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

Page 129 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

eek2121

Golden Member
Aug 2, 2005
1,800
1,982
136
For a while I've had an idea of reviewing CPU's/platforms with a completely subjective test suite. How snappy does it feel? Opening windows, closing them, opening everyday programs like browsers/office/etc and doing basic tasks in them and closing them. I think the problem with this would be consistency, although with some of the advanced scripting reviewers have done for other test suites before perhaps a "normal user" workflow could be scripted and the time it takes measured.

I think most people assume the fastest CPU/platform in benchmarks will be the most snappy and responsive for everyday computing, and that might be right.... but how do we know?
That is where Apple’s M1 shines. One of the biggest things you notice on an M1 based Mac is how all the UI wonkiness goes away.

I doubt the 5800X3D does much for this, however. Apple did it by optimizing their scheduler for the new chip.
 

ZGR

Golden Member
Oct 26, 2012
1,920
419
136
Do you have fTPM disabled in bios as well as AMD PSP 11.0 Device disabled through Device Manager? If not you need to do both to get rid of system stutter caused by fTPM and AMD PSP.. IF that's what's causing it.
Update. fTPM was enabled by default. Followed this article to disable it:

Looking inside device manager, I do see the Security Devices section containing AMD PSP 11.0. I just disabled it.

Re-running the Siege benchmark isn't exactly a fair test because I didn't have XMP enabled and was GPU limited at 1080p. Either way, I don't feel the stutters that I did when I ran it before. Thanks so much for the tip! This game is putting nearly 100% load across all cores.

siege2.jpg
 

biostud

Lifer
Feb 27, 2003
16,072
1,374
126
The new MSI BIOS is also a special one since it is based on the upcoming AMD AGESA 1.2.0.7 BIOS firmware. From the changelog provided, the AGES 1.2.0.7 offers:

Fix the fTPM issue that causes game stuttering.
300 series chipsets support the latest Ryzen CPU/APU such as Ryzen 7 5800X3D

 

dr1337

Member
May 25, 2020
183
299
106
Space Engineers
I actually have this game but never played it more than once so I reinstalled and fired it up. I was really hoping there would be an up to date benchmark suite on the steam workshop but sadly they're all old and none of them will load for me. Im very GPU limited in this game, my rx580 dies if I look at grass (max textures and model detail even at 4k is fine tho, lol), but with slightly tweaked settings I could maintain 1080p 120hz in the multiplayer racing map no problem. If you have a build or world in mind that is extremely CPU heavy I'm more than willing to fire it up and see how it does.
 

LightningZ71

Golden Member
Mar 10, 2017
1,236
1,250
136
Space engineers is a bit broken internally unfortunately. The developers are working on a complete game engine change that should improve things.
 

jamescox

Senior member
Nov 11, 2009
506
822
136
I believe he was saying that when I reference The geometric mean result from a Pseudo 5950X3D whaling on The regular 5950X. and Alder Lake The thing is that I did not take it from Phoronix but directly from the source


View attachment 60710
Why would you include a 16-core Epyc? They are not comparable at all. Even the normal 16-core Epyc has 128 MB L3 (4 CCX * 4 cores). The the 73F3 part has 256 MB (8 CCX * 2 cores). The 7373X is still 8 CCX * 2 cores for 768 MB. You could already get much larger cache per core with Epyc parts. It might be fun to see how the 72F3 (8 CCX * 1 core) with 256 MB for only 8 cores does, but it isn’t very relevant. A Ryzen 9 5950X3D, if it existed, would be 192 MB total. I don’t know if AMD would want such a part in the mix since it may compete with much more expensive Epyc parts in some cases. That might be part of why they didn’t produce it.
 

MadRat

Lifer
Oct 14, 1999
11,767
117
106
Why would you include a 16-core Epyc? They are not comparable at all. Even the normal 16-core Epyc has 128 MB L3 (4 CCX * 4 cores). The the 73F3 part has 256 MB (8 CCX * 2 cores). The 7373X is still 8 CCX * 2 cores for 768 MB. You could already get much larger cache per core with Epyc parts. It might be fun to see how the 72F3 (8 CCX * 1 core) with 256 MB for only 8 cores does, but it isn’t very relevant. A Ryzen 9 5950X3D, if it existed, would be 192 MB total. I don’t know if AMD would want such a part in the mix since it may compete with much more expensive Epyc parts in some cases. That might be part of why they didn’t produce it.
Maybe in the workstation arena, but it would be handicapped in the mid-range server arena, even with the best workstation chipset options. Offering a $1500 super duper 5950X3D gamer chip wouldn't be a bad margin to soak up if they can get around 1,000 sales a quarter. You aren't poaching other AMD customers at that price point, you are scalping Intel's top end gaming customers.
 

jamescox

Senior member
Nov 11, 2009
506
822
136
It's probably just that the product is not flawed, it's just very niche. Even I think spending more than $200 for a gaming cpu is stupid but apparently it's a market.
It's not a bad general cpu outside of gaming either and why would he be so hung up about overclocking but not mention amd's old statemend of something like "all ryzen am4 cpus are overclockable" which is now no longer true and that the market hasn't demanded overclocking for consumer cpus since intel locked away their K series.

All of charlie's recent articles (that I can read for free) have nose dived in quality and are petulant rants. It's good that he's calling these things out but it's done in a less professional manner or as impartially from what I would have expected of him from years past.
He seems to think that the lack of overclocking is due to thermal stability; that is, the bond between die can’t take the higher heat. This doesn’t really seem that likely to me. The denser process used for the cache die probably has a lower voltage limit than the high performance process used for the cpu, so the reason given is likely at least partially true. It is also more likely to have reduced life span from electro-migration at higher temperatures and higher voltages. Any kind of stacking does require a lot of engineering due to differing thermal expansion rates, so it is of concern, but anything so new is of concern.

I don’t think they are actually bonded by heating at all as Charlie mentions. It seems to be a cold weld. Metals will weld themselves together in a vacuum. Most things are held together by “just” Van Der Whaals forces; otherwise, everything would be one giant molecule. It would technically be a metallic bond, not just Van Der Whaals forces, although I don’t know if silicon participates in metallic bonds; any chemist around? The copper will cold weld together though. It is also probably surrounded by an epoxy for the packaging such that it cannot move.

The heat issue is that the cache will produce some heat of its own. This raises the temperature above the compute die. Heat transfer is proportional to the temperature difference, so this reduces the heat transfer, even if the thermal conductivity through the material and interfaces is roughly the same. I don’t think the thermal interfaces between the stacked die are an issue. The die are polished down to exceptional flatness. I think it will really behave like one piece of material. The copper TSVs may actually increase thermal conductivity over a device without it. The increased heat / lower thermal transfer could be gotten around by using a TEC, as in the AMD patent I posted about. That had top memory die that would be in more direct contact with the heat sink. The TEC layer would be in between the memory die and the bottom logic die(s), probably just to provide a slightly higher temperature differential to aid in heat transfer.

I think a lot of enthusiast over estimate the number of people that overclock. It was much better long ago when the differentiation between parts was often just clock speed. Look at the early Nehalem based extreme edition parts. It was ridiculously expensive just for higher clock. Now, with boost clocks and thermal limit settings, how often does anyone actually overclock? The boost clocks already are a kind of overclocking.

Also, the top end is always at a very high price to performance ratio. Is it a niche part? Yes, but so are just about any other high end processor. Most people don’t spend over $200 on their processors just like most people don’t have a 3090 gpu. A lot of people have a 3060 and maybe a 3070 though. This is a little more niche in that it mostly only makes sense as an upgrade part. I don’t see it as any more of a marketing part than any other high end part. A 3090 is mostly a marketing part; people see it at the top of the performance charts and all of the reviewer using it and then go buy a 3060.
 

Timorous

Senior member
Oct 27, 2008
825
1,012
136
Why would you include a 16-core Epyc? They are not comparable at all. Even the normal 16-core Epyc has 128 MB L3 (4 CCX * 4 cores). The the 73F3 part has 256 MB (8 CCX * 2 cores). The 7373X is still 8 CCX * 2 cores for 768 MB. You could already get much larger cache per core with Epyc parts. It might be fun to see how the 72F3 (8 CCX * 1 core) with 256 MB for only 8 cores does, but it isn’t very relevant. A Ryzen 9 5950X3D, if it existed, would be 192 MB total. I don’t know if AMD would want such a part in the mix since it may compete with much more expensive Epyc parts in some cases. That might be part of why they didn’t produce it.
The most likely reason AMD did not bother with 5900X and 5950X are as follows

1) 5900X and 5950X are productivity 1st parts and for most desktop usecases the 3d cache has little to no benefit so why would consumers pay extra for no gain in their primary workloads.

2) The concessions required for V-Cache currently mean a 5900X3D and 5950X3D would need lower clocks and better bins than their non 3D counterparts. With point 1 this probably means a regression in productivity performance (outside of a few edge cases) as we see with the 5800X3D vs 5800X.

3) It would use 2x the silicon and it seems unlikely AMD could charge $900 for a 5950X3D with the above caveats.

4) In gaming there is very little to choose between the 5800X, 5900X and 5950X. The same would be true for the 3D version of such parts.

So in conclusion why regress productivity performance on productivity 1st parts just to catch up to intel in gaming (on average) when you can release a gaming 1st part that does the same and is a lot cheaper and uses lower quality bins for the CCD and only 1 cache die leaving more usable parts for Milan-X.

I expect Zen 4 3D to have a different voltage domain for L3 cache vs core and it is possible AMD will stack the core on top of the cache as well rather than the other way meaning cooling is less of an issue. That would mean AMD can probably release parts with fewer caveats but it will still have the same issue of no productivity performance increase but it would mean AMD could release a no compromises all use case part which a 5950X3D would not be.
 

Shamrock

Golden Member
Oct 11, 1999
1,344
400
136

ZGR

Golden Member
Oct 26, 2012
1,920
419
136
I was checking out the 5800X3D owners thread on overclock.net and there are a few people who aren’t seeing 4550 mhz boost on one core. I am not seeing that either, goes up to 4450 mhz.

Anyone else having this issue?
 

ZGR

Golden Member
Oct 26, 2012
1,920
419
136
Also on the Overclock.net 5800X3D owners thread is Curve Optimizer/PBO Tuner 2 linked on post 13:


This tool is really simple, just lets user do some quick voltage offsets. I did this in about 5 minutes, no thinking, just did -20 offset for all cores. Here are my monitored clockspeeds and temps over a cinebench multicore run:

Stock voltage:

cinebenchstock.png

CPU maintaining around 4325 MHz, temps reaching 83C means thermal throttling.

Here is -20 offset for all cores with same test (I did -10 as a trial):

cinebenchCO.png

Temps never hit above 78C, and CPU maintained its max all core boost! I have to get ready for work, but I will try some lower voltages/actual stress tests when I get home. I saw some people with -30 offsets on some of their cores on the thread.
 
Last edited:

Timorous

Senior member
Oct 27, 2008
825
1,012
136
MSI have updated that they are skipping 1.2.0.6c for 300 series boards and going straight to 1.2.0.7. They have a beta out for some X570 boards so hopefully not to long for 300 series but my 5800X3D has been sat on the shelf for a week now and it does not even make a good paperweight.
 

lobz

Golden Member
Feb 10, 2017
1,945
2,593
136
I was more on the safer side when it came to sizing up the possibilities of this CPU before launch.

Am I the only one who thinks that this product ended up being such a massive boost for A LOT of AMD platform owners, whose primary focus is on gaming, requiring zero amount of exotic cooling and more often than not even less power than the CPU it'd replace, from 2000 to 5000 series..., all that at the cost of a BIOS update? And getting equal performance to a brand new, state of the art 12900K gaming rig on average?
If anything, this CPU is not even the worst for Intel... it's a nightmare for mobo, RAM and other system part manufacturers :D

I'm down to saying that AMD could produce a million of these and have them all sold at its original MSRP eventually.
 
Last edited:

biostud

Lifer
Feb 27, 2003
16,072
1,374
126
I was more on the safer side when it came to sizing up the possibilities of this CPU before launch.

Am I the only one who thinks that this product ended up being such a massive boost for A LOT of AMD platform owners, whose primary focus is on gaming, requiring zero amount of exotic cooling and more often than not even less power than the CPU it'd replace, from 2000 to 5000 series..., all that at the cost of a BIOS update? And getting equal performance to a brand new, state of the art 12900K gaming rig on average?
If anything, this CPU is not even the worst for Intel... it's a nightmare for mobo, RAM and other system part manufacturers :D

I'm down to saying that AMD could produce a million of these and have them all sold at its original MSRP eventually.
And hopefully the same will happen with AM5 🙏
 

ASK THE COMMUNITY