AMD Ryzen 5 7500F Desktop Processor Surfaces, Could this be Phoenix-2 on AM5?

[btarunr]

A screenshot from Puget Systems benchmark database reveals a new upcoming desktop processor model by AMD, the Ryzen 5 7500F. The screenshot details the 7500F as a 6-core processor, and the machine features an ASUS ROG Strix X670E-F Gaming motherboard, along with an RTX 4080 graphics card. At this point it's hard to tell what the "F" brand extension means in AMD nomenclature. On Intel, it denotes a lack of integrated graphics.

There are two possible theories on what the 7500F could be. One holds that it's a down-rated "Raphael" MCM with a disabled iGPU; while the other holds that it could be based on the 4 nm Phoenix-2 monolithic silicon. Detailed in an older article, the Phoenix-2 is a 137 mm² monolithic silicon that physically features no more than 6 "Zen 4" CPU cores, and an iGPU with just 4 RDNA 3 compute units, besides I/O that's identical to that of the regular 178 mm² 8-core/12-CU Phoenix silicon. Phoenix-2 on AM5 might just end up with a lower bill of materials than a single-CCD "Raphael" MCM.

Update 06:13 UTC: A Korean retailer has posted the first picture of the Ryzen 5 7500F in the flesh. They claim a street price of around $170-180 (KRW equivalent), and availability slated for July 7.



View at TechPowerUp Main Site | Source

 

[Minus Infinity]

IMO if it's Phoenix 2, I think the 7500F naming would mean it would be configured as 4 Zen 4 + 2 Zen 4c cores rather than 2 Zen 4 + 4 Zen 4c which I'd call a 7400F.

 

[NixZero]

IMO if it's Phoenix 2, I think the 7500F naming would mean it would be configured as 4 Zen 4 + 2 Zen 4c cores rather than 2 Zen 4 + 4 Zen 4c which I'd call a 7400F.

main difference between zen 4 and zen4c is the lower l3 size (16mb x 8 core ccx instead of 32mb x 8 core ccd *) so phoenix and phoenix 2 cores can be classified as 4c, there is no 32mb/ccd version.

(*) there are other differencies like the kind of sram cell used but those arent declared in specs so you can distinguish them only by a deep analysis.

 

[btarunr]

IMO if it's Phoenix 2, I think the 7500F naming would mean it would be configured as 4 Zen 4 + 2 Zen 4c cores rather than 2 Zen 4 + 4 Zen 4c which I'd call a 7400F.

Phoenix-2 doesn't have any "Zen 4c" cores. There's just one CCX with 6x "Zen 4" cores. The L3$ size is reduced to 8 MB, but there's no confirmation of this.

 

[vbq7qK68eyYAH4iR]

They could be pulling the "F" nomenclature from their server environment, where it means higher clocks and higher power.

 

[kapone32]

Just give me a Desktop verison of the 7940 HX or whatever. I want to build a SFF PC based on an APU. To be honest I would love to build my own handheld.

 

[Assimilator]

I'd be excited if this was an 8-core part, but 6 cores just isn't interesting in 2023, especially considering the higher platform cost of AM5.

 

[msroadkill612]

A rig that saves the costs of GPU, & associated power & heat downsides, yet provides entry level GPU perf, makes any incremental costs for an AM5 mobo academic.

 

[tabascosauz]

main difference between zen 4 and zen4c is the lower l3 size (16mb x 8 core ccx instead of 32mb x 8 core ccd *) so phoenix and phoenix 2 cores can be classified as 4c, there is no 32mb/ccd version.

(*) there are other differencies like the kind of sram cell used but those arent declared in specs so you can distinguish them only by a deep analysis.

I previously thought the same, but I'm pretty sure AMD intends for Zen 4c only to refer to the (so far only confirmed in enterprise) version of Zen 4 that is half cache + double core count (16) + floorplan reshuffled and space optimized to fit into almost the same area as the existing Zen 4 CCD. It's not explicitly stated or confirmed by AMD, but Semianalysis inferred that the same extreme space optimization that made Zen 4c impressive will also cause significant clockspeed penalties at the top end. Which immediately rules out Phoenix, though maybe not PHX2 as a budget product.

The Bergamo Zen 4c design looks to be very specifically focused on doubling core count while largely fitting into the existing layout of the CCD and IFOP connection, both of which are irrelevant considerations to PHX2 or any other APU.

So, at this point it's still 3: 32MB CCX/CCD (with TSVs in L3 for optional Vcache), 16+16MB Zen 4c CCD, and 16MB CCX for APU.

Maybe PHX2 does something further funky with reducing L3. It looks the rumor might be true, since the 8-core CCX has always taken up a lot of space on APU and most of the change comes from being skinnier than Phoenix. Raven Ridge/Renoir/Cezanne/Rembrandt all have roughly similar floorplans. Take out 8MB L3, remove 2/3 of the CUs, tidy things up and reshuffle some I/O on the edge and it looks about right.

 

[Jism]

I'd be excited if this was an 8-core part, but 6 cores just isn't interesting in 2023, especially considering the higher platform cost of AM5.

It's 12 threads, 6 cores.

The SMT with AMD always gains you extra performance.

 

[zenlaserman]

It's 12 threads, 6 cores.

The SMT with AMD always gains you extra performance.

This is like saying water is wet, but only from a certain tap. Anyway, thanks for the lesson, Cheddar Bob.

 

[Assimilator]

It's 12 threads, 6 cores.

The SMT with AMD always gains you extra performance.

SMT is still not a full core.

 

[Denver]

Very unlikely to be a faulty Phoenix chip. That APU hasn't even appeared in laptop market yet.

 

[SL2]

Funny that almost no one believes in the first theory..

Disable the GPU and add -F to the end, something that consumers might recognize. (No you don't have to explain how joe sixpack doesn't care what it's called)

That APU hasn't even appeared in laptop market yet.

Actually, it has, even though I really don't think that the 7500F is Phoenix 2.

Notebooks mit Codename AMD: Phoenix2 Preisvergleich Geizhals EU

Preisvergleich und Bewertungen für Notebooks mit Codename AMD: Phoenix2

geizhals.eu

 

[Assimilator]

Funny that almost no one believes in the first theory..

Disable the GPU and add -F to the end, something that consumers might recognize. (No you don't have to explain how joe sixpack doesn't care what it's called)

Nobody wants to believe the first theory because it's boring. One of the selling points of Zen 4 was the fact that it finally has an IGP, so it would be a little strange for AMD to discard that differentiating feature.

The theory also doesn't have much evidence behind it, because there's been no indication that iGPU yields are poor and as such, contributing to an unacceptable high defect rate for Zen 4 CPUs. If that was the case I'd have expected to see a die-harvested CPU much earlier in the product cycle, yet here we are coming up on the first anniversary of Zen 4 and this is the first indication of such.

I will say however, that given the platform cost of AM5, anything that AMD can do to get component prices down is a smart, and very overdue, choice.

 

[AusWolf]

F means f*ck. That's what some of us are gonna say when AMD announces that it's another USA and Microcenter only SKU.

 

[Mussels]

SMT is still not a full core.

Just skimmed a few videos and articles on it, it's nothing amazing but SMT definitely provides more of a boost than HT

The views on why are all over the place, just throwing it out there

 

[Darmok N Jalad]

Sure seems to me that it's got the iGPU disabled. If it had a beefier iGPU, they'd probably stick with the G suffix.

 

[Assimilator]

Just skimmed a few videos and articles on it, it's nothing amazing but SMT definitely provides more of a boost than HT

The views on why are all over the place, just throwing it out there

I'm not intending to imply that SMT is worthless, simply that the world has very much moved along from 6 cores (AMD themselves are mostly responsible for this due to the popularity of 5800X3D).

 

[SL2]

Nobody wants to believe the first theory because it's boring.

Yup, this is something I see too many times in so many situations these days. Not saying you're wrong, tho.

Either way, desktop Phoenix with gimped graphics is pretty boring as well. Imagine the how pissed customers would be if a cut down version of Phoenix would show up for AM5 before the real one..

My bet is still on Raphael, which is also what I've seen on other sites. The Korean video is gone tho, never saw it.

 

[Minus Infinity]

Phoenix-2 doesn't have any "Zen 4c" cores. There's just one CCX with 6x "Zen 4" cores. The L3$ size is reduced to 8 MB, but there's no confirmation of this.

The L3 cache is reduced because it's a 4c core. There are plenty of stories indicating Phoenix 2 is for lower end devices and will use 4 and 4c cores

Phoenix 2

Phoenix 2

 

[Mussels]

I'm not intending to imply that SMT is worthless, simply that the world has very much moved along from 6 cores (AMD themselves are mostly responsible for this due to the popularity of 5800X3D).

I know - i was adding my views in
4 HT threads might be worth half a core, while SMT is worth a full core (games performance wise, at least)

That said, the world has NOT moved along from 6 cores at all... we only have 6 or 8 core CPU's at present from AMD and intel, with addon cores that are slowed down with latency issues (Hence, the issues with ryzen 3000 and gaming with their 3 core per CCX CPU's
look at the 3300x for an example on how that worked, with the 4 monolithic cores being faster in games (and other low latency tasks) than the 3+3 setup of the "faster" cpus

We get SOME benefit from the extras, but it's like HT and SMT where the gains are nothing at all like the gains from larger CPU dies with higher monolithic core counts

 

[Division7800X3D]

Just give me a Desktop verison of the 7940 HX or whatever. I want to build a SFF PC based on an APU. To be honest I would love to build my own handheld.

AGREED! I want high end APU @ minimum of | 8c/16t | with 12+ (16CU MINIMUM) RDNA2+/RDNA3 (But (rdna3.5) would be better ( Strix Halo 40CU Monster is one i really want on DESKTOP NOW AMD LETS GO!)

(https://www.gigabyte.com/Motherboard/B650I-AORUS-ULTRA-rev-10#kf)
Ii have a 3rd itx GB B650 I Aorus Ultra (Wifi6) AM5 board i want to use as a NVMe Storage/File Server (up to 32GB/s ) with asus Hype Gen4 x4 x4 m.2 NVMe x16 PCI-e card ( Bifurcated to x4x4x4x4 )x4 for eacj of the 4x m.2 nvme slots!

and teh AMD am5 APUS better have good iGPU and able to use x16 pcie gen4 lanes so i can do this! I have the nvme card and 4x 7400Mbps m.2 ssds i just need a good apu to run it ;( (and use it it for light gaming

and this itx mb holds 3x m.2 nvme g4x4 ssd on the mb that a total of 7x nvme ssds ( with @ minimum of 7x 1TB 7400Mbps (7.4 GB Max Seq reads Gammix S70 Blade 1TB NVME 2280 ssds)

Max up to 7x 4TB nvme ssds 4x7=28GB of EXTREEME STRIPED NVMe STORAGE!! LOL ITS AN EXPERIMENT I NEED TO DO!

7.4GB/s x7 = Total throughput (Theoretical) 51.8GB/s WOW!!

 

[Mussels]

AGREED! I want high end APU @ minimum of | 8c/16t | with 12+ (16CU MINIMUM) RDNA2+/RDNA3 (But (rdna3.5) would be better ( Strix Halo 40CU Monster is one i really want on DESKTOP NOW AMD LETS GO!)

(https://www.gigabyte.com/Motherboard/B650I-AORUS-ULTRA-rev-10#kf)
Ii have a 3rd itx GB B650 I Aorus Ultra (Wifi6) AM5 board i want to use as a NVMe Storage/File Server (up to 32GB/s ) with asus Hype Gen4 x4 x4 m.2 NVMe x16 PCI-e card ( Bifurcated to x4x4x4x4 )x4 for eacj of the 4x m.2 nvme slots!

and teh AMD am5 APUS better have good iGPU and able to use x16 pcie gen4 lanes so i can do this! I have the nvme card and 4x 7400Mbps m.2 ssds i just need a good apu to run it ;( (and use it it for light gaming

and this itx mb holds 3x m.2 nvme g4x4 ssd on the mb that a total of 7x nvme ssds ( with @ minimum of 7x 1TB 7400Mbps (7.4 GB Max Seq reads Gammix S70 Blade 1TB NVME 2280 ssds)

Max up to 7x 4TB nvme ssds 4x7=28GB of EXTREEME STRIPED NVMe STORAGE!! LOL ITS AN EXPERIMENT I NEED TO DO!

7.4GB/s x7 = Total throughput (Theoretical) 51.8GB/s WOW!!

Now to get that same speed over a network...