AMD Ryzen 7 9700X

[LittleBro]

And all of this is worth $100 to you? A 50% higher price?
You want to spend extra $100 in order not to buy a robust cooler (a $35 Peerless Assassin is very robust, indeed). You actually need to buy a cooler for the 9600X, while the 7600 has one included (it will be loud, but it's free).
You want to spend extra $100 to have 20% better efficiency? It's not even true. The 7600 literally has better gaming efficiency, and in Cinebench the 9600X only wins by 10%. How many years will it take to save that $100 in energy costs?
But you get 5% more gaming performance and 11% productivity over the 7600? Wow, for $100 more, very impressive indeed.

I'd definitely pay $100 extra and go for 9600X instead of 7600X just to have my PC quieter and cooler for upcoming 3 years. Lower power draw = lower temps = less noise. The difference between 65°C and 85°C in terms of PWM fan control is enormous. For instance, my currect profile is set to spin fans at 10% PWM while CPU temp is being lower than 50°C, then raises proportionaly to 30% PWM until 68°C. Temps average during gaming at around 62-64°C, that's around 380-410 RPM, almost inaudible. I've never seen my CPU go beyond 73°C. I can't imagine that noise with 85°C and even beyond. It would take at least 70-80% PWM (1200-1300 RPM) to prevent temps from raising further. (Numbers presented apply to my fans.)

I like it quiet. Shush!

I have no idea where your logic comes from. Why not the 3600? Because the 9600X is 53% faster than the 3600. And the 7600 is 46% faster than the 3600. And the 3600 is on an old platform with inferior features (PCI-E, NVMe, USB). You're paying more, but you're getting a lot more. With the 9600X you literally get nothing over the 7600. Irrelevant improvement in efficiency and performance at a 50% higher price.

That reference to Ryzen 3600 was a kind of joke. But according to TPU it is the best bang for the buck right now.

An RTX 4090 has terrible performance per dollar, but it doesn't matter. Why? Because it's 33% faster than the next card in the line-up. If the 9600X offered 33% more performance at a 50% higher price, nobody would be complaining.

Everyone has this line of when it starts to matter defined differently.

 

[THU31]

I'd definitely pay $100 extra and go for 9600X instead of 7600X just to have my PC quieter and cooler for upcoming 3 years. Lower power draw = lower temps = less noise. The difference between 65°C and 85°C in terms of PWM fan control is enormous. For instance, my currect profile is set to spin fans at 10% PWM while CPU temp is being lower than 50°C, then raises proportionaly to 30% PWM until 68°C. Temps average during gaming at around 62-64°C, that's around 380-410 RPM, almost inaudible. I've never seen my CPU go beyond 73°C. I can't imagine that noise with 85°C and even beyond. It would take at least 70-80% PWM (1200-1300 RPM) to prevent temps from raising further. (Numbers presented apply to my fans.)

So you'd pay $100 instead of flicking a switch and changing the TDP profile from 105 W to 65 W in the BIOS, which the 7600 does by default, and is the one I was referencing, not the X model.

Because this is what the 9600X is. It's a 9600 for $280 and without a cooler.

I've heard enough.

 

[RandallFlagg]

TPU and HUB have bigger game testing suite compare to other tech outlets, TPU tested 14 games (10 + 4 RT) and HUB tested 13 games. Meanwhile Toms tested 7 and Kitguru tested 4 games.

It's not very encouraging to see 9700X slower than 7700 in 1% Low FPS
View attachment 358089

I had to go check some of the individual benchmarks and cross compare a little, and yeah that is exactly what is going on. Cyberpunk 2077 for example, without RT both Tom's and TPU show the 9600X performing very well - beating a 14700K by 8-13% depending which site does the test.

So they're not really in conflict, just none of them have a particularly large sample size of games and Zen 5 performance seems to vary pretty wildly from one game to the next.

 

[AnotherReader]

I get that for Zen 5 AMD went for efficiency, this will also help them more with their EPYC server CPUs that will have even more cores (up to 196 for Turin that will compete with efficient ARM CPUs)

Still, at 65W the 9700x is handicapped, the $360 price is too high and performance is not consistent and lower than expected is some areas.

AMD was generally spot on with their IPC estimates for previous Zen releases, but this time they promised a 16% IPC increase and the performance doesn't seem anywhere near that. Maybe things improve with new driver / AGESA updates, but we'll see.

They are still right about IPC increases, but they didn't mention the lower clocks in multithreaded workloads that almost nullify that increased IPC.

 

[W1zzard]

Granted, I have not checked every single review out there but still quite a few sites and TPU are literally the ONLY one who are saying that the 9700X can hold a candle to the X3D CPUs of the 7000 generation.

In fact, TPU said that the 9700X is faster than the 7950X3D (0.8% at 1080p). I can not even remotely find this result and these findings reflected in a single other review.
"All" other reviews show 7800X3D/7950X3D with a considerable lead over the 9700X.

This is why I would like to ask the slightly provocative question: What went wrong with benchmarking at TPU?

For the record, I think you are usually a great source with reliable data. But something does not add up here. Literally no one else but you sees 9700X even in the same ballpark as the 7800X3D and 7950X3D.

I figured it out .. when I tested 7950X3D I didn't have the Xbox Game Bar app installed & running, so it ran without game detection. Game bar is not "the Xbox app" which installs a lot of other junk, too. After installation there's some version mismatch so I had to reinstall the AMD chipset drivers, then everything worked ok. All games have been retested on 7950X3D and charts in both reviews have been updated.

 

[Mr_Engineer]

They are still right about IPC increases, but they didn't mention the lower clocks in multithreaded workloads that almost nullify that increased IPC.

Taking the TPU review (great reviews as always @W1zzard), productivity performance for stock 9700x at 65W, is 7.5% faster than the 7700 at 65W.

Toms Hardware review has the stock 9700x about 10% faster than the 7700.

The clocks are also similar, 9700x has 3.8MHz base - 5.5 boost, to the 7700 which has 3.8MHz base - 5.3MHz boost.

Overall it looks like Zen 5 offers 7-10% average IPC increase compared to Zen 4, while providing better power efficiency. Not bad but not great either.

AMD should have named this CPU "9700" and priced it $299 to make it a great buy.

 

[AnotherReader]

Taking the TPU review (great reviews as always @W1zzard), productivity performance for stock 9700x at 65W, is 7.5% faster than the 7700 at 65W.

Toms Hardware review has the stock 9700x about 10% faster than the 7700.

The clocks are also similar, 9700x has 3.8MHz base - 5.5 boost, to the 7700 which has 3.8MHz base - 5.3MHz boost.

Overall it looks like Zen 5 offers 7-10% average IPC increase compared to Zen 4, while providing better power efficiency. Not bad but not great either.

AMD should have named this CPU "9700" and priced it $299 to make it a great buy.

Clocks in computationally demanding multithreaded workloads are actually significantly lower. The IPC increase is significant, but the low clock speeds under load are wiping most of it away.

 

[RogueSix]

I figured it out .. when I tested 7950X3D I didn't have the Xbox Game Bar app installed & running, so it ran without game detection. Game bar is not "the Xbox app" which installs a lot of other junk, too. After installation there's some version mismatch so I had to reinstall the AMD chipset drivers, then everything worked ok. All games have been retested on 7950X3D and charts in both reviews have been updated.

Thanks for the update and I'm glad you figured it out . I hope you also had 'GameMode' turned on in Windows because AFAIK both is necessary for correct scheduling on dual CCD Ryzens. You need to run the Xbox GameBar (which tells the CPU when an *.exe is a game) and you also need to have GameMode turned on under 'Settings' - 'Gaming'.

As far as updates to your review are concerned, I think you might also want to errr... review the gaming conclusion as it still reads...

Our gaming tests show that AMD has definitely improved gaming performance over Zen 4—the 9700X beats almost any previous AMD processor, including the 7950X3D—only the 7800X3D is faster

^ It looks like this is no longer accurate after the updated results, right?

 

[EntropyZ]

If AMD was trying to get me to buy the Ryzen 7800X3D or any Zen 4 part for that matter, well they have certainly managed to convince me. Maybe there's not a lot of performance left to squeeze out of the current Zen architecture, but to call 9000 series Zen 5 and not Zen 4+ is a bit bold on their part considering everything. The 9600X/9700X are a pass for me until the prices drop down, the extra cost over 7000 series parts can't be worth it right now. I was hoping two years of time would have provided more improvements.

It's possible they will manage to release X3D parts on the same node with some slight changes. And those seem to be the ones to look out for, at least for gaming.

 

[Punkenjoy]

For servers it is extremely pathetic, as well, because the servers require maximum computational density, which a crappy 8-core CCD won't offer.
They needed a 16-core CCD at this point, a 24-core CCD at 3nm, and a 32-core CCD at 2nm, and preferably now.

It is actually not pathetic at all. For computational density, you need power efficiency. Also, having the possibility to cramp as much CPU into a rack unit is one usecase, but that is far from being the only usecase for servers.

Anyway, if you want to be able to acheive that, you still need to be able to put as much performance with the minimum amount of watt. We all focus on the number of cores for servers because its an obvious way of improving performance but its not the only one. ut

its great to have 128 cores at 2 GHz, but if you can outperform it with 64 cores at 3 GHz that have higher IPC, you better use that path.

A lot of software you run on server is licensed on cores. If you want to maximize your investisment with those kind of workload, you want a lot of performance for each of those core. You dont want to have more core.

The Zen 5 CCD is the same size as the Zen 4 CCD. that mean they will probably put the same amount of them on their EPYC lineup. But it will probably mean that for the same power, they will be able to deliver more performance and higher boost clock.

 

[Chry]

Why is an old BIOS version used? There are already two newer versions out:

 

[Zach_01]

If AMD was trying to get me to buy the Ryzen 7800X3D or any Zen 4 part for that matter, well they have certainly managed to convince me. Maybe there's not a lot of performance left to squeeze out of the current Zen architecture, but to call 9000 series Zen 5 and not Zen 4+ is a bit bold on their part considering everything. The 9600X/9700X are a pass for me until the prices drop down, the extra cost over 7000 series parts can't be worth it right now. I was hoping two years of time would have provided more improvements.

It's possible they will manage to release X3D parts on the same node with some slight changes. And those seem to be the ones to look out for, at least for gaming.

They "couldn't" call it 4+. You only saying that because you judge things from gaming performance perspective.
If you followed this conversation from start you would've know that there are a lot of architectural changes in Zen5.
The thing is that those changes affect gaming performance to the minimum compared to other work loads.

Some of us we are saying it for a while (a few years) now.
The Zen architecture(s) from start. and the way its build (chiplets) was/is primarily for server applications.
Its the field with most of the profit.

The Zen 5 CCD is the same size as the Zen 4 CCD. that mean they will probably put the same amount of them on their EPYC lineup. But it will probably mean that for the same power, they will be able to deliver more performance and higher boost clock.

Zen5 CCD is actually a bit smaller (~2% area) while having 30% more transistors in it if I'm not mistaken.

View attachment 358185
Why is an old BIOS version used? There are already two newer versions out:

View attachment 358186

And how is this relevant to this thread?

 

[Chry]

And how is this relevant to this thread?

It affects the comparison. I'd like to see all processors compared with the latest versions of their respective microcode.

 

[RandallFlagg]

If AMD was trying to get me to buy the Ryzen 7800X3D or any Zen 4 part for that matter, well they have certainly managed to convince me. Maybe there's not a lot of performance left to squeeze out of the current Zen architecture, but to call 9000 series Zen 5 and not Zen 4+ is a bit bold on their part considering everything. The 9600X/9700X are a pass for me until the prices drop down, the extra cost over 7000 series parts can't be worth it right now. I was hoping two years of time would have provided more improvements.

It's possible they will manage to release X3D parts on the same node with some slight changes. And those seem to be the ones to look out for, at least for gaming.

There's actually a lot to be squeezed out, because Zen 5 is fundamentally different from Zen 4 or anything prior.

Zen 5's architecture looks a lot more like an Apple M1 than it does Zen 4.

The increased number of pipelines means it will run at lower frequency, but will get more done per clock. Hence, more efficient. This is frankly the 'old school' way of getting efficiency from a CPU design. Narrow and deep pipes can clock higher, wide and shallow clock lower.

It's pretty clear that Zen 5 architecture is leaning hard towards server use cases, and also mobile use cases.

But in doing so, Zen 5 also eschews many workstation / performance desktop use cases.

 

[Zach_01]

It affects the comparison. I'd like to see all processors compared with the latest versions of their respective microcode.

That is true.
Reviews should be repeated in the future.

 

[mkppo]

AMD was generally spot on with their IPC estimates for previous Zen releases, but this time they promised a 16% IPC increase and the performance doesn't seem anywhere near that. Maybe things improve with new driver / AGESA updates, but we'll see.

They actually have higher than 16% IPC in SPEC, about the same in INT but 25% in FP. Omit the AVX512 workloads and you're still looking at 15% for INT and near 20% for FP. The single threaded gains are actually pretty damn high considering that they could've had more had they not focused so much on FP and the AVX512 path. The problem is this ST gain doesn't entirely materialise in MT workloads because the sweet spot in terms of power has been bumped up while at the same time they've reduced stock power draw and max voltages for all core boost. Run both on PBO, and you get 10% uplift in MT on average (it's 6% at 65W so you can see Zen 5 scales past Zen 4 in terms of power draw). MT workloads and especially games seem heavily memory and IO bottlenecked as well, which reduces that 15% gain to around 10%.

They really needed a new IO die for Zen 5 with a faster IF. There's easily double digits of performance gain to be had once they fix the back end and IO. Here's hoping X870 can do DDR5 8000, I'm hearing there will be EXPO kits out for plug and play 8000Mhz but there will still be bottlenecks because the IF is stuck at 2000mhz.

Clocks in computationally demanding multithreaded workloads are actually significantly lower. The IPC increase is significant, but the low clock speeds under load are wiping most of it away.

View attachment 358151

This is because at the same power, each Zen 5 core has more stuff to fire up. More registers, execution engines yada yada. So at the same power, there will be a clock regression but the IPC increase makes up for it. Crank both up to whatever the max performance is to erase the clockspeed difference (probably at around 90W for 7700X and 105W for Zen 5 - past this they stop scaling) and the MT performance increase goes up to 10% average - significantly higher in some workloads and less in others.

I figured it out .. when I tested 7950X3D I didn't have the Xbox Game Bar app installed & running, so it ran without game detection. Game bar is not "the Xbox app" which installs a lot of other junk, too. After installation there's some version mismatch so I had to reinstall the AMD chipset drivers, then everything worked ok. All games have been retested on 7950X3D and charts in both reviews have been updated.

Thank you for the hard work. I was wondering why the results are low a couple of months ago but chalked it up to game selection. After this review though, I realised it can't be that much slower than 14900K because most other review sites show the 7950X3D is faster. This result is more in line with what's expected.

On a side note, it's also interesting to see that with no game bar, it's pretty much identical to 7950x so you definitely need the game bar if you want the benefits of X3D.

 

[W1zzard]

View attachment 358185
Why is an old BIOS version used? There are already two newer versions out:

View attachment 358186

Yes, I spent a month on developing out new benchmark tests and finding test scenes for games in April, then a couple of weeks retesting 40 CPUs in May, so I could fly to Computex in early June, to Barcelona in mid-June, to Los Angeles in July, and be back to test Zen 5 with enough comparison data to write a proper review.

Are you saying I should have started retesting CPUs on July 12 with a day-zero BIOS? How about wait a week before I know it's stable .. that's Jul 19th, at this point it's just not possible to retest everything in time.

The other BIOS is beta and came out yesterday, I'm not allowed to use my time machine for benchmarking

 

[Neo_Morpheus]

I figured it out .. when I tested 7950X3D I didn't have the Xbox Game Bar app installed & running, so it ran without game detection. Game bar is not "the Xbox app" which installs a lot of other junk, too. After installation there's some version mismatch so I had to reinstall the AMD chipset drivers, then everything worked ok. All games have been retested on 7950X3D and charts in both reviews have been updated.

Given the “better” results observed by Phoronix, would you be interested in doing some benchmarks of those games but under Linux?

If you dont want to spend too much time configuring a system, just use Bazitte.

Would be interesting to see such test.

 

[W1zzard]

would you be interested in doing some benchmarks of those games but under Linux?

Absolutely not, not enough time and not a Linux desktop user, I only know my servers

 

[ShogoXT]

Level1tech handles Linux stuff too

Was this all a scam to make a 7700 successor without a cooler at higher cost? Incoming 9700xt and 9700xtx?

Yes, I spent a month on developing out new benchmark tests and finding test scenes for games in April, then a couple of weeks retesting 40 CPUs in May, so I could fly to Computex in early June, to Barcelona in mid-June, to Los Angeles in July, and be back to test Zen 5 with enough comparison data to write a proper review.

Are you saying I should have started retesting CPUs on July 12 with a day-zero BIOS? How about wait a week before I know it's stable .. that's Jul 19th, at this point it's just not possible to retest everything in time.

The other BIOS is beta and came out yesterday, I'm not allowed to use my time machine for benchmarking

I'm curious after all these years the strides you have made to automate the process. A lot of sites have definitely cut down on the variables so it must be difficult.

 

[W1zzard]

I'm curious after all these years the strides you have made to automate the process

Even assuming 100% automation, tests take time to run. Feel free to do the math, 40 CPUs with 49 apps + (10 games + 3 rt) x 4 resolutions

 

[trparky]

OK, so if faced with buying/building a new system today when these new chips become available, would it be better to go with the older generation (7000-series) or go with the newer (9000-series) despite it not being all that it's cracked up to be?

 

[Kapone33]

OK, so if faced with buying/building a new system today when these new chips become available, would it be better to go with the older generation (7000-series) or go with the newer (9000-series) despite it not being all that it's cracked up to be?

The first question would what is your use case and then what is the budget?

 

[Dyno]

Made the chart for you:

https://imgur.com/HLLGSXh

No plans for dynamic charts at this time, even though I get why you're liking them and how random combinations could be useful.


AMD says "Memory: Some processors may be able to achieve EXPO 6400 1:1 with manual settings. By default AGESA will set any memory profile above 6000 MT/s to 1:2 mode, but an end user may override this to 1:1. Stability of this configuration will vary based on the specific processor. A latency optimized 1:1 EXPO memory profile will provide the best performance in a wide range of applications. There is no need to make any other adjustments. If you desire to make further tweaks AMD recommends trying to tighten the timings as much as possible as AUTO:1:1 DDR5-6000 MHz remains as the “sweet spot” for price and performance"

6000 works really well, 6400 requires a bit of luck and some tweaking of voltages, i.e. making it non-stock. DDR5-8000 MHz is possible, but due to the 1:2 mode it won't be that much faster. I have a G.SKILL kit coming, so will have data on this soon (not until after the 2nd round of reviews)

This is the review that i was looking at. In regards to the 8000Mhz memory and the 9000 CPU benchmark.
AMD Ryzen 9000 Series Review - 9600X & 9700X - OC3D (overclock3d.net)

 

[trparky]

The first question would what is your use case and then what is the budget?

OK then...