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]

Few people will pair the R5-5600 or i5-12400 with a high end GPU. But if AMD is successful in portraying the image of the 5800X3D as the world's best gaming CPU, you can bet people will buy it over the 12900K/KS. Less heat/no E-core issues/no RAM tuning/no expensive RAM. So many headaches gone.

 

[nicalandia]

If Reviews are not released with 3D V-Cache Performance Optimizer then they are leaving performance on the table.




Also the TPU Review used a x570 MB which according to some posters(from the TPU Forums) it can't do much BCLK overclocking(capped at 101 mhz)

 

[Makaveli]

If Reviews are not released with 3D V-Cache Performance Optimizer then they are leaving performance on the table.

View attachment 59920


Also the TPU Review used a x570 MB which according to some posters(from the TPU Forums) it can't do much BCLK overclocking(capped at 101 mhz)

There are some good points in that thread and alot of clueless post.

AMD has this locked down for a reason I don't think switching to a B550 is going to make any difference. The only thing I can see is using faster lower latency memory.

This 3d V cache performance optimizer driver is only referenced for a Gigabyte board and no other boards yet. And the mention of a newer agesa won't matter until that is out.

 

[nicalandia]

This 3d V cache performance optimizer driver is only referenced for a Gigabyte board and no other boards yet.

It Would seem that others are implementing some form of Performance Enhancer for the 5800X3D on their Bios. The MSI does not mention any Performance Optimizer..

 

[Makaveli]

It Would seem that others are implementing some form of Performance Enhancer for the 5800X3D on their Bios. The MSI does not mention any Performance Optimizer..


View attachment 59927

This bios is the same one asus has out for my board

PRIME X570-PRO BIOS 4204
1. Update AMD AM4 AGESA V2 PI 1.2.0.6b

2. Improve system performance and stability
3. Improve system performance for AMD Ryzen 7 5800X3D

The bios tpu used

BIOS 7C37vAG2, AGESA 1.2.0.6c (5800X3D)

which is going to be pretty much the same thing.

The only reference to the next AGESA V2 P1 1.2.0.7 and it giving more performance than the bios above is from a tweet and nothing offical from AMD so even that has to be taken with a grain of salt right now.

 

[nicalandia]

The bios tpu used

BIOS 7C37vAG2, AGESA 1.2.0.6c (5800X3D)

This early tester says to stay away from that

PSA: Stay away from AGESA 1.2.0.6c, AMD restricting VCORE to 1.2v with BIOS PBO EDC above 140

Thought it was bad enough AGESA prior started capping VCORE at 1.425v if you ran above 140? How about 1.2v [https://i.imgur.com/Zh2iZk6.png](https://i.imgur.com/Zh2iZk6.png) I've ran a CB23 above, booted a game for a few minutes, opened/closed web browser and so on. Replicated this across...

www.fpshub.com

 

[Makaveli]

This early tester says to stay away from that

PSA: Stay away from AGESA 1.2.0.6c, AMD restricting VCORE to 1.2v with BIOS PBO EDC above 140

Thought it was bad enough AGESA prior started capping VCORE at 1.425v if you ran above 140? How about 1.2v [https://i.imgur.com/Zh2iZk6.png](https://i.imgur.com/Zh2iZk6.png) I've ran a CB23 above, booted a game for a few minutes, opened/closed web browser and so on. Replicated this across...

www.fpshub.com

This seems to be more of an issue for the non 3d chips as the new chip won't go over 1.35v and there is no OC,PBO etc on it.

 

[GodisanAtheist]

Best thing about this release is the apparent firesale of 5800x parts I'm seeing on used sites. Might be a great chance to pick up a dirt cheap used 5800x + a B550 board + carry over my DDR4 ram from current system: Huge core system upgrade for ~$350 bucks.

 

[Hans Gruber]

The only reason the 5800x3D will not OC or why the core speeds are low. The v-cache has a max voltage limit and somehow the CPU voltage directly correlates with the voltage of the v-cache. Otherwise the core speeds would be 300-500mhz higher.

 

[epsilon84]

Is the 5800X3D underperforming in the TPU review?

AMD Ryzen 7 5800X3D Review - The Magic of 3D V-Cache

The AMD Ryzen 7 5800X3D is the company's new flagship gaming processor. It introduces 3D V-Cache, a dedicated piece of silicon with additional L3 capacity. In our review, we're testing how much the larger cache can help intensive gaming workloads and applications and compare it to the Intel Core...

www.techpowerup.com

At 720P its ~10% faster than Zen 3, at 1080P that shrinks to ~7.5%, literally half of what AMD claimed the v-cache gains to be for 1080P. Granted, TPU is 'only' using an RTX 3080, you may see slightly better scaling with a 3090 Ti.

 

[epsilon84]

Few people will pair the R5-5600 or i5-12400 with a high end GPU. But if AMD is successful in portraying the image of the 5800X3D as the world's best gaming CPU, you can bet people will buy it over the 12900K/KS. Less heat/no E-core issues/no RAM tuning/no expensive RAM. So many headaches gone.

To be fair, the crowd that AMD is trying to target this at out aren't in it for the 'bang for buck'. That's not what the 'halo gaming' market is about. Think 3090 Ti owners here. They just spent $1500 on their GPU to gain 5% over a vanilla 3090, so spending a bit more (or less) on a CPU doesn't really matter that much as long as it is the 'fastest'. Trust me, I've seen enough 3090 Ti owners out there to see that most run 12900K or 5950Xs, CPU pricing isn't a concern for them. I do wonder if most 12900K/5950X owners would be tempted to 'sidegrade' to a 5800X3D just to get a few more fps though? Personally, I wouldn't, but I'm not just a gamer and use my PC for work as well, so I'll take 'moar corez' thanks

Also, I don't wish to derail this thread, but as an 12900K owner those 'headaches' with the 12900K aren't really true - for gaming, heat is a non issue, and I actually prefer having the E-cores to be available for background tasks while I game. RAM tuning? I think most people just set XMP and that's it (until recently I was in that camp too, the gains are small even with tuned subtimings). You may be exaggerating for effect, but honestly, I encountered none of those issues with my 12900K WR to gaming.

FWIW, DDR4 can still be expensive if you go for the high speed / low latency stuff. For my 12900K, I went with Trident Z Neo 'B-Die' 3600 C16 which was approaching the price of DDR5 kits (albeit entry level), and my DDR4 kit isn't even considered truly premium like those DDR4-4000 kits. Yes, when comparing fast DDR5 vs fast DDR4, DDR5 is obviously more expensive still, but I do see the prices trending down slowly for DDR5.

 

[gdansk]

I don't think it is under performing. It ends up around the 12900K. That's about what was expected.

It's just much later than we had hoped. Still a money saver for folks with previous AM4 boards.

 

[epsilon84]

I don't think it is under performing. It ends up around the 12900K. That's about what was expected.

It's just much later than we had hoped. Still a money saver for folks with previous AM4 boards.

I wasn't really comparing it to the 12900K, more to its Zen 3 counter-parts. AMD claimed on average a 15% increase ast 1080P over a 5900X, which I thought was a reasonable and realistic claim. The TPU results show a ~7.5% increase, which is actually a fair bit lower than what I expected, and as I said earlier, literally half of what AMD's estimated gains to be.

The fact that it can't truly beat the 12900K (according to TPU anyway) also means it is somewhat 'underperforming' since that is the whole point of the CPU - for AMD to regain the gaming performance crown. Not to draw, but to outright beat the 12900K.

I think I'll wait for more reviews with bigger game sample sizes before calling AMD out here for their 15% claim - I still think it's possible, but for all we know TPU might have just chosen a bunch of games that don't necessarily scale that well with larger caches.

 

[andermans]

I wasn't really comparing it to the 12900K, more to its Zen 3 counter-parts. AMD claimed on average a 15% increase ast 1080P over a 5900X, which I thought was a reasonable and realistic claim. The TPU results show a ~7.5% increase, which is actually a fair bit lower than what I expected, and as I said earlier, literally half of what AMD's estimated gains to be.

The fact that it can't truly beat the 12900K (according to TPU anyway) also means it is somewhat 'underperforming' since that is the whole point of the CPU - for AMD to regain the gaming performance crown. Not to draw, but to outright beat the 12900K.

I think I'll wait for more reviews with bigger game sample sizes before calling AMD out here for their 15% claim - I still think it's possible, but for all we know TPU might have just chosen a bunch of games that don't necessarily scale that well with larger caches.

I think that is going to be a clear problem this review cycle, there is such variance in how much games benefit from the cache. I wouldn't be surprised if one could get 5-20% vs. plain zen3 depending on the games that would be included in the average. We've clearly seen multiple 30%+ improvements.

 

[ondma]

To be fair, the crowd that AMD is trying to target this at out aren't in it for the 'bang for buck'. That's not what the 'halo gaming' market is about. Think 3090 Ti owners here. They just spent $1500 on their GPU to gain 5% over a vanilla 3090, so spending a bit more (or less) on a CPU doesn't really matter that much as long as it is the 'fastest'. Trust me, I've seen enough 3090 Ti owners out there to see that most run 12900K or 5950Xs, CPU pricing isn't a concern for them. I do wonder if most 12900K/5950X owners would be tempted to 'sidegrade' to a 5800X3D just to get a few more fps though? Personally, I wouldn't, but I'm not just a gamer and use my PC for work as well, so I'll take 'moar corez' thanks

Also, I don't wish to derail this thread, but as an 12900K owner those 'headaches' with the 12900K aren't really true - for gaming, heat is a non issue, and I actually prefer having the E-cores to be available for background tasks while I game. RAM tuning? I think most people just set XMP and that's it (until recently I was in that camp too, the gains are small even with tuned subtimings). You may be exaggerating for effect, but honestly, I encountered none of those issues with my 12900K WR to gaming.

FWIW, DDR4 can still be expensive if you go for the high speed / low latency stuff. For my 12900K, I went with Trident Z Neo 'B-Die' 3600 C16 which was approaching the price of DDR5 kits (albeit entry level), and my DDR4 kit isn't even considered truly premium like those DDR4-4000 kits. Yes, when comparing fast DDR5 vs fast DDR4, DDR5 is obviously more expensive still, but I do see the prices trending down slowly for DDR5.

Yea, but if price is no object, why not just wait for Zen 4. Get a new platform, faster I/O, DDR5, and unless the gains are far less than expectations, you could probably get gaming equal to or better than 5800X3D without the limitation of 8 cores.

 

[epsilon84]

Yea, but if price is no object, why not just wait for Zen 4. Get a new platform, faster I/O, DDR5, and unless the gains are far less than expectations, you could probably get gaming equal to or better than 5800X3D without the limitation of 8 cores.

Yeah, I'd agree with that. Zen 4 is close enough that it makes sense to wait, plus DDR5 should (hopefully) be ready for prime time then as well.

5800X3D is a nice last hurrah for AM4, but in grand scheme of things it will be a mere footnote for AMD historians in the future.

 

[deasd]

Sounds like it's still could be OC though if you're lucky:

https://twitter.com/x/status/1514063342440116229

 

[tamz_msc]

He said the 12900KS would be faster than 5800X3D with the exception of edge cases like SCII. Then he proceeded to make a point by show what we can probably consider an outlier by now: a CP2077 benchmark where ADL-S was up to 50% faster.

The argument was meant to be disarming, that's why I'm asking the question now, after getting proper DDR5 benchmarks in a variety of games. (including CP2077, obviously a different run than the one presented in the other video).

As much as you and others would like the CP2077 result to be an outlier, it really isn't. That's the norm in many RT heavy games when you enable raytracing. It's just another aspect that reviewers are oblivious to.

Anyway, if you don't like CP2077 because its an 'outlier', then feel free to look at Troy: A Total War Saga:



Then there's everybody's favourite Hardware Unboxed/Techspot, where they test The Riftbreaker:



Are these 'outliers' too?

It's too bad that Hardware Unboxed is already trying to paint the 5800X3D in a favourable light in their upcoming review by testing with DDR4-3200:



In the quest of testing which is the fastest gaming CPU, as per AMD's claims, crippling your competition only reinforces the prevalent notions some people have with their channel.

 

[gdansk]

What's the point? There's games where Golden Cove's ability to exploit greater instruction level parallelism will knock it out of the park. There's games where the 5800X3D's extra cache will knock it out of the park. In aggregate they seem rather equal. If somehow you know which games you'll play in the future then find specific benchmarks and buy based on that.

 

[tamz_msc]

What's the point?

The point is to test marketing claims. Isn't that what tech journalists are supposed to do?

 

[uzzi38]

It's too bad that Hardware Unboxed is already trying to paint the 5800X3D in a favourable light in their upcoming review by testing with DDR4-3200:

View attachment 59941

In the quest of testing which is the fastest gaming CPU, as per AMD's claims, crippling your competition only reinforces the prevalent notions some people have with their channel.

Yeah, that's a crock of BS. He said on Twitter that he actually leaned the other way.

https://twitter.com/x/status/1513655190821343233

He asked the question to see what the audience wanted, that's all. If he was trying to do what you're claiming, he wouldn't have said which way he leaned at all. Doing so inevitably ends up skewing the results.

 

[gdansk]

The point is to test marketing claims. Isn't that what tech journalists are supposed to do?

Not sure. I was asking what's the point of such "outlier" discussion. As the TPU benches show there are also scenarios where Zen3D out performs GC even when Alder Lake is equipped with memory more than twice as expensive. There's bound to be plenty of examples on each side. Soon you'll be able to throw selected graphs at each other all day but why bother. Consider a more complete picture.

 

[Insert_Nickname]

The Bios he is using is without the 3D V-Cache Optimizer Driver and unoptimized AGESA 1.2.0.6c

I'm wondering how much difference that driver will make. Here is to hoping someone tests with/without. There is bound to be some benefits from optimisation for such a large L3.

 

[coercitiv]

Meanwhile Techpowerup changed their 720p aggregated score. I'll refrain comments as I'm already a "Schrödinger's fan": both AMD and Intel biased depending on who reads this post.

As much as you and others would like

What annoys me the most on this forum is people telling me what I think. That must be the meta in terms of "forumsplaining". If you're so worried about my opinion on ADL-S vs 5800X3D why don't you just ask me? I made my prediction (and bet) long ago when I upgraded to 12700K.

Meanwhile when I ask you to analyze your claim of significant performance uplift using DDR5, you churn out a bunch of examples where ADL-S /w DDR4 already has a big lead. It should be clear to anyone in the forum that these results are not the product of DDR5 as much as they are the product of a stronger architecture. And if that's the case, then they should be counted against the other cases where 5800X3D wins by a landslide. As @gdansk already put it very nicely, consider a more complete picture.

As for your reaction towards HUB, i find it perplexing that you accuse them of something they MIGHT do while also posting DDR5 6400 CL36 results from their own testing. How the heck are they crippling the competition when they already made extensive testing with fast DDR5 on 12900KS?! It boggles the mind, really.

 

[lobz]

i2Hard tests with RT enabled. Unless you enable RT, CP2077 is mostly GPU-bound.

🤦🤦🤦🤦🤦