Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

[DisEnchantment]

Speculate at will

 

[eek2121]

The X3D parts could launch around gamescom. I haven’t heard anything specific about them, but I do hope we get a 9950X3D and I hope they improve the clocks on the chiplets with 3D V-Cache. If they could reach clock parity with non-3D parts, Intel won’t have a chance.

 

[CP5670]

I would get the X3D when it comes out, 15-20% is a decent bump and I like how AMD doesn't make you buy a new platform every time like Intel does. The 7800X3D is great but mine does not clock very well, tops out at 4.6-4.7ghz in most games. A fast single thread is still important for older games.

 

[Gideon]

Pretty decent scores is true:

https://twitter.com/x/status/1800185782814257253

 

[TitanicFreak]

Decent voltage for those clocks if that readout is correct.

 

[DisEnchantment]

Pretty decent scores is true:

https://twitter.com/x/status/1800185782814257253

If this is not an OC, clocks are great on the 65W 9600X (vs 5.3G on 105W 7600X)
Also ~11% gain/clock on a rather unfavorable benchmark (vs 1% on Z4, using 770 points for 7950X at 5.7G as reference)

 

[csbin]

If this is not an OC, clocks are great on the 65W 9600X (vs 5.3G on 105W 7600X)
Also ~11% gain/clock on a rather unfavorable benchmark (vs 1% on Z4, using 770 points for 7950X at 5.7G as reference)

It's obviously overclocked.

 

[DisEnchantment]

It's obviously overclocked.

Well, it is not obvious to me unless they mention it, because my 7950X spec says 5.7G but it runs 5.8G+ most of time in ST .

 

[eek2121]

Just wanted to add a few quick thoughts here, before my world gets super busy:

1. AMD has RDNA4 and X3D parts left to announce/launch.
2. A board partner stated that next-gen Radeon cards were launching at SIGGRAPH. AMD has done something similar before, but it was always an odd choice to me. Gamescom or a similar event would have made more sense.
3. X3D parts are also apparently "coming soon"
4. Recently it was claimed that partners at Computex said RDNA4 wasn't dropping until next year.

AMD doesn't really do multiple product launches at the same time, they like to spread them out to keep the buzz going, so I suspect one of two scenarios will happen:

1. RDNA4 at SIGGRAPH, possibly SIGGRAPH Asia, X3D parts at gamescom. (this would be my bet based on what I've heard and based on the fact AMD is still going ham with the driver updates for RDNA4)
2. RDNA4 at CES, X3D parts at Gamescom (or a custom event)

Note that if RDNA4 launches at SIGGRAPH Asia, availability would indeed not be until 2025.

All of this excludes the possibility of AMD just launching without attending any event at all. They've done that before.

I know this isn't an RDNA4 thread, but it relates due to scheduling, so I included the info here.

 

[eek2121]

Well, it is not obvious to me unless they mention it, because my 7950X spec says 5.7G but it runs 5.8G+ most of time in ST .

Note that unless you override it, the default limit is 5850mhz for the non-X3D 7950X. AMD has been adding a 150mhz buffer to these chips. They did this with Zen 3 as well (5050mhz for the 5950x).

My Zen 3 outperformed (I can add a +200mhz boost override and it'll do it! WITH curve optimizer as well!), my Zen 4 chip? It won't go above 5.75ghz at all. If I add a 25mhz boost clock override it crashes on boot. (I need to look and see how long AMD's warranty is, I hope to wait until the Zen 5 X3D parts drop, RMA it since it was damaged with the SOC voltage nonsense, and then sell the replacement or build a system for my daughter)

Zen 5 appears to be sticking with the same limits, FYI, but they CAN be overridden, provided the chip can handle it.

 

[CouncilorIrissa]

Pretty decent scores is true:

https://twitter.com/x/status/1800185782814257253

I absolutely loathe the fact that those with the CPUs in their possession are hell-bent on running the world's most useless benchmark.

 

[B-Riz]

I absolutely loathe the fact that those with the CPUs in their possession are hell-bent on running the world's most useless benchmark.

They have not discovered Phoronix or AT cpu reviews yet.

 

[fastandfurious6]

I think the blunder they made is not releasing 7745HX3D. That could've really given 13980HX/14900HX laptops a run for their money. 7945HX3D is too high end to be affordable.

nah, the price difference between mobile cpus per se is not too high

if u find laptop with 7945hx + 4060 it gonna be cheap

the problem is top-end cpus are forcecoupled only with top-end gpus which are obscenely expensive

e.g. 7945hx only with rtx 4080 or 4090, most times just 4090
price increase per gpu
integrated -> 4060: ~500
4060 -> 4070: ~300
4070 -> 4080: ~800
4080 -> 4090: ~1000

basically mobile 4090 costs ~2000-2500 just by itself

 

[fastandfurious6]

The X3D parts could launch around gamescom. I haven’t heard anything specific about them, but I do hope we get a 9950X3D and I hope they improve the clocks on the chiplets with 3D V-Cache. If they could reach clock parity with non-3D parts, Intel won’t have a chance.

more cache > higher clocks

imagine if games/apps actually micromanaged cache operations in a humongous 300mb l3 cache

100% cache hitrates

u mostly do 1 thing at a time on a computer anyway

 

[igor_kavinski]

https://twitter.com/x/status/1800185782814257253

Spoiler: My OCed 12700K for comparison

Intel Core i7 12700K @ 5285.74 MHz - CPU-Z VALIDATOR

[uqse3g] Validated Dump by Sonic PG Z790 12700K DDR5-6000 CL28 (2024-06-10 18:17:37) - MB: ASRock Z790 PG SONIC - RAM: 32768 MB

valid.x86.fr

Now for some CPU-Z IPC calculation:

ST (divide score by highest clock)

9600X 0.153 per MHz

12700K 0.162 per MHz

MT (divide score by core count)

9600X 591.383 per core

12700K 496.425 per core

That is MAD performance per core!

Delivering 19% higher perf per core against Alder Lake, in an unfavorable benchmark!

And this their entry level chip???

Intel better pray!

 

[CouncilorIrissa]

Spoiler: My OCed 12700K for comparison

View attachment 100899
View attachment 100898
View attachment 100897

Intel Core i7 12700K @ 5285.74 MHz - CPU-Z VALIDATOR

[uqse3g] Validated Dump by Sonic PG Z790 12700K DDR5-6000 CL28 (2024-06-10 18:17:37) - MB: ASRock Z790 PG SONIC - RAM: 32768 MB

valid.x86.fr

Now for some CPU-Z IPC calculation:

ST (divide score by highest clock)

9600X 0.153 per MHz

12700K 0.162 per MHz

MT (divide score by core count)

9600X 591.383 per core

12700K 496.425 per core

That is MAD performance per core!

Well the 12700K has E-cores, so of course its performance per core is lower.

 

[igor_kavinski]

Well the 12700K has E-cores, so of course its performance per core is lower.

Yeah but it's a comparison of the entire chip's capability.

Intel better pray to sell out those 12700Ks soon coz 9600X is coming to whoop its behind at HALF power use.

 

[invisible_city]

Arrow Lake would have to be very special. Qualcomm?? Too small in PC space to matter right now.

I mean if I were AMD and saw Lunar Lake, I would hedge that Arrow Lake would be real competition, and that Zen 5 may not be clearly better (but X3D may be)

 

[Det0x]

Pretty decent scores is true:

https://twitter.com/x/status/1800185782814257253

Hmm

9600X's advertised clockspeed is "5.4ghz"
Maybe all Zen5 SKUs will have fmax set +300mhz FMAX over "advertised clockspeed", but seems unlikely

Zen4 had set fmax +100 +150mhz over advertised clocks

 

[eek2121]

They have not discovered Phoronix or AT cpu reviews yet.

I would argue Phoronix has the most useful benchmark suite of them all, since it's based almost entirely on real world workloads. It is also open! The only drawback is that it is Linux based. I'm sure you could probably run it on Windows with some effort, but I haven't looked at it in quite a while.

I mean if I were AMD and saw Lunar Lake, I would hedge that Arrow Lake would be real competition, and that Zen 5 may not be clearly better (but X3D may be)

Unless Intel has decided to completely mislead, The "E" cores are about as fast as Zen 4 +-10%.

I think X3D chips will be the champs of gaming. I don't see Intel overcoming that. I think Intel will probably win this gen overall (minus gaming) unless AMD does something uncharacteristic with X3D, which is a very real possibility.

 

[uzzi38]

Zen4 had set fmax +100mhz over advertised clocks

+150MHz, to be exact. 7950X FMax is 5.85GHz.

There was a rumour that Zen 5 FMax is 6.1GHz. that could very well end up have been true given the clocks/voltage here, because that 5.7GHz at ~1.25v for relatively low end silicon implies that for the well binned stuff 5.7GHz at 1.2v may be possible. And in case you're wondering why 1.2v matters - that's V-Cache die territory.

 

[Abwx]

Unless Intel has decided to completely mislead, The "E" cores are about as fast as Zen 4 +-10%.

That s a myth to believe that a e core can have the same perf as RPL that is 2x the size, as for Zen 4 i just looked at Computerbase numbers the other day and the 7700X is 5.2% faster per clock than the 8 P cores of a 12900K, so this well extend to 22% for Zen 5 and they will be still ahead by 7% when Intel s next gen will land.

 

[exquisitechar]

+150MHz, to be exact. 7950X FMax is 5.85GHz.

There was a rumour that Zen 5 FMax is 6.1GHz. that could very well end up have been true given the clocks/voltage here, because that 5.7GHz at ~1.25v for relatively low end silicon implies that for the well binned stuff 5.7GHz at 1.2v may be possible. And in case you're wondering why 1.2v matters - that's V-Cache die territory.

If they increased clock speeds that much it's weird that the rated boost on the 9950X is the same as on the 7950X.

 

[invisible_city]

I would argue Phoronix has the most useful benchmark suite of them all, since it's based almost entirely on real world workloads. It is also open! The only drawback is that it is Linux based. I'm sure you could probably run it on Windows with some effort, but I haven't looked at it in quite a while.

Unless Intel has decided to completely mislead, The "E" cores are about as fast as Zen 4 +-10%.

I think X3D chips will be the champs of gaming. I don't see Intel overcoming that. I think Intel will probably win this gen overall (minus gaming) unless AMD does something uncharacteristic with X3D, which is a very real possibility.

If intel is right, the E Cores are faster than Zen 5 at the same power envelope (20% faster than Raptor Cove at low power). Discounting AVX512 workloads of course.

Edit: in lunar lake will be a bit weaker due to no l3, so maybe faster than zen5c and close to zen5.

 

[adroc_thurston]

If intel is right, the E Cores are faster than Zen 5 at the same power envelope (20% faster than Raptor Cove at low power). Discounting AVX512 workloads of course.

have you seen Z5 power curves? no? funny.

 

[invisible_city]

That s a myth to believe that a e core can have the same perf as RPL that is 2x the size, as for Zen 4 i just looked at Computerbase numbers the other day and the 7700X is 5.2% faster per clock than the 8 P cores of a 12900K, so this well extend to 22% for Zen 5 and they will be still ahead by 7% when Intel s next gen will land.

A myth? lol these are official figures from Intel. 40% lower power at same perf 1.2x perf at same power. And 20-80% better PPW at low power than Lion Cove. If Intel isn’t lying, Chadmont is roughly on par with Zen5 at low power.

Edit: updated perf data for completeness.