zen 4 AMD Ryzen 7000 Zen4 “Raphael” Desktop CPU Review Roundup: Ryzen 5 7600X, Ryzen 7 7700X, Ryzen 9 7900X & Ryzen 9 7950X Thoroughly Benchmarked

[porina]

8 hours ago, BiG StroOnZ said:

Ryzen 7 7700X DDR5 memory scaling review:

To clarify my earlier statement I was thinking mainly in bandwidth terms, not latency. With reasonably optimised kits fast or slow, the latency time (as opposed to cycles) kinda balances out, leaving bandwidth as the main variable. Their testing had relatively high latency slower configurations going against higher speed lower latency, so we can't really separate the contribution from that.

 

7 hours ago, leadeater said:

It's the exact same situation as with AM4. There are non-workstation boards that have ECC support on Ryzen CPUs, not Ryzen Pro. It's actually always been the situation and ECC support has actually always been a platform thing, kinda not a CPU thing unless someone does an Intel and microcode locks it out.

 

Motherboard vendors have to actually allow it to be functional otherwise it's off.

 

Basically nothing has changed at all.

At a practical level I guess I agree with you. I might be arguing over a detail here, but AMD was reported to state that although ECC functionality was present (not disabled) in their consumer AM4 CPUs, the functionality was unsupported. I'm not hearing exactly the same here for AM5. There may be some losses in detail from what I think was said by Ian, and what he heard from AMD. AMD doesn't seem to be outright saying it isn't supported by them this time around. If I needed system level ECC I'd still be cautious about using this vs. platforms that essentially have it available by default and known to be fully supported end to end. 

[Shzzit]

47 minutes ago, Stahlmann said:

Even though it's expensive for a 600-series CPU you could still argue it's a better gaming CPU than Intel's best. People need to realize that CPU names don't matter just like they don't matter with GPUs (see 3080 12G). What actually matters is performance requirements and budget.

Lol?  All the reviews show even the 7950x loosing to intel and last gen ryzen CPUs in a lot of games.  I don’t think new ryzen beats 5800 3d in any games.  All while at 95c.  We might need to wait for the 3d variant of this gen. 

[leadeater]

9 minutes ago, porina said:

At a practical level I guess I agree with you. I might be arguing over a detail here, but AMD was reported to state that although ECC functionality was present (not disabled) in their consumer AM4 CPUs, the functionality was unsupported. I'm not hearing exactly the same here for AM5. There may be some losses in detail from what I think was said by Ian, and what he heard from AMD. AMD doesn't seem to be outright saying it isn't supported by them this time around. If I needed system level ECC I'd still be cautious about using this vs. platforms that essentially have it available by default and known to be fully supported end to end. 

A lot of confusion happens when people hear or use "unsupported" or "not validated" etc. Something unsupported can be functional and work, thus the problem with the word "support" because it can actually mean different things depending on how it's used and the context. Ryzen 7000 supports AVX512 and Ryzen 5000 does not support AVX512, however neither Ryzen 7000 or Ryzen 5000 are supported for ECC either, only Ryzen Pro is. Support/supported being used very differently in these two situations. Ah how wonderfully confusing.

[Shzzit]

35 minutes ago, leadeater said:

Really depends on what you are doing. If you are playing games then you will never have all cores at 100% maximum power so you can use any current cool good enough for AM4/Ryzen 5000 CPUs and get the full performance of a Ryzen 7000 CPU.

 

Even if you become temperature bound for performance the loss is actually going to be very minimal, the performance scaling is really quite bad at the upper end of the scale.

 

Realistically it's not something I would worry about, all current cooler recommendations apply and you should not change your thinking on this unless you have a really good reason to. One reason would be knowing you will be regularly running sustained all core high power workloads and don't want to be temperature performance limited by that.

 

Honestly I don't think it's that big of a deal at all, just like it isn't on GPUs. Weird times ahead, going to be very interesting.

That makes sense, I keep having to remind myself that synthetic benchmark temps

and wattage is not even close to gaming.  Guess I’ll be waiting for the new 3d option or see how intels new CPUs  do gaming.  
 

Intel is probably gonna do the same thing with ghz scaling to temps like gpus? 
 

Exiting times right now, new CPUs and gpus all with in months of each other. 

[leadeater]

1 minute ago, Shzzit said:

Guess I’ll be waiting for the new 3d option or see how intels new CPUs  do gaming.  

Think I will do the same, 5800X3D was just too big of an improvement for me to go sinking money in to Ryzen 7000 knowing that is coming and how bad current pricing is for everything else.

[porina]

3 minutes ago, leadeater said:

A lot of confusion happens when people hear or use "unsupported" or "not validated" etc. Something unsupported can be functional and work, thus the problem with the word "support" because it can actually mean different things depending on how it's used and the context. Ryzen 7000 supports AVX512 and Ryzen 5000 does not support AVX512, however neither Ryzen 7000 or Ryzen 5000 are supported for ECC either, only Ryzen Pro is. Support/supported being used very differently in these two situations. Ah how wonderfully confusing.

100% agree with that. We need different terms for the two cases:

1, company support - if it doesn't work as expected, they'll help you fix it

2, feature support - it has it or it doesn't

[Stahlmann]

6 minutes ago, Shzzit said:

Lol?  All the reviews show even the 7950x loosing to intel and last gen ryzen CPUs in a lot of games.  I don’t think new ryzen beats 5800 3d in any games.  All while at 95c.  We might need to wait for the 3d variant of this gen. 

Yeah ok i misremembered it a bit, still it's punching way above it's weight. And again, the high temperature doesn't matter when the architecture was designed with it in mind. You're already hanging yourself up about the temperature as if it's a defect.

 

[porina]

Has anyone seen Aida64 mem bandwidth results on 7600X or 7700X? I have a theory I'd like to test.

[leadeater]

5 minutes ago, porina said:

Has anyone seen Aida64 mem bandwidth results on 7600X or 7700X? I have a theory I'd like to test.

https://www.thefpsreview.com/2022/09/26/amd-ryzen-5-7600x-cpu-review/4/

 

 

 

 

Edit:

Also https://overclocking.com/test-ryzen-r9-7950x-r9-7900x-r7-7700x-et-r5-7600x/8/

[porina]

4 minutes ago, porina said:

Has anyone seen Aida64 mem bandwidth results on 7600X or 7700X? I have a theory I'd like to test.

Thanks @leadeaterfor the link, I found some others as well. It answers one question but opens another!

 

On AM4 we know that, running synchronously, IF BW scales with ram BW. For a single CCD CPU, reads are matched, writes are half rate. I wondered if this still applied to Zen 4. I'm seeing just under 60 GB/s reads, which could be achieved by an AM4 IF clock of 1900 MHz or so. Two CCD AM5 models ball park 75 GB/s, so that's got to be outside the comfort zone of a single IF link.

 

The new weirdness is that single CCD Zen 4 is scoring ~80GB/s writes! Never mind the half speed writes of AM4, this just sounds too high for what understanding I have right now of how IF is arranged in Zen 4. AMD haven't told us everything yet!

[Shzzit]

21 minutes ago, Stahlmann said:

Yeah ok i misremembered it a bit, still it's punching way above it's weight. And again, the high temperature doesn't matter when the architecture was designed with it in mind. You're already hanging yourself up about the temperature as if it's a defect.

 

Oh I know, but it’s like normal nature to see those temps and think OMG.  Guess I’ll have to start deprogramming my brain lol. 

[Dogzilla07]

22 minutes ago, Shzzit said:

Oh I know, but it’s like normal nature to see those temps and think OMG.  Guess I’ll have to start deprogramming my brain lol. 

Der8auer and GamersNexus mentioned there's barely 50-100Mhz throttling per 25c gains. Outside specific small form factor scenarios, and professional work taxing all cores, there's hardly any difference between mediocre cheap-ish priced air tower cooler and top of the line, most expensive AIO.

 

Will require tinkering to get noise under control, fixed fan speeds, or custom curves with ramp-up > 95c that never happens. Or if u sacrifice some performance ECO mode, fixed clocks, undervolting, etc, ...

 

Gaming is not the same as synthetic/handbrake yet, so no power usage insanity. 7600x is a really good option for Gaming, for games that aren't memory bound, however as mentioned above in the thread, no reason not to get x3D chips instead when they come. Not having to worry about memory bottle-necks is an incredible boon.

[Spotty]

Has there been any discussion yet about the possibility of lower end 7000 series Ryzen CPUs, like a 7100X, 7200X, 7400X? Something to compete with Intel's i3 CPUs like the 12100F(/13100?) and 12400(/13400?)?

With the chiplets being 6c/12t design and the 7600X already being 6c/12t I'm not sure if AMD is going to lock them down further for lower end CPUs to compete with Intel's budget CPUs. Maybe lower clock speeds or less cache?

[Dogzilla07]

9 minutes ago, Spotty said:

Has there been any discussion yet about the possibility of lower end 7000 series Ryzen CPUs, like a 7100X, 7200X, 7400X? Something to compete with Intel's i3 CPUs like the 12100F(/13100?) and 12400(/13400?)?

Nope, AMD is pulling an Nvidia here. They're waiting for Intel to show their cards, and decide what to do. If Intel doesn't pressure them, it's gonna take ~22 months again for 5600/x to drop to 150 eur. If Intel does pressure them it's gonna be ~8-12 months to ~150 eur like for ryzen 2600/3600, or sooner.

 

As for the ~100 eur price range, Lisa hinted AMD will not compete no matter what with newest chiplets. Now maybe AMD's bluffing as well, but unlikely (They might do a monolithic APU again or something).

 

What intel has been saying doesn't look promising, but maybe Intel was bluffing. not only is 13100 looking to be carbon copy 12100. 13400, is looking to be just different composition 12600. + Intel has been signaling they have no choice but to raise prices after Q3 2022.

[Dabombinable]

Since I'm not even in a position where my 3700X is the bottleneck, I doubt I'll ever own anything from AM5. Having up to 16 cores and maybe said cores with vcache - no reason to change platform. Considering that over a year ago my 4790K was only just becoming a bottleneck (single and multi threaded).

[porina]

13 minutes ago, Spotty said:

Has there been any discussion yet about the possibility of lower end 7000 series Ryzen CPUs, like a 7100X, 7200X, 7400X? Something to compete with Intel's i3 CPUs like the 12100F(/13100?) and 12400(/13400?)?

I'd have to guess that'll be APU territory. I'd even go further and wonder how long they keep offering 6 and 8 core desktop chiplet models, with only 3D models making any sense there. I feel the main reason is the APU is far enough off they can't afford to wait for it to fill in the mid range.

 

13 minutes ago, Spotty said:

With the chiplets being 6c/12t design and the 7600X already being 6c/12t I'm not sure if AMD is going to lock them down further for lower end CPUs to compete with Intel's budget CPUs. Maybe lower clock speeds or less cache?

6/12??? They're still 8/16. You did make me do a double take there, since if it was true then 7700X would be 2 CCD and might explain some of the memory bandwidth results seen.

 

9 minutes ago, Dogzilla07 said:

not only is 13100 looking to be carbon copy 12100. 13400, is looking to be just different composition 12600

If memory serves me correctly, the lower end of 13th gen will still be Alder Lake. Raptor Lake is only going into the higher end.

[leadeater]

16 minutes ago, Dogzilla07 said:

As for the ~100 eur price range, Lisa hinted AMD will not compete no matter what with newest chiplets. Now maybe AMD's bluffing as well, but unlikely (They might do a monolithic APU again or something).

 

What intel has been saying doesn't look promising, but maybe Intel was bluffing. not only is 13100 looking to be carbon copy 12100. 13400, is looking to be just different composition 12600. + Intel has been signaling they have no choice but to raise prices after Q3 2022.

I suspect AMD has made a strategic move to service lower end, or rather lower market price points, with APUs only. I don't think the cost complexity of chiplets scales down to the low cost parts compared to monolithic.

 

The high end Ryzen CPUs now having iGPUs might factor in to this somehow, not exactly clear to me in what way though. Have mixed thoughts around that in terms of AMD going APUs only for low cost parts.

 

5 minutes ago, porina said:

6/12??? They're still 8/16. You did make me do a double take there, since if it was true then 7700X would be 2 CCD and might explain some of the memory bandwidth results seen.

haha yea me too. Was pretty sure there was no memo that CCDs were now 6 cores. Would be nice though, 6 core CCDs, 3 CCDs, 1 IOD with 4 channel memory, if only.

[porina]

Still watching Ian's stream replay. He just showed an AMD slide showing the IF connectivity, and it is 32B/cycle reads, 16B/cycle writes. I think this is same as AM4. So where is that Aida64 mem write bandwidth coming from??? Either I'm still missing something, or could the measurement be wrong in some way?

 

 

I got some Prime95 benchmark results for the 7950X from elsewhere, and I'll try using that as a real world workload to see what's happening.

 

[Dogzilla07]

43 minutes ago, porina said:

If memory serves me correctly, the lower end of 13th gen will still be Alder Lake. Raptor Lake is only going into the higher end.

Yup, this is quite disastrous. as it will not make AMD feel threatened.

 

36 minutes ago, leadeater said:

I suspect AMD has made a strategic move to service lower end, or rather lower market price points, with APUs only. I don't think the cost complexity of chiplets scales down to the low cost parts compared to monolithic.

Pretty much looks like that, yeah, the ball is firmly in Intel's court right now (one last time, before mainstream monolithic slides into history). it's going to be interesting how Intel approaches the lower end after they switch to chiplets, and whether they also starting doing monolithic APUs like AMD does.

[Spotty]

44 minutes ago, porina said:

6/12??? They're still 8/16. You did make me do a double take there, since if it was true then 7700X would be 2 CCD and might explain some of the memory bandwidth results seen.

Oops! I honestly didn't pay that much attention to the number of cores the CPUs had this generation.

[porina]

I've got Prime95 benchmark results from elsewhere for 7950X, and am comparing it to my 11700k and 7920X. This is interesting as, depending on FFT size, it could indicate efficiency of AVX implementation as well as cache/memory performance.

 

1K FFT

n cores n workers (1 task per core, scales well)

 

0.86/core - 13.7M 16 cores 7950X
0.74/core - 5.9M 8 cores 11700k - peak power observed 195W
0.47/core - 5.6M 12 cores 7920X - this defaults to power 140W limit. I can't observe actual power

 

Higher is better. This is not clock normalised so it is NOT IPC. I'd expect Zen 4 to be similar or better to Rocket Lake, and that seems to be the case here. The Skylake-X result is much lower than expected. I don't have an explanation for this, but my guess is that at a 1k FFT size, it isn't big enough to load the extra execution resource. I have requested tests at bigger FFT sizes that should load it better without being memory limited.

 

 

8064k

n cores 1 worker - 1 task using all cores. The expectation here is that the large FFT size pushes this mostly to memory limiting, but maybe not! More on that later.

 

318 7950X - dual channel 4400 
258 7920X - quad channel 3000 (2R per channel)
95 11700k - dual channel 2166 (4R per channel)

 

Higher is better. If this was pure memory bandwidth limiting, it should scale with ram. We're not seeing that, with the 7950X punching a good amount above the 7920X. Let's ignore the poor dual channel 11700k system.

 

So what could be happening here? The thing is, 7950X has 64MB in total of L3 cache, which appears enough to hold the workload's effective 63MB of data. In my own testing on Zen 2, and with assistance of others, testing on Zen 3, the split CCX nature gets in the way so scaling is less than ideal. However it might not be nothing, so that might give it the edge in this scenario. We need a workload that pushes it harder and... I have that data. I'll either edit it in later or post a new reply once I've compiled that data. 3D cache version would be really sweet for this use case.

 

Edit:

 

8064k n cores n workers (1 task per core)

 

180 7950X
250 7920X
94 11700k

 

Ok, now we're definitely pushing it hard into ram bandwidth limiting, and the 7950X drops below the 7920X as expected. Skylake-X has 39% more potential bandwidth, and scores 36% higher, so this seems about right. So regardless of Aida64 results, overall performance seems to be scaling within expectations.

[ZetZet]

3 hours ago, Stahlmann said:

Even though it's expensive for a 600-series CPU you could still argue it's a better gaming CPU than Intel's best. People need to realize that CPU names don't matter just like they don't matter with GPUs (see 3080 12G). What actually matters is performance requirements and budget.

7600X + platform is extremely expensive. 12600K is cheaper, 12700 has more cores. Gaming performance between them is splitting hairs. It's a very sad day when an AMD platform is more expensive. 

[Taf the Ghost]

2 hours ago, Dogzilla07 said:

Nope, AMD is pulling an Nvidia here. They're waiting for Intel to show their cards, and decide what to do. If Intel doesn't pressure them, it's gonna take ~22 months again for 5600/x to drop to 150 eur. If Intel does pressure them it's gonna be ~8-12 months to ~150 eur like for ryzen 2600/3600, or sooner.

 

As for the ~100 eur price range, Lisa hinted AMD will not compete no matter what with newest chiplets. Now maybe AMD's bluffing as well, but unlikely (They might do a monolithic APU again or something).

 

What intel has been saying doesn't look promising, but maybe Intel was bluffing. not only is 13100 looking to be carbon copy 12100. 13400, is looking to be just different composition 12600. + Intel has been signaling they have no choice but to raise prices after Q3 2022.

Historically, AMD has supported their previous DDR platform in the medium range economies for a solid 5+ years. There's a LOT of DDR3 Bulldozer parts running around all over the world. I suspect, since AMD will be still producing Zen3 for a while, will simply allow those parts to run their course and make up their "lower range" products for upwards of a year. Intel does their entire product line refresh because of yearly OEM updates, AMD isn't as tied to the OEM market as Intel and has more flexibility. (Realistically, until DDR5 prices are way down from the current spot, very budget sensitive buyers will still go AM4. There's also a note that the "majority of sales" transition point is normally about 9 months after launch. Zen2 was an oddity where that flipped over in like 4 months.)

 

Also, AMD will still be making APUs for laptops.

[Stahlmann]

1 hour ago, ZetZet said:

7600X + platform is extremely expensive. 12600K is cheaper, 12700 has more cores. Gaming performance between them is splitting hairs. It's a very sad day when an AMD platform is more expensive. 

Just wait for B-series motherboards which should start around $125 from what i heard. The B-series chipsets typically have everything a gamer could ever need. X-series chipsets were always in a weird place in terms of price / performance.

 

If you're looking for the best deal, you shouldn't be looking at the newest products.

[williamcll]

I wonder will there be a 7300X or a 7500 lower spec processor.