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"

 

[nicalandia]

There really isn't a clear winner since both have a few games where they just crush the other. The average is close enough that it may as well be margin of error.

However, at this point in time the AMD CPU and DDR4 memory are considerably cheaper. Even if you do play one of the titles where Intel does win by double digits, is it worth the $400+ it will cost for that?

If you already have an AM4 board there's good odds the 5800X3D is supported. HUB did some testing for Zen 3 on old X370 boards and found that there's maybe a 1% performance loss on average over the newest boards.

Unless you're a competitive CS player, it's hard to justify the extra cost for Intel. You'd be better off just getting something like a 12400 which will get you most of the performance at a much lower price and event upgrading to a 13000-series CPU.

AMD already bumped prices with Zen 3, to some degree because they could, but their costs aren't going down with Zen 4. The chiplets aren't any really smaller even with the move to 5nm and the IO die is using a much more expensive process now as well. Maybe AMD uses the Zen 3 cost jump to eat those extra costs, but there's no guarantees there. However, I don't think Intel can keep charging several hundred dollars more than AMD if they hope to be any kind of winner beyond on paper.

Had they included MS Flight Simulator the % Would have been larger lead for AMD

 

[lobz]

Huh??? Those results are all over the place. Based on AT alone, I certainly dont see a clear winner.

I don't think he meant that comment 100% literally

 

[jamescox]

There really isn't a clear winner since both have a few games where they just crush the other. The average is close enough that it may as well be margin of error.

However, at this point in time the AMD CPU and DDR4 memory are considerably cheaper. Even if you do play one of the titles where Intel does win by double digits, is it worth the $400+ it will cost for that?

If you already have an AM4 board there's good odds the 5800X3D is supported. HUB did some testing for Zen 3 on old X370 boards and found that there's maybe a 1% performance loss on average over the newest boards.

Unless you're a competitive CS player, it's hard to justify the extra cost for Intel. You'd be better off just getting something like a 12400 which will get you most of the performance at a much lower price and event upgrading to a 13000-series CPU.

AMD already bumped prices with Zen 3, to some degree because they could, but their costs aren't going down with Zen 4. The chiplets aren't any really smaller even with the move to 5nm and the IO die is using a much more expensive process now as well. Maybe AMD uses the Zen 3 cost jump to eat those extra costs, but there's no guarantees there. However, I don't think Intel can keep charging several hundred dollars more than AMD if they hope to be any kind of winner beyond on paper.

Isn’t there going to be some desktop APUs with up to 8 cores? I would expect lower end parts to use those rather than the chiplet plus IO die. They likely will not clock as high, but the monolithic chip may have a latency advantage.

 

[Mopetar]

Had they included MS Flight Simulator the % Would have been larger lead for AMD

Sure, but had they included some game where Intel outperforms AMD by a similar amount the average would swing the other direction. Or if they include both, it remains roughly the same because they offset. I forget who had done it, but someone did a 50 game comparison and with that many games even adding a 30% outlier on either side doesn't even move the average a full percentage point.

Isn’t there going to be some desktop APUs with up to 8 cores? I would expect lower end parts to use those rather than the chiplet plus IO die. They likely will not clock as high, but the monolithic chip may have a latency advantage.

The IO die GPU is just a nice to have because it means you don't need a discrete GPU to run a desktop Ryzen anymore. If you just want to crunch numbers and aren't going to game at all, that saves you the trouble of having to spend extra for a low-end GPU, which over the past year could easily cost half as much as a top-end 16-core CPU.

We'll still see single chiplet Ryzen parts just because some IO dies will be defective in a way that wouldn't allow them to connect two chiplets. Having two separate dies also allows for greater flexibility in creating bins. The APUs may also have lower power targets which means performance will be constrained by TDP.

 

[Markfw]

Sure, but had they included some game where Intel outperforms AMD by a similar amount the average would swing the other direction. Or if they include both, it remains roughly the same because they offset. I forget who had done it, but someone did a 50 game comparison and with that many games even adding a 30% outlier on either side doesn't even move the average a full percentage point.



The IO die GPU is just a nice to have because it means you don't need a discrete GPU to run a desktop Ryzen anymore. If you just want to crunch numbers and aren't going to game at all, that saves you the trouble of having to spend extra for a low-end GPU, which over the past year could easily cost half as much as a top-end 16-core CPU.

We'll still see single chiplet Ryzen parts just because some IO dies will be defective in a way that wouldn't allow them to connect two chiplets. Having two separate dies also allows for greater flexibility in creating bins. The APUs may also have lower power targets which means performance will be constrained by TDP.

I think the point is that no matter how you look at "who wins more games", the 5800x3d certainly wins at least 1/2 or more and does so at HUNDREDS of $$ less than the Intel it competes with.

 

[Mopetar]

I think the point is that no matter how you look at "who wins more games", the 5800x3d certainly wins at least 1/2 or more and does so at HUNDREDS of $$ less than the Intel it competes with.

That was my initial point. Unless you specifically play one of those games where Intel does have a 20%+ edge, the extra $400+ you'd need to spend to get it just isn't worth it.

 

[MadRat]

$400 for a 5% edge when money is on the line for top gamers, is cheap. Its worth having both systems if you need them. You never want to lose because of your system.

 

[igor_kavinski]

I think the point is that no matter how you look at "who wins more games", the 5800x3d certainly wins at least 1/2 or more and does so at HUNDREDS of $$ less than the Intel it competes with.

And at much lower power. Hard not to love this CPU for gaming.

 

[Mopetar]

$400 for a 5% edge when money is on the line for top gamers, is cheap. Its worth having both systems if you need them. You never want to lose because of your system.

On average it's $400 (possibly closer to $600 once board and RAM are accounted for) for maybe a .5% edge. You'd also need to be quite wealthy to have separate systems and getting a top-end GPU for both now makes this a matter of closer to $4,000 once the full system cost is calculated. Realistically if you're going that far you'd need 4 machines since some games see large performance differences based on the GPU as well and there are a few titles where a 6900XT will beat NVidia's flagship GPU.

Anyone playing in professional competitions only needs whichever CPU is best suited for their game of choice. Sometimes it doesn't matter if the competition supplies the machines since everyone is on even footing in those cases. The number of people who'd buy both is low. For most consumers the $400 you can save with Zen 3D is $400 that can be spent on other components.

There are some people who do primarily play just a single title even if it's not competitively or professionally and for them it may make sense to prefer one CPU over another. However, in those cases it may make more sense to get a 12700K or even a 12600K given that those can still beat a 5800X3D and won't require paying a premium for it.

If you're a dyed in the wool Intel fan who would rather be dead than team red at least you've got some options. It's nothing nearly as bad as the Bulldozer years where the best you could hope for is matching up against a midrange part, but for anyone who doesn't have a really good reason (or just a strong inclination) for picking Intel, it's hard to recommend doing so when it comes to a top end gaming PC.

 

[MadRat]

They were talking about 20% margins not being worth $400. But in the past people spent buku on a 5% edge. They are going to buy the best for the experience. $400 is chickenscratch for those types.

 

[Markfw]

They were talking about 20% margins not being worth $400. But in the past people spent buku on a 5% edge. They are going to buy the best for the experience. $400 is chickenscratch for those types.

but (arguably) The 5800x3d is the best...And less expensive.

 

[Insert_Nickname]

$400 is chickenscratch for those types.

Ah. The more money then sense crowd.

It still a relatively cheap hobby compared with some others.

 

[MadRat]

Ah. The more money then sense crowd.
It still a relatively cheap hobby compared with some others.

Quite the opposite. The idea that someone would go cheaper to have a lesser system for the game they want to dominate is laughable. Its a small percentage that actually play at that level.

AMD doesn't need to waste energy on that crowd. Those same types will dump Intel for AMD if thats the better system for the game, even if it means they lose money.

I never underestimate how preoccupied with performance it is for those hardcore gamers. To suggest otherwise demonstrates naïvety.

 

[Insert_Nickname]

I never underestimate how preoccupied with performance it is for those hardcore gamers. To suggest otherwise demonstrates naïvety.

You don't just find that type in gaming BTW. I don't begrudge people who do what floats their boat. So long as they make their own money they're free to spend it however they see fit.

I just find it foolish. It's the same crowd who'd switch off textures in Quake2 matches 20-odd years ago just to chase those last few FPS. Didn't matter squat then, doesn't matter now. I can still hit them just fine.

 

[blckgrffn]

One thing I have to say about the 5800X3D - and I concede my board just might be the best in this regard - is that it's temp can spike hard and bring fan speeds up with it quickly.

It's super annoying to me... I just installed a Thermalright Frost Commander (it's a beast) and still can't keep it from spiking fast, which leads me to believe A) so much power in one CCD stacked is just going to have thermal dissipation problems (I mean the 5800X had this without the cache) or B) my board has fan ramping issues/power delivery issues that make this worse somehow.

I replaced a 5900X with the 5800X3D and DX 12 results for Borderlands 3 at my settings (2k,DX12, mostly maxed settings on my Radeon 6800) the benchmark dropped from 109 fps to 99 fps, but way less hitching and stuttering, butter smooth in most places and the hitching still present was less impactful. I guess it's smoothness vs avg frames in this case.

 

[nicalandia]

Sure, but had they included some game where Intel outperforms AMD by a similar amount the average would swing the other direction.

Show me a Game that was Not Included with the a similar difference in Frames as MS Flight Simulator but to the Advantage of Intel... Show me and this forum that such game exist and it was "conveniently left out"..

 

[igor_kavinski]

the hitching still present was less impactful.

That might go away completely if the 5800X3D is paired with CL14 RAM.

 

[Mopetar]

Show me a Game that was Not Included with the a similar difference in Frames as MS Flight Simulator but to the Advantage of Intel... Show me and this forum that such game exist and it was "conveniently left out"..

I don't know if there are any games that see that much of a difference, only in Intel's favor. I haven't looked into it extensively, nor do I care about trying to defend Intel. However, I think some people are exaggerating the effect of the results that adding MS Flight Simulator will have.

LTT benchmarked the game and saw a 22% (at 1080p) advantage for the 5800X3D over the 12900KS.



Add it to the 40 game benchmark from Techspot/HUB (who I believe said they didn't include the game because they couldn't get it running correctly on their Intel platform) and it doesn't change the overall average by a noticeable amount. This comped the 12900K to the 5800X3D, so it's not quite Intel's best, but I don't think it changes the overall results by much. Maybe it's 1% in Intel's favor, or even just flat even.

From their results it wouldn't even be the biggest outlier in the list. Doesn't really move the overall average either because they have too many titles for even one more game, even one heavily favoring one CPU by 20% to matter. MS Flight Simulator certainly is a strong outlier, even among other outliers, it's not as though there aren't games where Intel does better that on average it comes out in the wash.

The only way it matters is if you throw it into a six game benchmark in which case it can swing it by 4% (or more depending on what you compare it against) but a small sample size like that isn't as good because of that very reason. Unless you're going to contend there are a lot of games like MS Flight Simulator where AMD wins hard enough to make it an extreme outlier that are being excluded then your own argument can be used against you. Maybe there's some game out there that heavily favors Intel by that much, but I don't care to go looking for it. If it does exist, it may be just as or even more niche than Flight Simulator so I'm not eve sure you could make a compelling argument to include it as part of a benchmark suite other than to show the extremes that might exist at the other end.

My conclusion is still the same. MS Flight Simulator doesn't matter as much as people think if we're just considering averages. By all means get the 5800X3D if that's your game because it's hands down the best you can get. Otherwise adding it to a massive list of other benchmarks isn't going to move the average more than a fraction of a percentage point. Because it can't meaningfully change that average result the argument shifts to other factors and AMD kills it on cost, especially if you already have an AM4 board.

 

[igor_kavinski]

https://twitter.com/x/status/1534355166656331776

What does Microsoft want with the 3D V-cache team lead?

 

[Timorous]

https://twitter.com/x/status/1534355166656331776

What does Microsoft want with the 3D V-cache team lead?

Xbox Series XL?

 

[JoeRambo]

What does Microsoft want with the 3D V-cache team lead?

To provide 3D stacking know-how to Microsoft's rumored silicon projects?

I think several years before on this forum people were already reaching conclusion that past war of Clock gave way to Core count war and that is about to give way to Cache size "war". And we have seen now companies like AMD, Nvidia to fully embrace it to give what is essentially one time boost to performance and perf/watt.

Does not take rocket scientist to see why a cloud provider like MS would not want to keep competitive with Amazon and bulding ARM SoCs + advanced packaging them is the way to go. Imagine something like Graviton 3, but instead of anemic L3 it has 1GB of LLC? Cloud providers dream chip.

 

[moinmoin]

https://twitter.com/x/status/1534355166656331776

What does Microsoft want with the 3D V-cache team lead?

The better question is what and where is Microsoft Silicon? It was rumored to be Azure's counterpart to AWS's Graviton. But likely due to Azure's excellent partnership with AMD getting new Epyc gens first the urge to push an in-house Arm solution isn't there right now. There also had been rumors the Arm effort may be in partnership with AMD as well, with AMD's recent mention that they are looking at semi custom solutions containing ARM cores. If nothing else Agarwal's hire shows the Microsoft Silicon project still exists. But I wonder when we will see a public result from that.

 

[LightningZ71]

At some point soon (or just recently) the exclusivity agreement with Qualcomm expires. It is my opinion that Microsoft wants to move forward with a big WARM (windows on ARM) initiative to better compete with Apple. They are establishing the MS Surface line as a premium solution with similar levels of "nearly impossible to repair or upgrade" associated with each one. They have a store like Apple's walled garden. If they can take control of their silicon, they can make themselves independent of Intel and AMD and operate on their own schedule. Vertical integration is the name of the game at the top. The only piece that MS will be missing is mobile phones. Google isn't too far from that level of integration either. They have the phones, they have the Chromebook ecosystem, and they have a software stack. They are moving back into the tablet space now and are getting a smart watch as well.

I feel strongly that we are moving to a world that's largely built around Apple, MS and Google/Alphabet from the top down. Where that leaves Linux/Intel/AMD is anyone's guess.

 

[igor_kavinski]

I wouldn't worry too much about it. ARM is mostly used by technically clueless people. It's only success story is Apple and Android and the bulk of those sales are phones that no one in their right mind would use for serious work.

I'm not saying that nothing will change in the future but for ARM to invade the desktop, they have to work a LOT harder and come up with a really effective way to emulate legacy x86/x64 software, including PC games. Maybe Microsoft will accelerate the transition with their x86 emulation in Windows ARM edition but it will be years before the emulation speed is acceptable enough for people to stop wanting an x86 CPU.

 

[LightningZ71]

Remember, MS is rolling their own ARM chips (though, they likely have partners in the segment). Apple proved with Rosetta 2 that you can couple custom software with hardware tweaks to achieve exceptionally good emulation/translation performance from x86 to ARM. This is all about MS and how much time/money they want to throw at the problem. For 90% of the total market, an X1/3xA710/4xA510 processor with a decent memory subsystem and iGPU is all they'll ever need in a computer/laptop/tablet/phone/etc. It's all in the platform execution.