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

Page 76 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Kedas

Senior member
Dec 6, 2018
355
339
136
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"
 
Last edited:
  • Like
Reactions: Tlh97 and Gideon

MadRat

Lifer
Oct 14, 1999
11,946
265
126
You only have so many connections in a home. Unless you bought a business connection I doubt you see any difference between 6 and 12 cores in that scenario.
 

ozzy702

Golden Member
Nov 1, 2011
1,151
530
136
In my kid's case, he tends to stream out to his friends, while hosting a game server, while also playing the same games, while havong another monitor filled with discord, youtube, etc. I've seen his task manager. He could use more cores than his 2700x has.

Which games and servers? We talking minecraft? My son is starting to get into the same stuff and his i7-7700 is probably not going to cut it for long.
 

LightningZ71

Golden Member
Mar 10, 2017
1,798
2,156
136
His biggest time sink recently has been "From the depths", which, with it's recent switch to a recent Unity engine, is much friendlier to multi-core processors due to better multi-threading. However, he has certainly played his share of Minecraft, and, of course, indulges in Fortnite with his siblings fairly often. (as an aside, having 5 kids, and watching them all go to town in Fortnite together is just both hilarious and amazing) He plays a lot of other games on steam and in other launchers too. His brother just bought Ace Combat 7 and they've been playing that. He has a long running Factorio challenge with one of his friends going too.

The 2700x actually holds up quite well in most of his games. It pairs well with his 1070ti in his desktop and it always gives him playable framerates at 1080p. He hasn't been using his desktop as much lately though as he recently upgraded his laptop (Christmas!) to a 5900hx/6800m model and, well, it is just faster across the board.
 

Joe NYC

Platinum Member
Jun 26, 2021
2,535
3,466
106

nicalandia

Diamond Member
Jan 10, 2019
3,331
5,282
136

Zucker2k

Golden Member
Feb 15, 2006
1,810
1,159
136

BorisTheBlade82

Senior member
May 1, 2020
680
1,069
136
@BTRY B 529th FA BN
Of course there has not been any official information from AMD. Also Chips and Cheese did not explicitly mention it when they measured its latency.
But seeing that latency only increased a bit I would guess that associativity stayed the same.
 

nicalandia

Diamond Member
Jan 10, 2019
3,331
5,282
136
Sorry if this has been discussed before but will the 5800X3D L3 cache be 16-way or will that change?

Thanks to the most recent Leak(like the one where some mad lad was able to OC 128 Cores of Milan X to 4.8 Ghz on air) we can see that a single 64C/128T Milan-X is using 768 L3$ with 16-Way, as expected as they are trying to make the 3D V Cache to appear as just additional L3$ to the OS


Milan-X with 768 Total L3$(512 From a Single Stack 3D V Cache)
1644346170382.png


Also going from the Zen4 Leaked data(Gigabyte) the L3$ on Zen4 is still 16-way, so any further 3D V Cache on Zen4 will remain the same at 16-way

1644347913236.png

 
Last edited:

Saylick

Diamond Member
Sep 10, 2012
3,531
7,858
136
Someone overclocked dual Milan-X with V-Cache to 7773X to 4.8 GHz on all 128 cores.


WCCFTech didn't include this in their article (I wonder why...) but Videocardz clarified the OC:
The creator is using a special EPYC Milan/Rome ES/QS overclocking tool by Uat4 and ExecutableFix. This tool unlocks power limits and increases maximum all-core frequencies and voltages much higher than AMD intended, but only for engineering samples. However, this is not true overclocking. ExecutableFix explained to us that the frequency reported by CPU-Z software is not a real frequency, but rather a target. This is probably why Kenaide did not provide any benchmark results using a 4.8 GHz OC session, it would have simply shown smaller than expected performance uplift.
 

StinkyPinky

Diamond Member
Jul 6, 2002
6,886
1,103
126
Someone overclocked dual Milan-X with V-Cache to 7773X to 4.8 GHz on all 128 cores.


How is that even physically possible? I was also under the impression the stacked cache ads a bit more heat since the 5800X3D is clocked lower.
 

nicalandia

Diamond Member
Jan 10, 2019
3,331
5,282
136
Someone overclocked dual Milan-X with V-Cache to 7773X to 4.8 GHz on all 128 cores.


Looks like that someone is Chinese content creator 穷三代kenaide科技 and he is the one that purchased the ES/QS sample it was being offered by the Chinese Taobao retail outlet.

This is the screen shot of the original

1644360108361.png
 
  • Like
Reactions: lightmanek

RnR_au

Platinum Member
Jun 6, 2021
2,062
5,006
106
That's . . . interesting. I don't think we're going to learn much more unless there are head-to-head comparisons done between B0 and B2-stepping Vermeer @ static clocks/volts.
This might be coming shortly. Planet3DNow managed to pick up a 5800X B2 from retailer in Germany. Seems to be easier to cool.

However, what is immediately noticeable is that the CPU is far easier to cool than the old version of the 5800X . This model in particular has been tricky so far, since its 105 W TDP or 143 W PPT is concentrated on just one compute die in addition to the IO die. With the 5900X , for example, which is classified in the same TDP class, the heat dissipation is distributed over two compute dies. The 5800X was correspondingly hot-headedso far on the way. A copy I installed in August 2021 could not be tamed with a budget cooler, peak loads immediately caused the die temperature to shoot up to over 90 °C, so I had to install a Scythe Mugen 5 to get below 90 °C CPU temperature to stay. With the B2 stepping, on the other hand, the CPU runs at a peak of 85 °C with Prime95, although only a comparatively cheap Arctic Freezer 34 CO is installed. It's not like he's not hitting his PPT limit. Almost 140 W are also on the clock here:

German article autotranslated - https://www-planet3dnow-de.translat...ch=http&_x_tr_sl=de&_x_tr_tl=en&_x_tr_hl=ende

Will be interesting to see what AMD has done, if anything, to this new B2 stepping.
 

Saylick

Diamond Member
Sep 10, 2012
3,531
7,858
136
This might be coming shortly. Planet3DNow managed to pick up a 5800X B2 from retailer in Germany. Seems to be easier to cool.



German article autotranslated - https://www-planet3dnow-de.translat...ch=http&_x_tr_sl=de&_x_tr_tl=en&_x_tr_hl=ende

Will be interesting to see what AMD has done, if anything, to this new B2 stepping.
Bodes well for V-cache variants then, if all V-cache versions use the B2 stepping. Seems odd now that AMD had to reduce base and boost clocks at all if the B2 stepping is easier to cool. Must be a power thing, or they didn't really want to spend the time validating V-cache with the same clocks so it was just easier to rate it at lower clocks and call it a day. Regardless, it does sound like you can overclock B2 parts easier, with and without V-cache.
 

moinmoin

Diamond Member
Jun 1, 2017
5,064
8,032
136
They'd best use the same motherboard + firmware revision to compare the B2 stepping chip to an existing B0 chip.
Already noted in the article: "Unfortunately I don't have a B0-Stepping 5800X at hand right now, with which one could do direct comparison tests on the same mainboard."
 

Joe NYC

Platinum Member
Jun 26, 2021
2,535
3,466
106
Will be interesting to see what AMD has done, if anything, to this new B2 stepping.

I am going to guess that the most important thing (aside for enabling V-Cache) will be on the server side. Google Cloud and I believe also Azure have been itching to release "Confidential Computing" on AMD Milan platform, but it keeps saying "Coming Soon" for some time.

There was some errata in the SEV instructions in earlier generations, and I would bet those got fixed and will finally allow these Cloud providers to fully enable Confidential Computing into production.

Just today, Google Cloud released a new instance C2D based on Milan, and again, it says "Coming Soon" for Confidential computing. So maybe these companies just need to accumulate enough of the B2 stepping Milan parts, so that when customers turn on the "Confidential Computing" flag, they have enough nodes with Milan B2 to serve them.
 
  • Like
Reactions: Vattila

eek2121

Diamond Member
Aug 2, 2005
3,100
4,398
136
Bodes well for V-cache variants then, if all V-cache versions use the B2 stepping. Seems odd now that AMD had to reduce base and boost clocks at all if the B2 stepping is easier to cool. Must be a power thing, or they didn't really want to spend the time validating V-cache with the same clocks so it was just easier to rate it at lower clocks and call it a day. Regardless, it does sound like you can overclock B2 parts easier, with and without V-cache.

AMD already stated it was a power thing:

AMD CPUs in 2022: Zen 4 in Second Half, Ryzen 7 5800X3D with V-Cache by Spring (anandtech.com)

The frequencies are lower than the regular 5800X? The cache does add a few watts to the power both in terms of idle and load. Rather than bin a stricter chiplet, the decision was made to reduce the frequency a little, but still allow overclocking. The chip, while listed at 105 W, still has the 142 W package power tracking for motherboards that support it.