses) and ship them to schools instead of trashing them. We disable mitigations on older ones because they need all the help they can get. Context switch time is also heavily affected on some models which impacts every multithreaded load, which modern web most definitely is.
All in all, I disagree - you definitely can feel certain mitigations on certain hardware IRL.

I doubt you have anything to back up that claim. It's hard to measure who was impacted the most and by how much. Just counting the number of CVEs doesn't tell you much, you have to correlate that with the severity/complexity of each CVE. Even if you did that, how much the 7th Core generation or the 1st Zen generation was impacted has little to do with the impact on AL or Zen3+/4.

There's plenty of benchmarks on Phoronix targeting different segments for each heavily publicized vulnerability. You're right, the older mitigations have no impact on ADL, but how many more is it going to get before the generation is out of support?. BHI also affects 10th and 11th gens on desktop + a few more. What will be the impact on them? We'll have to wait for benchmarks.

 

[TheLostSwede]

Maybe not strange to you, but it's strange as f to me. Especially now since their controller is on a chiplet on its own so it's very easy to swap.

I thought you were a smart guy that a thing or two about how these things are made, but clearly not.

I'm not sure what "weird issues" you have in mind, DDR4 is working perfectly fine here. I can even run 1T command rate on sticks rated for 2T.
(I know I'm a sample of 1, I'm just saying I haven't heard of issues before.)

For one, it's apparently not easy to overclock DDR4 on Alder Lake. There was also a lot of reports of various RAM related bugs early on, but that might have been fixed by now.

DDR4 performance is up there with much faster DDR5 sticks. I would like to know what you would deem adequate if this is "pretty terrible".

Only if you have very low latency modules at 3600MHz. Most people don't own that kind of RAM.

 

[NDown]

Only if you have very low latency modules at 3600MHz. Most people don't own that kind of RAM.

It honestly aint that bad, in this video he used CL14 3200 DDR4 (which isnt so rare) and CL 38 5200 DDR5

Well, depends on your definiton of "up there" tho

 

[Deleted member 24505]

It honestly aint that bad, in this video he used CL14 3200 DDR4 (which isnt so rare) and CL 38 5200 DDR5

Well, depends on your definiton of "up there" tho

Interesting, very close twixt DDR4 and 5. I would have probably gone for 5 myself, but when i got the ADL setup it was still pretty new, and DDR5 was very expensive, and hard to get hold of. I don't really regret going with DDR4 tbh.

 

[Cutechri]

I don't really regret going with DDR4 tbh.

And you shouldn't, early adopting new RAM standards is a bad idea

 

[bug]

I thought you were a smart guy that a thing or two about how these things are made, but clearly not.

I know they're not drop-in replacements and I have a hunch AMD thought they were smart not forking the cash for another SKU, testing and stuff. You may give them a pass on that, I don't.

For one, it's apparently not easy to overclock DDR4 on Alder Lake. There was also a lot of reports of various RAM related bugs early on, but that might have been fixed by now.

Again, not sure what you're talking about. Running my RAM at 3600 already counts as overclocking. All I had to do was flip on XMP. And then I was even able to change command rate from 2T to 1T. Sure, maybe it doesn't work that well for more extreme values for DDR4, but that hardly qualifies as "weird issues".

Only if you have very low latency modules at 3600MHz. Most people don't own that kind of RAM.

Low latency DDR4 is still cheaper than low latency DDR5.

AMD dropped (will be dropping, more exactly) the ball on this one. Let it rest.

 

[seth1911]

Ryzen 3 4100 is out of garbage from renoir with 8 PCIE Lanes to the first 16x slot,
this thing will be pretty good against a 12100F for 95€ with16x PCiE 4 or 5 (depending on motherboard).

This Ryzen 3 4100 for 100$ is the same garbage for fanboys of AMD like the great 6500 XT


But yeah if u have ure pink glasses up on ure face AMD is great

 

[Vader]

why would a 8c/16t zen 3 chip lose to a 6c/12t zen 3 chip at gaming?

Go and read any review of the 5700G, it's likely the same chip as the 5700, but with iGPU. What hurts these chips seems to be the lower L3 cache

 

[stimpy88]

Since you clearly can predict the future, what's next week's lotto numbers?

It's not as if AMD could change their plans last year, when it became apparent that there would be supply issues. The issue isn't even making the actual DDR5 dies, the issue right now is shipping, packaging and testing, shipping, power regulation components, shipping, retail packaging and shipping.

Intel managed it perfectly fine.

 

[TheLostSwede]

Intel managed it perfectly fine.

Intel has always had two memory controllers in their transition SKUs, AMD not so much.

I hope all you engineers and sages are applying for jobs at AMD, so they never make any more flawed predictions.

 

[NDown]

Intel has always had two memory controllers in their transition SKUs, AMD not so much.

I hope all you engineers and sages are applying for jobs at AMD, so they never make any more flawed predictions.

AMD dropped (will be dropping, more exactly) the ball on this one. Let it rest.

 

[95Viper]

Anymore insults and off topic BS will be dealt with.
Take the arguing to PMs.

Thank You

 

[Assimilator]

AMD dropped (will be dropping, more exactly) the ball on this one. Let it rest.

I don't know if that's entirely fair:

* Zen 4 was a DDR5 design from day one
* It took the launch of ADL to demonstrate how dire the DDR5 situation is
* Zen 4 engineering samples showed up about the same time as ADL launched, which implies that the Zen 4 design is completed (except for obvious errata) and available to partners to start buildjng their products around

In other words, for AMD to decide to change the Zen 4 memory controller, or shoehorn a DDR4 version in there, would be a massive - perhaps impossible - change. Even if it is possible, for the company to make that change so (relatively) late in the day would require huge amounts of rework, which would have corresponding knock-on effects for their board partners. AMD had a poor reputation with partners after the Bulldozer days, they've somewhat rebuilt it with the Zen series but it's definitely not in the same place as the relationships those partners have with Intel, so pissing off said partners with last-minute changes would be problematic for AMD.

These "new" chips, which are just current and old-generation die harvests with different coats of paint, are very obviously a way for AMD to buy time by staving off ADL for a few more months at least. Whether that's to add DDR4 to Zen 4 (unlikely IMO) or just an attempt to allow the DDR5 situation to improve, it is definitely a somewhat desperate move that is unlikely to endear the company to consumers who have (justifiably) been ticked off about how long AMD has milked Zen 3 prices. And of course it's AMD's prerogative to do that, because they are a company and companies exist to make money, it's just that enthusiasts tend to have long memories.

 

[bug]

I don't know if that's entirely fair:

* Zen 4 was a DDR5 design from day one
* It took the launch of ADL to demonstrate how dire the DDR5 situation is
* Zen 4 engineering samples showed up about the same time as ADL launched, which implies that the Zen 4 design is completed (except for obvious errata) and available to partners to start buildjng their products around

In other words, for AMD to decide to change the Zen 4 memory controller, or shoehorn a DDR4 version in there, would be a massive - perhaps impossible - change. Even if it is possible, for the company to make that change so (relatively) late in the day would require huge amounts of rework, which would have corresponding knock-on effects for their board partners. AMD had a poor reputation with partners after the Bulldozer days, they've somewhat rebuilt it with the Zen series but it's definitely not in the same place as the relationships those partners have with Intel, so pissing off said partners with last-minute changes would be problematic for AMD.

These "new" chips, which are just current and old-generation die harvests with different coats of paint, are very obviously a way for AMD to buy time by staving off ADL for a few more months at least. Whether that's to add DDR4 to Zen 4 (unlikely IMO) or just an attempt to allow the DDR5 situation to improve, it is definitely a somewhat desperate move that is unlikely to endear the company to consumers who have (justifiably) been ticked off about how long AMD has milked Zen 3 prices. And of course it's AMD's prerogative to do that, because they are a company and companies exist to make money, it's just that enthusiasts tend to have long memories.

Of course it was too late to change course after AL happened.
Imho, they should have had a DDR4 controller as a backup the moment they put Zen4 on the drawing board. Of course nobody could have predicted DDR5 shortages. But everybody knew DDR5 would be more expensive and barely faster at first. If Zen4 launch drew close and DDR5 looked rosy, they could have silently scrapped DDR4 support.

 

[Selaya]

i mean isnt one of the big advantages of the mcm that said parts are like ... modular and interchangable?
could just do an am4 zen4 release w/ a ddr4 iod/imc, heck maybe even just recycle the current zen2/zen3 one?

 

[bug]

i mean isnt one of the big advantages of the mcm that said parts are like ... modular and interchangable?
could just do an am4 zen4 release w/ a ddr4 iod/imc, heck maybe even just recycle the current zen2/zen3 one?

They're not exactly puzzle pieces you swap out, but yes, if they decided to drop DDR4, they wouldn't have to trow away the other chiplets.

 

[Bomby569]

A 5700X3D would fill that big gap

 

[cyberloner]

my 5800x end of life @@"

 

[bug]

This is actually rather terrible news.

Since it's unlikely AMD will release SKUs with a lifespan of less than 1 year, it's more likely Zen4 will come as a handful of high-end, expensive SKUs, while these one just announced/leaked will be sold alongside Zen4. Just like they did with Zen3+.