5800X3D - A good send off for AM4?

[AGTDenton]

A number of reviews are now published for the 5800X3D


Club 386

PC Gamer

Techspot

I personally can't see the attraction based on the scores, but this 3D cache seems to be a taste of things to come with AM5

[CAT-THE-FIFTH]

I made a post on OcUK forums that in some games the improvements were very large:
https://forums.overclockers.co.uk/posts/35559674/

[[GSV]Trig]

Same price, on paper at least, as the 5800X, wonder if there is more to come or if this is just testing the waters for AM5 as it seems odd to have this 3d cache but then slower speeds than the stock 5800x, also be interested to see how hot it runs given the 5800x isnt exactly a 'cool' chip..

[Saracen999]

.... as it seems odd to have this 3d cache but then slower speeds than the stock 5800x, also be interested to see how hot it runs given the 5800x isnt exactly a 'cool' chip..

Yeah, but look at the power draw on equivalent Intel 12th Gen.

The explanation I saw somewhere (Gamers Nexus, I think) was that by slapping the extra L3 cache on-board, you get the speed boost from that but, as it somewhat insulates the CPU itself from cooling, they had to drop the base speed a bit too. The inference seemed to be that the 5800X etc were tuned/tweaked pretty close to the limit, hence presumably, the entire new platform. So final performance depends very much on what you're doing, and perhaps more importantly, how whichever game you're playing works, because that can mean significant cains from the extra L3, on even slowing down from lower base speed. It makes choosing a bit bit tricky, don't it?

[AGTDenton]

I made a post on OcUK forums that in some games the improvements were very large:
https://forums.overclockers.co.uk/posts/35559674/

That's quite a post!

[CAT-THE-FIFTH]

That's quite a post!

It seems any applications which are Cache bound can show massive improvements in performance.

[AGTDenton]

It seems any applications which are Cache bound can show massive improvements in performance.

Yeah it's quite an interesting release, a bit of weighing up based on what you plan on doing.

One thing's for certain, AMD are not asleep

[DanceswithUnix]

It seems any applications which are Cache bound can show massive improvements in performance.

Cache bound???

If an application is limited by memory speed, then it likely still will be and there won't be much improvement.
If an application fitted in cache already, then the extra cache is close to useless and there won't be any improvement.

But if an application didn't quite fit entirely in cache before, then now it will and an entire level of memory access goes away. That's a huge jump, but for a small subset of programs.

My biggest take away from these reviews has been that the 5600 looks like quite nice value and a worthy successor to the 3600 that I used to buy

[Kovoet]

I will be getting this.

[CAT-THE-FIFTH]

Cache bound???
....

When I talk about it being cache bound I mean limited by cache size(as in upper boundary).

One of the applications I linked to is Lightroom as I use it on and off,and the other is DxO Photolab which I use too. It's why I linked to both. We have probably had this discussion before when the Zen2 reviews dropped.

Photolab does not seem to care too much about the size of Zen caches because Zen3 and Zen3 APUs show similar performance. Lightroom does,and shows significant performance degradation between the CPUs and APUs. It also didn't show huge improvements between Zen2 and Zen3 when it came to export performance,and any differences seem more akin to clockspeed differences.

But it shows a large improvement when moving to 192MB of L3 cache. Lightroom can use more threads and tends to traditionally prefer Intel CPUs(but Adobe CBA to properly support it on AMD 16C CPUs,as the performance is bugged,so even Zen2/Zen3 performance isn't optimal).

Yet an 8C/16T Ryzen 7 5800X3D is the fastest pure 8C CPU ever made for Lightroom exports.

Likewise,it also showed a massive improvement in performance with Zen2 CPUs(and their much large L3 caches) over Zen/Zen+ to the extent you saw nearly a doubling of performance over a Ryzen 7 2700X in exports,and it thrashed even a Core i9 9900K. You also saw the same with HEDT Intel CPUs(larger L3 caches) showing the same relative to the equivalent consumer CPUs. Its not even a memory bandwidth thing either looking at earlier results and some other comparisons.

My biggest take away from these reviews has been that the 5600 looks like quite nice value and a worthy successor to the 3600 that I used to buy

Not as good value as my £185 Ryzen 7 5700X! Though I still think a Core i5 12400F is better value as a CPU - even seen it dip below £150 a few times,even £100ish one! Having said that Intel dropped the ball on the motherboard side!

[badass]

Yeah, but look at the power draw on equivalent Intel 12th Gen.

The explanation I saw somewhere (Gamers Nexus, I think) was that by slapping the extra L3 cache on-board, you get the speed boost from that but, as it somewhat insulates the CPU itself from cooling, they had to drop the base speed a bit too. The inference seemed to be that the 5800X etc were tuned/tweaked pretty close to the limit, hence presumably, the entire new platform. So final performance depends very much on what you're doing, and perhaps more importantly, how whichever game you're playing works, because that can mean significant cains from the extra L3, on even slowing down from lower base speed. It makes choosing a bit bit tricky, don't it?

I wonder if they are looking into placing the SRAM die and shim down first then the CPU die on top - effectively where it was before without the shim that affects thermal transfer. I assume AMD have already looked into it and it's either way more complex and needs more development or is simply not viable with current tech.

[DanceswithUnix]

When I talk about it being cache bound I mean limited by cache size(as in upper boundary).

Thinking about it like that lets us programmers off the hook, so I don't think that's a good label

A program has a "working set" size in memory. As a programmer, you have quite a bit of control over that if you know what you are doing, so you should try to tile your workload into the cache size of your existing processor.

It is interesting that AMD don't seem to be releasing a 5950X3D as that would have increased cache from 96MB of the 5800X3D to 192MB. But then the 5950X already had 64MB of L3, so the 5800X3D is only 50% bigger than that.

Not as good value as my £185 Ryzen 7 5700X!

Well yeah, that's pretty hard to beat. Would be tempted to get one myself at that price, but they seem to be generally £100 more than that

[Rob_B]

If I had the budget I'd go for one, but using an A320 board I'd need to update that too so then we come onto not such a simple 1 part upgrade (although I'm sure it would probably work on the A320) I'm not as blown away as some hype had suggested but it trades blows with chips costing significantly more so I guess it's a good sign of a possible future AM5 option.

I'm thinking of upgrading my rig by picking up a cheap 3xxx chip as a drop in upgrade as those prices will drop due to the new 5xxx releases (or splurge and grab a 5600) Although as a gaming only machine with a 1070 I think I'm GPU bound even with a 2600X so maybe it's not such a bright idea

...Not as good value as my £185 Ryzen 7 5700X! ...

Gutted I missed the misprice, that would have been a no-brainer that would keep me right for a good few years to come.

[DanceswithUnix]

I wonder if they are looking into placing the SRAM die and shim down first then the CPU die on top

That would mean routing all the power and signal lines for the CPU through the 3D cache dies to get the into the carrier and out onto the pins. Doesn't sound like a flyer to me.