Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

[Vattila]

Except for the details about the improvements in the microarchitecture, we now know pretty well what to expect with Zen 3.

The leaked presentation by AMD Senior Manager Martin Hilgeman shows that EPYC 3 "Milan" will, as promised and expected, reuse the current platform (SP3), and the system architecture and packaging looks to be the same, with the same 9-die chiplet design and the same maximum core and thread-count (no SMT-4, contrary to rumour). The biggest change revealed so far is the enlargement of the compute complex from 4 cores to 8 cores, all sharing a larger L3 cache ("32+ MB", likely to double to 64 MB, I think).

Hilgeman's slides did also show that EPYC 4 "Genoa" is in the definition phase (or was at the time of the presentation in September, at least), and will come with a new platform (SP5), with new memory support (likely DDR5).

What else do you think we will see with Zen 4? PCI-Express 5 support? Increased core-count? 4-way SMT? New packaging (interposer, 2.5D, 3D)? Integrated memory on package (HBM)?

Vote in the poll and share your thoughts!

 

[leoneazzurro]

Are these really a surprise? We have still general component shortages, and PCIe 5.0 needs high quality PCBs (more layers, more strict quality requirements). We already saw this with ADL mainboards. And what we are seeing today are the X670E/X670 boards, that is, high end to very high end. I saw these prices coming from afar. B650/B650E chipset is probably where the sweet spot is and we will certainly see cheaper MB based on this chipset. But there is definitely a trend for higher MB cost, either for AMD and Intel, and it will not cease soon.

 

[coercitiv]

I was referring to the max temp the CPU will get: it was 95º in GN's review, and they stated this would be so, REGARDLESS of the cooler used.

THIS is the value i wanted to be possible to change in the BIOS.

Same mate, same. I want to bring down temps, because I hate fan noise and heat, it interferes with my concentration.

We now have confirmation the temperature target is configurable via the UEFI. In the case of the Asus firmware it's called Platform Thermal Trottle Limit:



Info available via SkatterBencher.

 

[DrMrLordX]

Grab them while you can. The servermarket is going to take all with that efficiency, unless they can wait for zen5.

AMD has stated that there won't be supply problems of consumer Raphael chips.

Tidy uplift!

Curious. Phoronix Test Suite has been somewhat unkind to desktop AMD CPUs in the past (at least among the tests that Phoronix chooses from the entire suite). See 5950X vs 12900k. Raphael has really gained ground with PTS.

According to Computerbase, when set to the same TDP limit as a 5950x, the 7950x only loses 5% of its average multicore performance.

Kind of makes me wonder at some of the benchmarks showing the 5950X to be more efficient? Ain't no way. Yeah at ~230W the 7950X isn't doing itself many favors, at least not on a 360mm AiO.

It also looks like Zen 4 continues the trend that undervolting is the new overclocking. With an undervolt they showed both less power consumption and higher clocks.

Some of us got that behavior with 12c and 16c Matisse. My 3900X upped its MT clocks after I used vdroop chicanery to trick it into undervolting itself. ST clocks suffered though.

First impression is these chips should've all been X3D from the start.

While stacked cache on N5-family nodes will be ready for use by AMD in a much shorter time frame than stacked cache on N7-family nodes, it's still going to take time to get Genoa-X/Raphael-X to market. AMD certainly wasn't going to delay Genoa, and what else do you expect them to do with the CCDs that won't bin well enough for the server room?

Perhaps lapping will also yield a major improvement. I'm certainly not going to delid and also use a special bracket. Too much hassle and risk. But if I can improve temps by >=10° C from lapping alone, I probably will (if I get a Ryzen 7000).

Lapping is today more risky than a delid, thanks to the use of specialized tools. Though some consider the heating process risky, and I might agree with them. But after my rather expensive failure when lapping a 3900X three years ago, I can honestly tell you that soapy water and metal shavings are very difficult to control.

 

[moinmoin]

AMD did not increase platform power limits (and cost) on AM5 to accommodate power hungry Zen 4 chips pushed past any reasonable limit for a meagre ~5% performance gain. Seems obvious that AM5 power delivery was bumped for the sake of potential future products that raise the core count and might need room to stretch their legs. Zen 4 only consumes as much power as it does because why shouldn't it when there's free performance on the table and 5% could make the difference between winning or losing to Raptor Lake, which will also be over-juiced? I do wish that AMD had defaulted to sane TDPs with the option to use higher TDPs for the extra ~5%, rather than the other way around, though.

This and we so far only got expensive high end "extreme" boards. Come October and beyond I fully expect cheaper boards to appear that don't support 170W TDP but "only" 105W or something along that line. And they will be perfectly fine as well.

OK......MSI internal materials that knocking my head quite hard again:

View attachment 68198


limiting temperature to just 65C, only 5% performance lost.

DIY cooler, no perf lost.........DAFAK

Excellent. With a way to limit it in most BIOSes confirmed limiting the temperature will be the way to go imo. Just need some more detailed benchmarks on what's the effect of different limits and how low one can go. In the above example it looks like there's still room to go even lower than 65°C package power wise.

 

[Shivansps]

4CUs on the iGPU would have been so good here, so people can actually play something while saving for a gpu. That would have been enoght to match the 5600G.

 

[moinmoin]

4CUs on the iGPU would have been so good here, so people can actually play something while saving for a gpu. That would have been enoght to match the 5600G.

It was intended as the bare minimum. It is what it is.

 

[Shivansps]

It was intended as the bare minimum. It is what it is.

It was intended to match what Intel has, nothing more, and it is exactly that. Thats why it runs at that very high clock.

 

[moinmoin]

It was intended to match what Intel has, nothing more, and it is exactly that.

I do wish AMD matched Intel in AV1 encoding support.

 

[TESKATLIPOKA]

It was intended to match what Intel has, nothing more, and it is exactly that. Thats why it runs at that very high clock.

If you want to play, then I think you can OC the IGP to ~3GHz.

 

[nicalandia]

All Hail The New King..!

AMD Ryzen 9 7950X Review: The New Performance King

AMD Ryzen 9 7950X Review: The New Performance King

The Ryzen 9 7950X is the new performance king and the jack of all trades, apart from maybe power consumption and pricing, of course. The Zen 4...

www.techspot.com

 

[evident]

What's the best processor to get for energy efficiency yet still have good performance? I game but not as often as i'd like.

 

[Hitman928]

What's the best processor to get for energy efficiency yet still have good performance? I game but not as often as i'd like.

If you need multi core performance, then the 7950X in eco mode (88W limit or 142W limit depending on your preference). If it's just for gaming, then I'd probably go for the 7700X.

 

[BorisTheBlade82]

In case you are interested: I got a some results for the R9 7950X in the PES | Assessing Power and Performance Efficiency of x86 CPU architectures thread.

 

[JustViewing]

I may be wrong but, It seems DDR5 will not have significant bandwidth improvement for single CCD. It won't have double the bandwidth of DDR4 because CCD to IOD bandwidth hasn't increased much from Zen3 to Zen4. For me 230w is too much, when I buy one I would surely use it with ECO105 or ECO65. Has anyone tested how R9 7950X will perform if we simply use 125w air cooler? Will it throttle to something like 400MHz or will it throttle gradually to 125TDP and does not loose much performance? Also AVX512 is a surprise, it is performing way better than I expected. I would love to tryout some ASM coding with it.

 

[Hitman928]

I may be wrong but, It seems DDR5 will not have significant bandwidth improvement for single CCD. It won't have double the bandwidth of DDR4 because CCD to IOD bandwidth hasn't increased much from Zen3 to Zen4. For me 230w is too much, when I buy one I would surely use it with ECO105 or ECO65. Has anyone tested how R9 7950X will perform if we simply use 125w air cooler? Will it throttle to something like 400MHz or will it throttle gradually to 125TDP and does not loose much performance? Also AVX512 is a surprise, it is performing way better than I expected. I would love to tryout some ASM coding with it.

Theoretically, with a 125W air cooler it should throttle by less than 10% under heavy load. Under lighter loads, including gaming, it shouldn't throttle at all.

 

[Racan]

Looking at Intel's Raptor Lake gaming benches it should be faster that Zen 4 and Zen 3 3D overall, but Zen 4 3D should get back the performance crown without much trouble.

 

[eek2121]

I may be wrong but, It seems DDR5 will not have significant bandwidth improvement for single CCD. It won't have double the bandwidth of DDR4 because CCD to IOD bandwidth hasn't increased much from Zen3 to Zen4. For me 230w is too much, when I buy one I would surely use it with ECO105 or ECO65. Has anyone tested how R9 7950X will perform if we simply use 125w air cooler? Will it throttle to something like 400MHz or will it throttle gradually to 125TDP and does not loose much performance? Also AVX512 is a surprise, it is performing way better than I expected. I would love to tryout some ASM coding with it.

Since the memory interfaces are on the IOD, The CCDs don't need the full bandwidth. They will never use it. IF speeds did improve, however. A lot of the 'lackluster' results with DDR5 relate to application optimization and a very slight latency regression. Desktop also wasn't really bandwidth constrained to begin with.

 

[eek2121]

Looking at Intel's Raptor Lake gaming benches it should be faster that Zen 4 and Zen 3 3D overall, but Zen 4 3D should get back the performance crown without much trouble.

This is not an Intel thread, besides, Intel compared Raptor Lake to the 5950X in their own slide deck. Looking at those numbers alone I can tell you that they will not be ahead of the 7950x.

 

[inf64]

Looking at Intel's Raptor Lake gaming benches it should be faster that Zen 4 and Zen 3 3D overall, but Zen 4 3D should get back the performance crown without much trouble.

Looking at their numbers it barely outperforms Zen 3D on average, so not that impressive to be honest. It's basically balls to the wall OCed part and if their numbers hold up, it should be like ~5-7% faster in games than Zen 4 on average (but needs to be confirmed). Vcache version will dispose of it easily. In productivity 7950X should be faster and consume the same or a bit less power.

Intel's problem is the fact that only the top end model has the super high boost clocks, while the rest have lower clocks than Ryzen counterparts. So 7900X should outperform 13700K across the board, while 7700X should win ST, gaming but lose the MT. 7700X should also outperform 13600K but lose in perf./$ (as intel part is cheaper). 7600X is in the odd spot as it will go against 13600K.

edit:
Here is igor's lab review, just add 30% to the 5950X's number if we are to believe intel and 13900K should score ~111pts on this chart (~4% faster than 7950X).

 

[coercitiv]

Has anyone tested how R9 7950X will perform if we simply use 125w air cooler? Will it throttle to something like 400MHz or will it throttle gradually to 125TDP and does not loose much performance?

Both Intel and AMD modern CPUs throttle gradually to meet their power or temperature targets. The boost algorithms ensure the cores are effectively coasting around the equilibrium point for each workload. Rest assured you will find the appropriate combo of lower TDP and temperature limit, you just need to start from the maximum fan speed you would like and then work your way towards temperature limits and then the power limit.

For example, assume you want a maximum of 50% fan speed at 80C: configure a fan curve that maxes out at 50% @ 80C, set your throttle temp to 80C, then run an intensive MT workload and watch as the CPU gradually throttles towards an equilibrium point.... let's assume it's 135W. This means 170W is too much for your cooler, but something like 142W should be ok in combination with the temperature limit you just set previously.

In a way I'm having so much fun with these 2 CPU launches. Ever since Comet Coffee Lake Refresh I have behaved like a madman on these forums, warning of impending doom & gloom, a time when all CPUs will use way too much power for consumer comfort. Well, that time is finally here, and I am now uniquely equipped to guide unsuspecting souls across the atomic wasteland that Intel and AMD have created.

Repent, sinners. Repent and learn the holy ways of power management!

 

[Leeea]

I have watched a bunch of Ryzen 7000 content now.

Excited about buying Ryzen 5000 soon.

 

[nicalandia]

I have watched a bunch of Ryzen 7000 content now.

Excited about buying Ryzen 5000 soon.

5800X3D or Bust....

 

[Leeea]

5800X3D or Bust....

The 1% lows on the 5800x3d are impressive.

 

[lightmanek]

Both Intel and AMD modern CPUs throttle gradually to meet their power or temperature targets. The boost algorithms ensure the cores are effectively coasting around the equilibrium point for each workload. Rest assured you will find the appropriate combo of lower TDP and temperature limit, you just need to start from the maximum fan speed you would like and then work your way towards temperature limits and then the power limit.

For example, assume you want a maximum of 50% fan speed at 80C: configure a fan curve that maxes out at 50% @ 80C, set your throttle temp to 80C, then run an intensive MT workload and watch as the CPU gradually throttles towards an equilibrium point.... let's assume it's 135W. This means 170W is too much for your cooler, but something like 142W should be ok in combination with the temperature limit you just set previously.

In a way I'm having so much fun with these 2 CPU launches. Ever since Comet Coffee Lake Refresh I have behaved like a madman on these forums, warning of impending doom & gloom, a time when all CPUs will use way too much power for consumer comfort. Well, that time is finally here, and I am now uniquely equipped to guide unsuspecting souls across the atomic wasteland that Intel and AMD have created.

Repent, sinners. Repent and learn the holy ways of power management!

I rock my 480mm + 360mm custom loop and look forward to 7950x build, but I like quiet operation for work and don't mind a bit of fan noise for gaming as then I wear headphones. My current 5950x spends 90% of time underclocked to 2.6GHz across all cores and just 0.8V with occasional bump to 3.6GHz for older games or stock operation for more CPU demanding games. For 7950x I see even more opportunities to tune it to perfection with temperature and power targets to get me most performance at almost no noise! Obviously for fun I will try to push it to 6GHz if possible on water and do few animation renders or video productions when I need.
But that purchase will be after 13900k launches as I expect good competition and maybe even prices to budge a bit downwards

 

[Shivansps]

I do wish AMD matched Intel in AV1 encoding support.

I think i saw that this igpu does support AV1 un a slide.