Info 64MB V-Cache on 5XXX Zen3 Average +15% in Games

[Kedas]

Well we know now how they will bridge the long wait to Zen4 on AM5 Q4 2022.
Production start for V-cache is end this year so too early for Zen4 so this is certainly coming to AM4.
+15% Lisa said is "like an entire architectural generation"

 

[igor_kavinski]

Sad to see that people still don't agree with the only thing I had asked AMD for this generation: higher prices!

Not only prioritizing the server market, they really needed to make as much money as possible while the competition was still weak.

They are making enough from their server business. Don't give them ideas!

We don't want AMD to become fat and bureaucratic like Intel. AMD must remain a nimble hand-to-mouth company. Humans function best with limited resources. Give them too much and you get delays, laziness and all around incompetence.

 

[Hitman928]

There will be almost a year between the provided gaming performance numbers (still the same) they gave and the launch date.
I think we all feel like this is released later than what they internally planned.
And they first had almost certain more versions in mind than 8 core only.
(based on communication almost a year ago, later saying "it's hard")

I'm wondering if this 1.35V limit (a non issue for servers) is the reason for the later than expected release since they needed to figure out how they would handle that.

Obviously getting closer to Zen4 launch isn't really helping to still put a lot of effort in making more versions, 6 months from now '5800X3D' will be a thing from the past.

Do you think an upper voltage limit will remain on 5nm stacking? (Zen4)

A Zen3D SKU was never going to launch before March/April time frame. The 3D stacking technology wasn't scheduled to be ready for volume manufacturing until very late 2021. So by the time they build/test/assemble/package/ship the chips, you get to late March at the very earliest. Anyone saying it should have come before then just didn't know what they were talking about.

 

[maddie]

A Zen3D SKU was never going to launch before March/April time frame. The 3D stacking technology wasn't scheduled to be ready for volume manufacturing until very late 2021. So by the time they build/test/assemble/package/ship the chips, you get to late March at the very earliest. Anyone saying it should have come before then just didn't know what they were talking about.

Logic, how dare you. Pfft.

 

[igor_kavinski]

A Zen3D SKU was never going to launch before March/April time frame.

AMD should have mentioned that in their original announcement.

AMD's revolutionary V-Cache for Ryzen: Everything you need to know | PCWorld

There are no hard details on when you’ll be able to buy a CPU with V-Cache. About as close we can get is “production later this year,” so we’d guess it’ll arrive before the holiday shopping season begins.

AMD didn't object to these speculations. They let the FUD about V-cache CPU availability run rampant.

Intel didn't start touting their Zen 3 beating ADL gaming performance 10 months before launch. They launched the CPUs and let the reviews speak for themselves. From a PR perspective, AMD failed with V-cache.

 

[Hitman928]

AMD should have mentioned that in their original announcement.

AMD's revolutionary V-Cache for Ryzen: Everything you need to know | PCWorld



AMD didn't object to these speculations. They let the FUD about V-cache CPU availability run rampant.

Intel didn't start touting their Zen 3 beating ADL gaming performance 10 months before launch. They launched the CPUs and let the reviews speak for themselves. From a PR perspective, AMD failed with V-cache.

Why do you feel that AMD owes you a release date for a product they are showcasing at a trade show? They told us that it would go into production later in the year (which of course means retail availability will come some time after that) and a release date would be announced later. If the tech "journalists" did any research, they would have known that AMD was using TSMC's stacking tech which didn't even go into volume production until very late 2021 which gives you at best a late March release date. Seems like you should be upset at the people you look at for sources for not doing their jobs rather than AMD who didn't lie or mislead on this point at all.

 

[coercitiv]

Intel didn't start touting their Zen 3 beating ADL gaming performance 10 months before launch. They launched the CPUs and let the reviews speak for themselves. From a PR perspective, AMD failed with V-cache.

You know what Intel was doing 10 months before the ADL launch? They were getting ready for their Rocket Lake launch. And did they leave gaming reviews to speak for themselves? Suuuuuure they did.



Selective memory goes so well with cognitive dissonance. They just click!

 

[igor_kavinski]

You know what Intel was doing 10 months before the ADL launch? They were getting ready for their Rocket Lake launch.

Time frame between showing those benchmarks and actual launch was less than 4 months. AMD is LATE. They miscalculated. They should have done better. Next time, until and unless they are sure that their product will launch soon, they better keep their mouths shut.

By the way, I will shut up about this if AMD gives me a free 5800X3D for testing my patience

 

[maddie]

AMD should have mentioned that in their original announcement.

AMD's revolutionary V-Cache for Ryzen: Everything you need to know | PCWorld



AMD didn't object to these speculations. They let the FUD about V-cache CPU availability run rampant.

Intel didn't start touting their Zen 3 beating ADL gaming performance 10 months before launch. They launched the CPUs and let the reviews speak for themselves. From a PR perspective, AMD failed with V-cache.

"AMD didn't object to these speculations. They let the FUD about V-cache CPU availability run rampant."

My take.

AMD, how dare you let the idiots continue speaking.

 

[maddie]

Time frame between showing those benchmarks and actual launch was less than 4 months. AMD is LATE. They miscalculated. They should have done better. Next time, until and unless they are sure that their product will launch soon, they better keep their mouths shut.

By the way, I will shut up about this if AMD gives me a free 5800X3D for testing my patience

As an aside. Do you really believe what you write, or is it a way to encourage reactions by some members? Truly interested.

 

[igor_kavinski]

AMD, how dare you let the idiots continue speaking.

Yes, AMD. Please DO shut me up!

 

[igor_kavinski]

As an aside. Do you really believe what you write, or is it a way to encourage reactions by some members? Truly interested.

I want 5800X3D reviews. Not reactions from forum members.

 

[Hans Gruber]

I just want to point out that I have seen some reviews/leaks. Might be accurate and might not be accurate. At what point will they test AMD CPU's with either 3600mhz or 3800mhz with Cas 16 vs Alder Lake processors running DDR5? AMD runs much better with fast DDR4 sticks. When official 5800x3D reviews are published next month. They better not use 3200mhz memory, nobody runs AMD rigs below 3600mhz.

 

[Thibsie]

Sad to see that people still don't agree with the only thing I had asked AMD for this generation: higher prices!

Not only prioritizing the server market, they really needed to make as much money as possible while the competition was still weak.

They want AMD to make money, but never from their pockets.

 

[Thibsie]

Yes, AMD. Please DO shut me up!

If only they could

 

[Mopetar]

Well, I am very dissapointed that voltage and core clock control will be blocked for this chip, at least for the time being. So the chip may not scale well with voltage, I don't see that as a legitimate reason to block this control, which is always optional and YMMV anyway, but a good thing to have if you want it. Hopefully, AMD will change their mind in the future, like they did with Vermeer on B350 boards (finally).

I don't know how the silicon behaves, but it's a new product using a new manufacturing process and that likely makes it a bit more fragile than people are used to. If exceeding the limits has a higher than normal tendency to brick products, then AMD probably made the correct call locking it down.

People grumbling about being limited is better than three reviews about how they tried to OC their CPU and wound up with an expensive paperweight. Then people would be posting here asking why AMD didn't do the sane thing and lock down the voltage. They weren't going to win this either way, so it's a matter of choosing the option where they lose the least.

Time frame between showing those benchmarks and actual launch was less than 4 months. AMD is LATE.

Not at all. If a competitor has a superior product available on the market that you know for fact you'll be able to beat your in four months you scream about it as loud as you can from the highest mountain you can climb.

AMD has been consistent in their messaging about this product. It's a gaming CPU that may have a few other niches where it excels, but isn't going to replace their regular Zen 3 parts otherwise. Since Intel retook the gaming crown with Alder Lake there was no risk of AMD suffering an Osborne effect from talking about Zen 3D well before it was available.

It's also as others have pointed out a halo product. Most gamers aren't going to get this, or a competing top-end i9 from Intel either. A 5600 or a 12400 are more than good enough for most people who primarily game with their PCs and available at half the cost.

 

[maddie]

I want 5800X3D reviews. Not reactions from forum members.

"Strange" strategy.

 

[Doug S]

Although, I have to wonder how much 5 nm capacity will be available going forward with Apple making all of their M1 variants. Those are large chips and represent chips that would have been made by Intel previously.

Apple sells 20 million Macs a year, but most have M1 which is only 120 mm^2. It isn't like they are going to sell millions of M1 Ultra Macs, and probably don't even reach a million M1 Max Macs. So you're probably talking something like 5K wafers per month. Apple didn't decide at the last minute to transition to Apple Silicon. They knew several years ago they would need these extra wafers, and therefore TSMC knew when planning N5 production quantities.

There will another hit when TSMC starts fabbing Apple modems, replacing the Samsung capacity currently used for Qualcomm standalone modems. Qualcomm's warning they would provide "only 20% of modems for the 2023 iPhone" is a bit hard to interpret. Either iPhone 14 starts using Apple's modem this fall (so only the SE and 13 are left using Qualcomm modems by fall 2023...not sure if that's 20% of sales?) or 80% of iPhone 15s sold will use Apple's modem and 20% will still use Qualcomm. I could see that if there are enough markets remaining where LTE/5G coverage is so poor that 2G/3G support is a must. I know there will be some, but I'm skeptical such markets could account for 20% of unit sales of a new iPhone model (yeah UK/EU are keeping 2G/3G around for many years but that's due to legacy need like alarm systems, not because there isn't LTE/5G coverage so there's no reason phones need to continue supporting 2G/3G in those markets)

So the timing is unclear, and we also can't be sure how big Apple's modem will be. Mainly because I'm not sure how big Qualcomm's X60 is, which would be useful to know as a "won't be exceeded" size given that Apple's will be made on a denser process and do less assuming it drops 2G/3G. But again, Apple has been planning this for a long time, so TSMC has already planned for those wafers.

 

[Hitman928]

I want 5800X3D reviews. Not reactions from forum members.

Reading your posts, this is the only thing I can think of. . .

 

[gdansk]

I presume my RTX 3070 is held back a bit on my old 1700X. Let's say I used a 1080p monitor.
What percentage of increase do we expect in CSGO from Zen 1 at 3.8 to Zen 3D at 4.5? More than 50%? Anyone hazard some napkin math before we see reviews?

 

[Hitman928]

I presume my RTX 3070 is held back a bit on my old 1700X. Let's say I used a 1080p monitor.
What percentage of increase do we expect in CSGO from Zen 1 at 3.8 to Zen 3D at 4.5? More than 50%? Anyone hazard some napkin math before we see reviews?

What fps are you getting now?

 

[Saylick]

I presume my RTX 3070 is held back a bit on my old 1700X. Let's say I used a 1080p monitor.
What percentage of increase do we expect in CSGO from Zen 1 at 3.8 to Zen 3D at 4.5? More than 50%? Anyone hazard some napkin math before we see reviews?

Just purely from an IPC standpoint, Zen 1 to Zen 3 is about a 40% improvement, then factoring in the boost in clocks from 3.8 to 4.5, that's another 18% improvement. CSGO doesn't really benefit from V-cache it seems (it allowed the 5800X3D to tie a 5900X), so if there were no other bottlenecks in your system, I think it's fair to get a 40-50% bump in performance.

 

[gdansk]

What fps are you getting now?

It's more a thought experiment. I haven't ever run it at 1080p.

 

[Hitman928]

It's more a thought experiment. I haven't ever run it at 1080p.

This guy shows a 51% increase from a 2700x to a 5600x using a Radeon 5600XT as the GPU on CS:GO at 1080p Ultra. So with a more powerful GPU and an even bigger difference in CPUs, the difference should probably be closer to 70% I'd imagine.

 

[jamescox]

Apple sells 20 million Macs a year, but most have M1 which is only 120 mm^2. It isn't like they are going to sell millions of M1 Ultra Macs, and probably don't even reach a million M1 Max Macs. So you're probably talking something like 5K wafers per month. Apple didn't decide at the last minute to transition to Apple Silicon. They knew several years ago they would need these extra wafers, and therefore TSMC knew when planning N5 production quantities.

There will another hit when TSMC starts fabbing Apple modems, replacing the Samsung capacity currently used for Qualcomm standalone modems. Qualcomm's warning they would provide "only 20% of modems for the 2023 iPhone" is a bit hard to interpret. Either iPhone 14 starts using Apple's modem this fall (so only the SE and 13 are left using Qualcomm modems by fall 2023...not sure if that's 20% of sales?) or 80% of iPhone 15s sold will use Apple's modem and 20% will still use Qualcomm. I could see that if there are enough markets remaining where LTE/5G coverage is so poor that 2G/3G support is a must. I know there will be some, but I'm skeptical such markets could account for 20% of unit sales of a new iPhone model (yeah UK/EU are keeping 2G/3G around for many years but that's due to legacy need like alarm systems, not because there isn't LTE/5G coverage so there's no reason phones need to continue supporting 2G/3G in those markets)

So the timing is unclear, and we also can't be sure how big Apple's modem will be. Mainly because I'm not sure how big Qualcomm's X60 is, which would be useful to know as a "won't be exceeded" size given that Apple's will be made on a denser process and do less assuming it drops 2G/3G. But again, Apple has been planning this for a long time, so TSMC has already planned for those wafers.

While I am not an Apple fan, I do have Apple devices. They have obviously generally been significantly more expensive for the performance, but they have a premium brand reputation. To some extent, that is earned. I still have a 2009 17 inch Mac Pro that I still use. It needs a new battery again, but otherwise works fine for what I use it for. It will struggle with playing high resolution video due to no hardware acceleration for some modern codecs. I added added a fast SSD, which is probably a large part of why it is still usable. I also use Firefox with noscript, which reduces web page bloat significantly and has the side effect of blocking most ads without running an explicit ad blocker.

Anyway, I think Apple sales are going to be increasing with them leaving Intel behind. While they were generally lower performing for the cost, I don’t know if that is true any more, at least when power consumption and form factor are considered. I also think that the PC makers are going to need to respond to Apple devices. This is just the M1; the first generation of “desktop” level parts. What will the M2 or M3 look like? This is also reflected in server parts. I saw a phoronix review a while back where it seemed to be mostly AMD at the top, some ARM based processors second, and intel often third.

I have asked for a long time why we can’t get a laptop made more like a console. At this point, I would like an APU, perhaps with separate gpu and cpu chiplets and HBM memory. At least a memory system closer to a gpu with many DDR or GDDR channels seems reasonable. If you are going to solder memory on the board / package, then at least make it fast. Given the way memory systems have evolved lately, a change in hierarchy seems like it is over due. I guess we might just get an APU with extra SRAM cache.

 

[jpiniero]

I have asked for a long time why we can’t get a laptop made more like a console.

No need since you can just use dGPUs.