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"

 

[Topweasel]

Even more nuts when you factor in the possibility that even with a very optimistic 30% IPC uplift, Zen 4 may not be able to match the 5800X3D in MSFS, let alone beat it. Higher DDR5 speeds might help narrow the gap or even overcome it.

Maybe not. Do we know what the cache structure of Zen 4. The 12 and 16 core 5900 and 5950 respectively already come pretty close to the 5800x3D cache size. Will a dual ccd Zen4 CPU exceed the cache of the 5800x3d.

 

[Makaveli]

Not really a fan of BCLK OC for any platform, but I might give this a go if I am desperate for more performance in a few months. Gonna let some 5800X3D owners really push their CPU while I sit back.

i'm on the same boat for BCLK overclocking as so far everyone that has done it doesn't have any long term testing done. So I would be reluctant to do it on my brand new cpu.

 

[DrMrLordX]

bclk OC makes PCIe devices unstable. Really annoying. Nobody has good bus locks for PCIe.

 

[Tuna-Fish]

Maybe not. Do we know what the cache structure of Zen 4.

We do. The L2 is doubled in size, the L3 remains the same.

The 12 and 16 core 5900 and 5950 respectively already come pretty close to the 5800x3D cache size. Will a dual ccd Zen4 CPU exceed the cache of the 5800x3d.

They don't. Each core can normally only make use of the cache on the same chiplet, dual chiplet Zen 3 is more like 2x 32MB cache rather than 64MB. There are a lot of workloads where this really matters.

 

[ZGR]

MC has shipped my CPU, time to start buying parts. Hope to see some of ya on the Ryzen 5000 builders thread!

 

[nicalandia]

Do we know what the cache structure of Zen 4. The 12 and 16 core 5900 and 5950 respectively already come pretty close to the 5800x3D cache size. Will a dual ccd Zen4 CPU exceed the cache of the 5800x3d.

This is what we know so far.

Zen4 will have twice as much L2$ which is Faster than L3 by an order of Magnitude and this is one of the most profound change as the L2 has remained the same size since Zen1. And it's believed to be bigest source of IPC Boost. L3$ Remains unchanged from Zen3

Games, in general, do not benefit from Cache$ that is on another CCD, for example, the 5900X3D Prototype Had 192 Total$ and it offered the same Gaming performance as the 5800X3D

 

[BTRY B 529th FA BN]

Ridiculously fast in UT4

EDIT: @nicalandia Sorry I'm gonna have to pass on that type of oc'ing. I'd have to setup an SSD drive for it but read my above statement

 

[nicalandia]

Ridiculously fast in UT4

EDIT: @nicalandia Sorry I'm gonna have to pass on that type of oc'ing. I'd have to setup an SSD drive for it but read my above statement

So, I take it you are playing UT4 right about now? Any ScreenShots? Or is that from the reviews done recently online?

 

[BTRY B 529th FA BN]

So, I take it you are playing UT4 right about now? Any ScreenShots? Or is that from the reviews done recently online?

I just played a quick Elimination match. The difference in responsiveness from my previous setup to this, granted I gained IF and Mem frequency speed from 1866/3733 to 1900/3800 TM5 stable (checking for WHEA errors later) is more than a generational leap. From my 3800X to my 5800X was a nice increase in speed. This is like 10 fold ... however I feel I must recognize I could be experiencing the excitement over new hardware but I'll know for sure in about a week or so.

 

[nicalandia]

From my 3800X to my 5800X was a nice increase in speed. This is like 10 fold ...

So from a 1800X to a 5800X3D would be like Night and Day? all while keeping the same X370 MB

 

[BTRY B 529th FA BN]

So from a 1800X to a 5800X3D would be like Night and Day? all while keeping the same X370 MB

I never had an 1800X so I couldn't speculate on that. Games always feel super great when playing them for the first time on new stuff. It's fast, but I have to give it time to settle in.

 

[BTRY B 529th FA BN]

I'm about 50m into OCCT looking for WHEA errors at 1900/3800. With the larger amount of L3$ would there be a better test to use? I'm using AVX2, Large, Extreme, Variable, 224 MB, for OCCT settings which is suppose to find errors the fastest.

 

[igor_kavinski]

You can also try running Prime95(Blend or in-place FFTs) in parallel. Can you post screenshot of core temps?

 

[BTRY B 529th FA BN]

Unfortunately I didn't monitor temps, and I think the OCCT version I used doesn't have sensor support yet for the 5800X3D? I'm not sure. All I know is they weren't showing up. They did with my 5800X, though

 

[Markfw]

Did you update your bios ? It may work, but no sensors on your current bios.

 

[BTRY B 529th FA BN]

@Markfw I've got the latest bios for my board, 1.2.0.6 B F36a. Apps like HWinfo64 show the temps etc

 

[dr1337]

You can also try running Prime95(Blend or in-place FFTs) in parallel. Can you post screenshot of core temps?

Heres mine, temps shoot up at least 2 degrees at a time and bounce around till it settles in at 90c and begins to throttle. This is on a d15, and temps instantly drop back into the 30's after stopping the test so I'm pretty sure I didn't mess up the pasting and mounting. And while playing ArmA 3 the temps never crack 55c, its interesting. It acts a lot like my launch day 3600 it replaced.

 

[RnR_au]

Sold out in Australia apparently...

 

[Shamrock]

AAAAAANNNNNDDDDDDD...They're gone!

 

[krumme]

Many in stock here still today. Get mine today hopefully shipped yesterday. I think its to expensive but hey, us 1% min fps, 280 we absolutely need it lol.
In time this will hopefully go 30% down in price.

 

[Carfax83]

Instead of memory speeds (besides larger L2), isn't the real reason the extra ALU in Alder Lake allowing it to do more work?

I would imagine so. The UE5 Matrix City demo is apparently heavily single threaded due to being unoptimized and also because Lumen and Nanite are single threaded as well, so apparently, no current CPU can achieve a stable 60 FPS.....including the 12900K. The 12900KS should be able to hit the 60 FPS mark, but it will not sustain it.

But when you compare it to Zen 3, Zen 2 and Skylake, Golden Cove is obviously much more potent:

(I hope Zen 4 will also come with a fifth ALU, right?)

No idea to be honest. Zen 4 has so much hype behind it, that if it doesn't add additional ALUs yet manages to hit the reputed 20%+ IPC, it will be seen as an extremely impressive achievement considering Golden Cove is arguably a much wider design yet would ostensibly come up short against Zen 4. Now Raptor Cove on the other hand should improve the inefficiencies in Golden Cove's design and make it more likely to hit its full potential in terms of IPC I would think.

 

[jamescox]

This is what we know so far.

Zen4 will have twice as much L2$ which is Faster than L3 by an order of Magnitude and this is one of the most profound change as the L2 has remained the same size since Zen1. And it's believed to be bigest source of IPC Boost. L3$ Remains unchanged from Zen3

Games, in general, do not benefit from Cache$ that is on another CCD, for example, the 5900X3D Prototype Had 192 Total$ and it offered the same Gaming performance as the 5800X3D

View attachment 60321

I don’t think the 5800X3D will beat Zen 4 in much of anything, although there are occasional applications that have a resident set size where performance increases massively if you manage to fit it in cache. The L2 improvements are probably the major component, but there are a lot of other things.

I think I read that Zen 3 actually only gets the minimum latency out of the L3 with a set size of 8 MB or less since it overflows the TLB at that size. If they enhance the TLB (very likely), then average L3 latency could go down significantly, even with no L3 size increase. They may have been able to reduce the access latency itself with the switch to 5 nm, although it likely will have second generation v-cache components on the die from the start. We probably don’t get that until a little later though. Someone mentioned Bergamo, but I think it is a completely different thing, although I wouldn’t rule out it using v-cache in addition to other special sauce. Bergamo may not have sufficient un-core area such that v-cache would cover cores. It is supposed to be low power though, so covering cores may not matter. Some rumors say 2x16 MB L3 / 8-core CCX made on a more dense and lower power process. Where would you stack the v-cache? There will certainly be a v-cache Genoa.

They will have a lot more transistors to play with. If they upgrad the floating point with AVX512, then that requires a lot more bandwidth to the core; all of the data paths are likely widened. DDR5 may still help a lot, even with large caches. I haven’t read up too much on DDR5, but going to 4x32 channels instead of 2x64 should be interesting. That can decrease latency due to better granularity. The extra bandwidth can also allow for more aggressive prefetch. Even though Zen 4 is the same family, we will be getting a lot of improvements just about everywhere with the new platform and new process. It is quite amazing how well the 5800X3D can do with such an old platform.

 

[Asterox]

Ok, it is now official.

https://twitter.com/x/status/1517087945710256129

 

[Hitman928]

I would imagine so. The UE5 Matrix City demo is apparently heavily single threaded due to being unoptimized and also because Lumen and Nanite are single threaded as well, so apparently, no current CPU can achieve a stable 60 FPS.....including the 12900K. The 12900KS should be able to hit the 60 FPS mark, but it will not sustain it.

But when you compare it to Zen 3, Zen 2 and Skylake, Golden Cove is obviously much more potent:

No idea to be honest. Zen 4 has so much hype behind it, that if it doesn't add additional ALUs yet manages to hit the reputed 20%+ IPC, it will be seen as an extremely impressive achievement considering Golden Cove is arguably a much wider design yet would ostensibly come up short against Zen 4. Now Raptor Cove on the other hand should improve the inefficiencies in Golden Cove's design and make it more likely to hit its full potential in terms of IPC I would think.

The fact that a 5900x is outperformed by a 2700x in average frame rates and a 3600x outright tells me there are some major issues to work out on the AMD CPU side.

 

[Makaveli]

The fact that a 5900x is outperformed by a 2700x in average frame rates and a 3600x outright tells me there are some major issues to work out on the AMD CPU side.

Those numbers do look weird but it also also be Single CCD vs dual CCD at least with the Zen 3 parts.

There are sometimes when the 5600X and 5800X will out perform the 5900X and 5950X due to the latency in single vs Dual CCD aslong as your workload is not bottlenecked by core count.